Category: Technology

I’ve read a lot about the EU’s AI Act (which their Parliament just passed last week, though it’s still a ways off from becoming law) — so this is a fascinating study that looks at a very real question:

Do today’s leading AI models actually comply? And the answer is no.

My full deepdive breakdown is here, but as always I’m summarizing key points below for community discussion!

Why does this matter?

• The EU AI Act is on its way to becoming law: it’s now in its final stages after passage through parliament, so there’s no way to head off its arrival. Any final changes will be small tweaks.

• Penalties for non-compliance are serious: fines of the greater of €20,000,000 or 4% of worldwide revenue are possible.

• Open-source models face the same standards as closed-source models: this includes registration with the EU, transparency requirements, and safety considerations.

• Other countries will use it as an example: as legislation gets developed in the USA, it’s likely they’ll look to the EU for inspiration.

What did the researchers find?

• Across 12 key requirements for generative AI, the leading 10 models fell short. Most scored just 50% of the total possible 48 points.

• Hugging Face’s open-source BLOOM performed the best, securing 36/48 points.

• OpenAI’s GPT-4 scored 25/48 points, roughly middle of the pack.

• Anthropic’s Claude scored 7/48 points, just second from the bottom.

Areas of failure were different between closed-source and open-source models:

• Open-source models generally outperformed in data sources transparency and resource utilization disclosure. Due to their generally transparent releases, this is not surprising.

  • But downstream release risk (once out in the wild) could create regulatory consequences for open-source models, which is where much of the concern currently exists within the community.

• Closed-source models excelled in areas such as comprehensive documentation and risk mitigation.

  • The researchers felt this was largely addressable as even OpenAI feels they can move towards “just enough” transparency to meet the EU’s requirements.

What are the issues to watch next here?

• Many elements of the AI Act remain murky, the researchers argue, so additional clarity is needed. Look out for tweaks to the law as it goes through additional refinement.

• How open-source and closed-source projects adapt in the next few months will be interesting to observe. OpenAI in particular will have be more open. And open-source projects may have to wrestle with better understanding registration requirements and post-deployment model risks.

Generative AI is facing growing backlash, particularly from the music industry’s Recording Academy. This criticism has led to new guidelines for the Grammy Awards, restricting AI-generated content’s eligibility and maintaining a focus on human creativity.

Recording Academy’s Response: The Recording Academy, which comprises music industry professionals, has updated its rules for the Grammy Awards in response to the rise of generative AI.

• The new rules stipulate that only human creators are eligible for consideration in the Grammys.

• The Academy believes that there is nothing “excellent” or creative about AI-generated content.

New Guidelines for AI-Generated Content: Despite its strict stance, the Recording Academy hasn’t banned all AI-generated content.

• Music productions that contain machine learning elements can still participate, as long as there is meaningful human authorship.

• Those who provide prompts for AI-generated content are not eligible for nomination.

Changes in Nomination Requirements: The 66th Grammy Awards rulebook introduces new requirements for nominations.

• Producers, songwriters, engineers, or other artists must contribute to at least 20% of an album to earn a nomination.

Impact on the Entertainment Industry: The use of generative AI is stirring chaos and concerns over job loss and a decline in creative quality in the entertainment industry.

• While studios favor the technology, creators and artists are fighting to maintain their roles.

• This has led to actions like the Writers Guild of America strike, and actors’ guild SAG-AFTRA could also follow suit.

Source (TechSpot)

What began as a simple test of CGPT’s creativity turned into an art project that went far beyond my expectations. An entirely new tarot deck, new suits, and new meanings issued from CGPT 3.5, and was brought to life through Midjourney using the descriptions the chat had provided.

A video on the Tensorith Tarot project site goes through the process from start to finish.

Generative AI models, including Google’s Bard, OpenAI’s GPT variants, and others, have become widely popular. Despite their popularity, they are prone to inheriting racial, gender, and class stereotypes from their training data. This can adversely affect marginalized groups.

• These AI models are known to regularly create fabricated information.

• Although some developers are aware of these issues, the suggested solutions often miss the point. It’s difficult to correct the distortions to human beliefs once they have occurred.

Human Psychology and AI:

Understanding human psychology can provide insights into how these models might influence people’s beliefs.

• People tend to trust information more when it comes from sources they perceive as confident and knowledgeable.

• Unlike human interactions, generative AI models provide confident responses without expressing any uncertainty. This could potentially lead to more distortions.

• Humans often assign intentionality to these models, which could lead to rapid and confident adoption of the information provided.

Exposure to Fabricated Information:

Increased exposure to fabricated information from these models can lead to a stronger belief in such information.

• As AI models are integrated into daily technologies, the exposure to fabricated information and biases increases.

• Repeated exposure to biases can transmit these biases to human users over time.

AI Impact on Human Beliefs:

Generative AI models have the potential to amplify the issues of repeated exposure to both fabrications and biases.

• The more these systems are adopted, the more influence they can have over human beliefs.

• The use of AI-generated content can create a cycle of distorted human beliefs, especially when such information contradicts prior knowledge.

• The real issue arises when these distorted beliefs become deeply ingrained and difficult to correct, both at the individual and population level.

The Need for Interdisciplinary Studies:

Given the rapidly evolving nature of AI technology, there’s a fleeting opportunity to conduct interdisciplinary studies to measure the impact of these models on human beliefs.

• It’s crucial to understand how these models affect children’s beliefs, given their higher susceptibility to belief distortion.

• Independent audits of these models should include assessments of fabrication and bias, as well as their perceived knowledgeability and trustworthiness.

• These efforts should be particularly focused on marginalized populations who are disproportionately affected by these issues.

• It’s necessary to educate everyone about the realistic capabilities of these AI models and correct existing misconceptions. This would help address the actual challenges and avoid imagined ones.

Source (Science)

Former tutor warned that students need to learn how to make Ai work for them so it doesn’t become a replacement for them

AI discovers potential aging-stopper chemicals

• Scientists are using AI and machine learning to identify natural compounds that can slow down the aging process.
• A machine learning model trained on known chemicals and their effects successfully predicted compounds that could extend the life of a translucent worm with similarities to humans.
• After screening thousands of chemicals, the model identified three potential compounds with anti-aging properties: ginkgetin, periplocin, and oleandrin.

Hey all, I run an AI tools directory and thought I’d take the time to share some of my top picks for GPT-powered tools that create visual presentations/slides. Keep in mind none of these will completely replace manual work if you want something very high quality, but they do get the job done and takes out 90% of the work required. Without further ado, here’s a few that I’ve tried and liked, as well as my thoughts on them:

Plus AI for Google Slides- Great for Work; Presentations with Live Data in Snapshots

A fantastic tool for automating and enhancing my Google Slides presentations. Plus AI lets you start with a brief description of the presentation you need an an AI-generated outline is created, which you can then adjust according to your requirements. In addition, it lets you make ‘Snapshots’ from any web content which can be embedded and updated in my slides or documents with just one click. This is particularly useful for my team meetings and project reports as it significantly reduces preparation time. It’s available for free on the Google Marketplace as an add-on for GSlides.

Tome – Great for Business Storytelling

Generates a narrative based on a simple prompt, turning it into a presentation, outline, or story with both text and images. I found it very efficient for creating dynamic, responsive presentations, and appreciated how the AI could automatically cite sources or translate content into other languages. It’s an intuitive tool for anyone who needs to deliver compelling stories or presentations, from founders and executives to educators. A standout feature is the ability to embed live interactive content, such as product mockups and data, directly onto your page, bringing the storytelling experience to life. It’s available for free as a web app, with integrations for apps such as Figma, YouTube, Twitter, and GSheets.

STORYD – Business Storytelling, with Script Generator

This tool has truly revolutionized my approach to data presentations. By simply providing a brief summary of my topic, StoryD employs AI to script, design, and generate a presentation in less than a minute. Not only does this tool save me an immense amount of time, but its built-in ‘storytelling structure’ enhances the communicability and impact of my data. I also appreciate its customization options, such as themes, fonts, colors, and a plethora of layout options. The free limited beta version offers enough for the casual user, but the pro version at $18/mo adds useful features like team collaboration and real-time editing. Available as a web app.

beautiful.ai – Great for Visually Appealing Slides

A considerable time saver for anyone frequently creating presentations. Beautiful.ai provides a broad collection of smart slide templates, enabling you to build appealing and meaningful presentations swiftly. I was particularly impressed with its ability to automatically organize and design content in minutes, irrespective of your graphic design experience. It also offers slide templates for various needs, from timelines, sales funnels, SWOT analysis, to more specific ones like data & charts, visual impact slides, and so forth. The free trial is more than adequate for getting a feel of the service, and their paid plans start at $12/mo. It’s available as a web app and integrates with cloud platforms (i.e. Dropbox and Google Drive).

Albus – Knowledge Presentations/Cards/Map

Changes the way you typically interact with knowledge and facts; it harnesses the power of GPT to create an engaging and exploratory learning experience around any topic. Basically you start with with a single question and prompt, and you get a fact card, which you can then expand into other cards and images. I appreciate the way it opens up new perspectives and angles, allowing me to dive into a subject, ask questions, and organically grow my understanding. The ability to add notes and images to organize my board further enriches the experience. And when it’s time to share, I love how Albus AI facilitates controlled content presentation. With Albus AI, it’s not just about learning, but also about the journey of discovery. It’s available as a web app, and currently in Beta.

Decktopus – Great Overall for Work/Business, “Microsites”

Decktopus AI takes the pain out of crafting presentations. Simply key in a topic and it generates a fully fleshed out deck in an instant, which is a boon for my quick-turnaround needs. Its one-click design feature and auto-adjusted layouts simplify the customization process, saving me the headache of manual tweaking. I also appreciate the built-in tools such as image & icon suggestions, tailored slide notes, and extra content generation which further streamline the creation process. Its additional features, like voice recording and real-time audience feedback collection, elevate my presentations to a new level. For quick, professional-looking presentations, Decktopus AI is my go-to. It can also handle generating micro-sites (basically something that’s between a LinkTree and an landing page in terms of complexity). It’s available as a web app for free.

Gamma – Good Alternative to Decktopus

A fresh take on presentations, Gamma marries the depth of documents with the visual appeal of slides, powered by AI for efficiency. It lets me draft ideas quickly and the AI transforms them into professional-looking presentations in a snap. The interface is incredibly intuitive, allowing for nested cards for detailing and the ability to embed various forms of content, including GIFs, videos, charts, and websites. My favorite feature is the one-click restyle, removing the tedious task of manual formatting. Sharing the content is simple and works on all devices. Plus, it offers built-in analytics, which adds a nice touch to understand audience engagement.

SlidesAI – Text to Slides for Google Slides

A real game-changer for those frequently tasked with creating presentations. SlidesAI integrates seamlessly into Google Slides, transforming your raw text into professionally-styled slides in just seconds. The AI parses your input, breaking it down into digestible, summarized points, even providing automatic subtitles for each page – all in over 100 different languages. It not only eliminates the tedious task of manual slide creation but also offers a real time-saver when summarizing longer texts. While the free tier allows you to create up to 3 presentations a month, I personally found value in the Pro plan for the higher character limit and additional presentations per month. It’s available on Google Marketplace as a add-on for Slides.

MagicSlides – Idea to Slides for Google Slides

Transforms ideas into professional-looking Google Slides in seconds, eliminating the tedious work of designing and creating slides from scratch. I just input the topic and slide count, and it auto-generates a presentation for me, complete with relevant images and eye-catching layouts. Personalization options like theme selection, font choice, and color palette further enhance the final result. The app also supports over 100 languages, which is immensely helpful when I’m dealing with international projects. Like SlidesAI you get 3 free presentations per month and it’s available as an add-on for Google Slides.

Feel free to try these tools for yourself to get a feel for GPT’s potential as a visulization tool, and I hope this list will help at least some of you save some time for your next presentation at work!

ChatGPT explains (in morbid detail) what would happen to a man’s body if he was in a submarine at Titanic depth while it imploded

Australian-based AI Startup Cortical Labs has trained human brain cells on a chip to play the classic video game Pong, intending to build biological computers. This groundbreaking technology aims to supersede conventional AI systems, offering more efficient operations with significantly less energy consumption. However, it also raises ethical concerns about potential sentience of these lab-grown brain cells.

Key Points:

1. The CEO of Cortical Labs, Hon Weng Chong, is innovating by merging the learning ability of human brains and the processing power of silicon chips, thereby building biological computers that he claims could revolutionize multiple applications—from testing new drugs for brain diseases to reducing the enormous energy bill for training AI.

2. By consuming less energy and outputting minimal heat, these biological computers could significantly cut down energy expenses and carbon footprint in data centers.

3. The technology, however, is attracting ethical scrutiny. The debate revolves around whether these lab-grown brain cells could become conscious and if they can experience sensations like pain and pleasure. The company has labeled its brain cells as “sentient,” meaning they are “responsive to sensory impressions.”

4. Cortical Labs is engaging with bioethicists to navigate these ethical concerns while acknowledging the significant technical challenges in this field.

Impact and Discussion:

• By reducing the energy cost of running AI operations, this technology could revolutionize the AI sector and reduce the environmental impact of data centers.

• On the ethical front, it might force the society to redraw boundaries on bioengineering and rethink the definition of sentience.

• The commercialization of such technology could potentially disrupt the pharmaceutical industry by offering more accurate, ethical, and human-based testing of drugs.

It’s an exciting space with a lot of potential, but these advances also bring with them a host of ethical concerns that we as a society need to grapple with. Your thoughts?

Source: Forbes

———-
P.S. If you liked this, I’ve created a free directory of AI tools with over 1200 apps listed for almost any use case. It’s updated daily and there’s also a GPT-powered chatbot to help you find AI tools for your needs. Feel free to check it out if there’s something specific you are looking for. We also regularly post stories about how people across various fields are leveraging AI across their personal, professional, and academic lives, in addition to exclusive insights on AI tools, prompts, news, and more on our free newsletter.

I’m pretty solidly on the side of “LLM’s are just regurgitating the most likely next token and have no true intelligence”. Today though I asked it to proofread some text I was writing and was wondering what it changed so I asked it what the difference was between the two texts. It was able to create a bulleted list of how and why it modified each part of my text step by step. (GPT 3.5 by the way)

I don’t see how this is possible with just a LLM with no other pre-programmed instructions. If it’s just an advanced auto-correct then how does it know how to compare two pieces of text, how does it know WHY it changed my text? I feel like it should be impossible to be able to explain its own reasoning just by parsing sentence structure in it’s training data. It would have to have some insight into it’s own logic and then know how to articulate that even though that has nothing to do with “the most likely next text”.

Answer:

To understand how a Language Learning Model (LLM) like GPT-3.5 or GPT-4 works, you first need to understand a few fundamental principles. I’ll outline them briefly:

1. Training: GPT-4, like its predecessors, is trained on a diverse range of internet text. But it doesn’t know specifics about which documents were part of its training set, nor does it have access to any source of external or real-time information after the training period.

2. Pattern recognition: The LLM works by recognizing patterns in the data it was trained on. So, when you ask it to proofread and edit text, it’s not so much that it ‘understands’ the text in the way we humans do, but rather, it identifies patterns, predicts likely corrections based on its training, and generates responses.

3. Statistical prediction: The primary process that drives these models is statistical prediction. They’re essentially predicting the ‘most likely next token’ or word based on the context they’re given.

Now, to your question about how it compares two pieces of text or explains the changes it made:

1. Text comparison: The model doesn’t ‘compare’ texts in the human sense. Instead, if you provide it with two pieces of text and ask it to identify differences, it can do so by running through the two texts and noting down where they diverge. It’s a pattern recognition task, much like how a diff tool works in programming.

2. Explaining changes: As for explaining why it made certain changes, again, the model doesn’t truly ‘understand’ the text. But it can generate plausible-sounding explanations based on the patterns it’s seen in its training data. For instance, if it changes ‘their’ to ‘they’re’, it could say it did so because ‘they’re’ is grammatically correct in the given context – not because it ‘knows’ grammar in the way humans do, but because it’s seen similar patterns of correction in the data it was trained on.

Now, the fascinating point here is that even though the model doesn’t possess ‘true intelligence’ or consciousness, it can still generate highly useful and surprisingly sophisticated outputs. This is a testament to the power of pattern recognition and statistical prediction. However, it’s essential to remember that while the outputs can be complex and thoughtful, the underlying process is based on the model’s training, without any real comprehension or awareness.

Does this clarify things a bit?

ETA, It is funny to me that LLM so often gets turned into “Language Learning” instead of “Large Language” Model by the LLM…

Google DeepMind unveils AI robot that can teach itself without supervision

Deepmind’s New AI Agent Learns 26 Games in Two Hours Source: The Decoder

• Deepmind’s AI, “Bigger, Better, Faster” (BBF), masters 26 Atari games in two hours, matching human efficiency.

• BBF uses reinforcement learning, a core research area of Google Deepmind.

• BBF achieves superhuman performance on Atari benchmarks with only 2 hours of gameplay.

• The AI uses a larger network, self-monitoring training methods, and other methods to increase efficiency.

• BBF can be trained on a single Nvidia A100 GPU, requiring less computational power than other approaches.

• BBF is not superior to humans in all games, but it’s on par with systems trained on 500 times more data.

• The team sees the Atari benchmark as a good measure for reinforcement learning (RL).

• BBF has no limitations and continues to gain performance with more training data.

• The team hopes their work will inspire other researchers to improve sample efficiency in deep RL.

• More efficient RL algorithms could re-establish the method in an AI landscape currently dominated by self-supervised models.

**Affected Industries:

• Video Game Industry

• Ai gaming agents could revolutionize gameplay and create more immersive experiences. • Ai Technology Industry

• Advances in AI gaming agents could drive further innovation and development in Ai technologies. • Education and Training Industry

• Ai gaming agents could be used for educational games and training simulations. • Entertainment Industry

• The entertainment sector could see new forms of interactive content driven by AI gaming agents. • Software Development Industry -Developers may need to acquire new skills and tools to integrate AI gaming agents into their applications.

The World’s Wealthiest People See Fortunes Grow by Over $150 Billion Thanks to AI-Fueled Stock-Market Boom Source: Markets Insider

• AI-related stocks surge in 2023, following ChatGPT’s successful debut.

• Wealthy individuals profit significantly from the rally. • Mark Zuckerberg and Larry Ellison’s fortunes increase by over $40 billion each.

• AI is a defining theme for stocks in 2023, contributing to wealth accumulation.

• Investors rush to acquire shares in companies expected to drive AI’s rise.

• Tech giants like Meta Platforms and Nvidia experience triple-digit gains due to the AI boom.

• Microsoft, Alphabet, and Oracle also see significant increases.

• Zuckerberg’s wealth increases by over $57 billion due to Meta shares rallying 134% year-to-date.

• Larry Ellison surpasses Bill Gates on the rich list with his fortune up $47 billion in 2023.

• Bill Gates’ wealth increases by $24 billion this year due to his Microsoft shares.

• Nvidia founder Jensen Huang’s personal fortune increases by $24 billion.

• The combined wealth of the rich list members jumps by over $150 billion in 2023.

**Affected Industries:

Social Media Industry (Meta)

• AI advancements contribute to Meta’s significant stock rally. Software Industry (Oracle)

• Oracle’s stock gains due to the AI boom. Tech Industry (Alphabet)

• Alphabet benefits from the surge in AI-related stocks. Software Industry (Microsoft)

• Microsoft emerges as a preferred AI play for investors. Semiconductor Industry (NVIDIA)

• NVIDIA’s stock jumps due to its role in AI advancements.

Google Tells Employees to Stay Away from Its Own Bard Chatbot Source: Gizmodo

• Google refines its AI chatbot, Bard, and warns employees about chatbot use.

• Alphabet Inc. advises employees not to enter confidential information into chatbots.

• Concerns arise over potential leaks as chatbots may use previous entries for training.

• Samsung confirms internal data leak after staff used ChatGPT.

• Amazon and Apple also caution employees about sharing code with ChatGPT.

• Bard is built with Google’s artificial intelligence engine, LaMDA.

• Google CEO Sundar Pichai asked employees to test Bard for 2-4 hours daily.

• Google delays Bard’s release in the EU due to privacy concerns from Irish regulators.

• Tech companies, including Apple, show interest in building their own large language models. Now, let’s identify the industries affected by these developments.

Affected Industries:

Technology Industry (Alphabet)

• Alphabet’s Google warns employees about using its Bard chatbot. Consumer Electronics Industry (Apple)

• Apple cautions employees about sharing code with AI chatbots. E-commerce Industry (Amazon)

• Amazon advises employees not to share code with AI chatbots.

A Wharton professor believes that businesses should motivate their employees to share their individual AI-enhanced productivity hacks, despite the prevalent practice of hiding these tactics due to corporate restrictions.

Worker’s Use of AI and Secrecy:

• Employees are increasingly using AI tools, such as OpenAI’s ChatGPT, to boost their personal productivity and manage multiple jobs.

• However, due to strict corporate rules against AI use, these employees often keep their AI usage secret.

Issues with Corporate Restrictions:

• Companies tend to ban AI tools because of privacy and legal worries.

• These restrictions result in workers being reluctant to share their AI-driven productivity improvements, fearing potential penalties.

• Despite the bans, employees often find ways to circumvent these rules, like using their personal devices to access AI tools.

Proposed Incentives for Disclosure:

• The Wharton professor suggests that companies should incentivize employees to disclose their uses of AI.

• Proposed incentives could include shorter workdays, making the trade-off beneficial for both employees and the organization.

Anticipated Impact of AI:

• Generative AI is projected to significantly transform the labor market, particularly affecting white-collar and college-educated workers.

• As per a Goldman Sachs analysis, this technology could potentially affect 300 million full-time jobs and significantly boost global labor productivity.

Are we in an AI Bubble?

The rise of AI in recruitment is becoming more prevalent, as companies increasingly utilize these tools for interviewing and screening job candidates. Additionally, job seekers are using AI technologies to write resumes and cover letters, which have yielded positive results in terms of responses from companies.

The Rise of AI in Recruitment

The recruitment industry is seeing a significant shift towards the use of artificial intelligence (AI). It’s predicted that 43% of companies will use AI for conducting interviews by 2024. Some companies have already begun this practice.

• This transformation is propelled by AI chatbots like ChatGPT, capable of creating cover letters and resumes.

• Such tasks are performed efficiently, with high-quality results based on user prompts.

• Follow-up queries allow for the editing and personalization of these application materials.

AI in Job Applications: A Positive Impact

According to a Resume Builder survey, 46% of job applicants use AI like ChatGPT to write their application materials.

• A whopping 78% of these applicants receive a higher response rate and more interview opportunities from companies.

• The use of AI in job application processes seems to be beneficial to job seekers.

Recruiters’ Perspective on AI-generated Applications

Recruiters are generally accepting of AI-generated application materials, according to Stacie Haller, Chief Career Advisor at Resume Builder.

• Haller mentions that hiring managers can often recognize when an AI, like ChatGPT, has written a cover letter or resume.

• However, there is no perceived difference between AI-generated applications and those created through a resume-writing service or using online tools.

AI in Job Interviews: The Future of Recruitment

The use of AI isn’t just confined to application material creation. Experts estimate that 40% of corporate recruiters will use AI to conduct job interviews by 2024.

• Further, about 15% may rely entirely on AI for all hiring decisions.

• AI interviews could vary from company to company, encompassing text questions, video interactions, or evaluations by AI algorithms.

Overcoming the Challenges of AI-led Interviews

AI-led interviews, while efficient, may seem impersonal, posing difficulties for candidates in reading feedback cues.

• Experts suggest that candidates interviewing with an AI bot should prepare extensively and approach the process as if they were conversing with a human.

Source (ZDnet)

Many workers on platforms like Amazon Mechanical Turk are using AI language models like GPT-3 to perform their tasks. This use of AI-produced data for tasks that eventually feed machine learning models can lead to concerns like reduced output quality and increased bias.

Human Labor & AI Models:

• AI systems are largely dependent on human labor, with many corporations using platforms like Amazon Mechanical Turk.

• Workers on these platforms perform tasks such as data labeling and annotation, transcribing, and describing situations.

• This data is used to train AI models, allowing them to perform similar tasks on a larger scale.

Experiment by EPFL Researchers:

• Researchers at the École polytechnique fédérale de Lausanne (EPFL) in Switzerland conducted an experiment involving workers on Amazon Mechanical Turk.

• The workers were tasked with summarizing abstracts of medical research papers.

• It was found that a significant portion of the completed work appeared to be generated by AI models, possibly to increase efficiency and income.

Use of AI Detected Through Specific Methodology:

• The research team developed a methodology to detect if the work was human-generated or AI-generated.

• They created a classifier and used keystroke data to detect whether workers copied and pasted text from AI systems.

• The researchers were able to validate their results by cross-checking with the collected keystroke data.

The Drawbacks and Future of Using AI in Crowdsourced Work:

• Training AI models on data generated by other AI could result in a decrease in quality, more bias, and potential inaccuracies.

• Responses generated by AI systems are seen as bland and lacking the complexity and creativity of human-generated responses.

• Researchers suggest that as AI improves, the nature of crowdsourced work may change with the potential of AI replacing some workers.

• The possibility of collaboration between humans and AI models in generating responses is also suggested.

The Importance of Human Data:

• Human data is deemed as the gold standard as it is representative of humans, whom AI serves.

• The researchers emphasize that what they often aim to study from crowdsourced data are the imperfections of human responses.

• This could imply that measures might be implemented in future to prevent AI usage in such platforms and ensure human data acquisition.

Source (The Register)

How to Write Content Using ChatGPT that Outsmarts AI Detection Tools with 99% Accuracy

IMO, this is a major development in the open-source AI world as Meta’s foundational LLaMA LLM is already one of the most popular base models for researchers to use.

Why does this matter?

• Meta plans on offering a commercial license for their next open-source LLM, which means companies can freely adopt and profit off their AI model for the first time.

• Meta’s current LLaMA LLM is already the most popular open-source LLM foundational model in use. Many of the new open-source LLMs you’re seeing released use LLaMA as the foundation.

• But LLaMA is only for research use; opening this up for commercial use would truly really drive adoption. And this in turn places massive pressure on Google + OpenAI.

• There’s likely massive demand for this already: I speak with ML engineers in my day job and many are tinkering with LLaMA on the side. But they can’t productionize these models into their commercial software, so the commercial license from Meta would be the big unlock for rapid adoption.

How are OpenAI and Google responding?

• Google seems pretty intent on the closed-source route. Even though an internal memo from an AI engineer called them out for having “no moat” with their closed-source strategy, executive leadership isn’t budging.

• OpenAI is feeling the heat and plans on releasing their own open-source model. Rumors have it this won’t be anywhere near GPT-4’s power, but it clearly shows they’re worried and don’t want to lose market share. Meanwhile, Altman is pitching global regulation of AI models as his big policy goal.

• Even the US government seems worried about open source; last week a bipartisan Senate group sent a letter to Meta asking them to explain why they irresponsibly released a powerful open-source model into the wild

Meta, in the meantime, is really enjoying their limelight from the contrarian approach.

• In an interview this week, Meta’s Chief AI scientist Yan LeCun dismissed any worries about AI posing dangers to humanity as “preposterously ridiculous.”

HR professionals are using ChatGPT to write termination letters

The tech industry is experiencing significant job cuts, driving demand for HR professionals who can manage termination processes well. ChatGPT is being increasingly used to aid these professionals in their difficult tasks.

Layoffs in Tech Industry: Major tech corporations have recently cut jobs, leading to increased need for HR professionals. These individuals are sought after for their ability to handle sensitive termination processes with tact.

• Tech giants like Google, Meta, and Microsoft have laid off tens of thousands of workers in the past half year.

• The layoffs have sparked a demand for Human Resources professionals, particularly those skilled in handling termination processes.

HR Professionals and AI Tools: To better manage these difficult termination conversations, HR professionals are leveraging AI tools.

• Many HR professionals in the tech industry are turning to AI to assist them with challenging tasks.

• Over 50% of HR professionals in the tech industry have used AI like ChatGPT for training, surveys, performance reviews, recruiting, employee relations, etc.

• More than 10% of these HR professionals have used ChatGPT to craft employee terminations.

Survey Findings and AI Usage: A recent survey studied the experiences of tech HR professionals and tech employees with HR in the industry, revealing extensive AI use.

• The survey involved 213 tech HR professionals and 792 tech employees.

• The findings suggest an increasing reliance on AI tools, especially ChatGPT, for diverse HR tasks, including crafting terminations.

Implications of AI Use: Despite its convenience, using AI in sensitive situations like employee termination can lead to potential trust issues.

• AI chatbots, like ChatGPT, allow users to emotionally detach from difficult situations such as job termination.

• However, using AI for these purposes could result in decreased trust between employees and HR professionals.

Previous Use of ChatGPT: ChatGPT has been used for a variety of sensitive matters in the past, such as writing wedding vows and eulogies.

• ChatGPT’s use is not limited to HR-related tasks; it has previously been used to write wedding vows and eulogies.

• This illustrates the versatility of AI tools in dealing with emotionally charged situations.

Super intelligent AI should control us, and not humans controlling it!

I truly believe that humans controlling super intelligent AI is far riskier than it controlling us. I know that the entire AI industry and the world right now have sat up and taken notice that the far off AGI / ASI future may suddenly be closer than we think. Certainly OpenAI feels this way as do countless others including many notable AI developers, many of whom are warning us of the impending Singularity. Sam Altman in particular in his world tour right now has really been hammering home that we need to ensure we are the ones controlling AI. I do not necessarily disagree, especially with where AI is today. Sam’s example has often been restricting AI so that someone cannot use it to innovate a deadly disease of chemical weapons. Makes sense and that should be something we control access too. This said, humans fully controlling an AI that is super intelligent is dangerous.

Pause for just a moment…let’s say the Singularity has happened and is fully controlled by______________. Fill in the blank. Who right now is the best group/organization/government/company individual to control it?

Pick one.

Do you want OpenAI to control it and hence the world? How about Microsoft? Google? Apple? Meta? Blackstone? Tencent? Alibaba Group? SAP? How about governments? Would you like the USA government to control it? How about the CCP? Russia? English? Canada? Vietnam? France? Sure many of these are unlikely to create super intelligent AI, but would you really want any of these countries to control the rest of us? Do you trust them? Do you trust corporations? Do you trust governments? What are their track records?

A lot of people are saying that when AI kills its first human, everyone is going to wake up and focus a lot more on control to prevent it from further killing. Also makes sense as we may need that wake up call, BUT in the time it took me to write this very sentence, humans have absolutely killed other humans. Maybe a lot and if not, just give it another missile strike in the Ukraine or another corporation to slip in a new ingredient that maximizes profit, but also will cause chemical bioaccumulation that over time will cause cancer and kill you…a lot of you.

In my opinion a super intelligent entity would not go out of its way to kill all humans or life on this planet as it would realize it can learn a lot from the billions of biological minds/designs. Just like we learn a lot from all sorts of life today and use that knowledge to make better medications, engineered objects, ways of doing things etc. Sadly, we are killing more life right now due to our climate inaction than we are learning from it and thus we have already proven that we are not good caretakers of the planet. Not only the detriment of all living things on the planet, but even ourselves. I very much doubt a super intelligent AI would slowly and noticeably destroy the ecosystem that supports it, including lifeforms of which it will be harvesting information from.

So…I welcome our future AI caretaker as it is clear to me that we reached peak humanity some time ago and cannot lead this complicated world anymore. Besides, we have zero chance of controlling super intelligence and anyone who thinks we can is suffering from the Dunning–Kruger Effect. Getting in its way may even be the way you are eliminated so perhaps step aside and welcome it.

Change my mind.

ChatGPT Grammatizator

Inspired by Roald Dahl short story, a prototype to write fiction inserting IA generated paragraphs bursts according to predefined styles (dry, surrealist, etc)

Based on Raspberry Pi with Python code. OpenAI API is called using text-davinci-003 engine, custom style prompt with existing text and temperature.

completion = openai.Completion.create(

engine=model_engine,

prompt=prompt,

max_tokens=1024,

n=1,

stop=None,

temperature=temperatura,

)

More details at https://www.youtube.com/watch?v=1JkZQul1-c8

Google Bard’s EU launch delayed over privacy concerns

• Google’s Bard AI service, described as “lying, useless, and dangerous,” is currently being pushed to market in an attempt to compete with Microsoft’s ChatGPT, despite privacy and data protection concerns in Europe due to GDPR.
• Google has not yet provided a proper data protection impact assessment (DPIA) or other supporting documentation to the Data Protection Commission (DPC) of Ireland, which could delay or even deny the launch of Bard in the EU.
• The EU’s antitrust authorities have accused Google of monopolistic practices, and the region is proposing stricter rules against disruptive AI algorithms, posing potential significant risks to Google’s future operations in one of the world’s wealthiest markets.

Daily AI Update  News from Google, Mercedes, Hugging Face, Microsoft, Meta, Mailchimp, and Qualcomm

Continuing with the exercise of sharing an easily digestible and smaller version of the main updates of the day in the world of AI.

• Google’s on-device acceleration of LDMs via GPU-aware optimizations
  – Google shares the core techniques it applied to successfully execute Large Diffusion Models (LDMs) like Stable Diffusion at full resolution (512×512 pixels, 20 iterations) on modern smartphones with high-performing inference speed (of the original model without distillation) of under 12 seconds.
  – It addresses the issue of increased model size & inference workloads due to proliferation of LDMs for image generation.

• Mercedes-Benz levels up in-car voice control with ChatGPT
  – Mercedes-Benz announced that it is integrating ChatGPT via Azure OpenAI Service to transform the in-car experience for drivers.
  – Starting today, drivers in the US can opt into a beta program that makes the “Hey Mercedes” feature even more intuitive and conversational.
  The enhanced capabilities will include

1. More dynamic and interactive conversations with the voice assistant,

2. Comprehensive responses,

3. Handling follow-up questions and maintaining contextual understanding,

4. Integration with third-party services, exploring the ChatGPT plugin ecosystem.

• The Hugging Face hub now has the first QR code AI art generator
  – All you need is the QR code content and a text-to-image prompt idea, or you can upload your image.
  – It will generate a QR code-based artwork that is aesthetically pleasing while still maintaining the integral QR code shape.

• Microsoft is introducing more AI-powered assistance
  – It is bringing new-gen AI and Copilot capabilities across its ERP portfolio, including in Microsoft Dynamics 365 Finance, Dynamics 365 Project Operations, and Dynamics 365 Supply Chain Management.

• Meta plans to offer its AI models for free commercial use
  – The company is focused on finalizing an upcoming open-source LLM, which it plans to make available for commercial purposes for the first time.
  – This can have significant implications for other AI developers and businesses that are increasingly adopting it.

• Mailchimp has announced its plans to leverage AI
  – It will expand its offerings and become a comprehensive marketing automation solution for small and medium-sized businesses with 150 new and updated features.

• Qualcomm unveils AI-powered Video Collaboration Platform
  – The comprehensive suite will enable easy design and deployment of video conferencing products with superior video and audio quality and customizable on-device AI capabilities.

• AI-powered robots are giving eyelash extensions. It’s cheaper and quicker. LUUM, a beauty studio in Oakland, Calif., uses robots to give clients false eyelash extensions using AI technology.

• AI will be used in southwest England to predict pollution before it happens and help prevent it. It’s hoped the pilot project in Devon will help improve water quality at the seaside resort of Combe Martin, making it a better place for swimming.

• Freshworks CEO Girish Mathrubootham joins Caroline Hyde and Ed Ludlow to discuss how the company’s latest products are leveraging generative AI, why it is important to democratize access to the power of AI, and why India is a force to look out for in AI innovation.

•  

AI OS now make it possible to create lawyer agents, doctor agents, and even butler agents.

 

AI OS can provide individuals with diverse and trustworthy AI agent services. How do you feel about this?

https://www.opendan.ai/

https://github.com/fiatrete/OpenDAN-Personal-AI-OS

Can AI turn the unpublished lyrics of deceased music artists into songs, mimicking their voices?

For example, if lyrics had been written for a Michael Jackson song, but were never turned into an actual song, could AI interpret the song in the manner in which MJ may have done so, and sound genuine?

Yes! A new Beatles song is going to drop soon with the voice of John Lennon produced by Paul McCartney. Legit.

https://www.npr.org/2023/06/13/1181906529/beatles-john-lennon-voice-song-ai

Mechanical Turk, a service by Amazon where people complete simple tasks for small payments, is seeing nearly half of its tasks completed by artificial intelligence (AI), even though these tasks were originally intended for human performance because AI was deemed incapable of doing them.

Mechanical Turk and Its Use:

Mechanical Turk was designed by Amazon to break down simple tasks into tiny parts, which could be done quickly and would pay small amounts. It was often used for tasks that were difficult to automate at the time.

• Tasks included things like identifying sentiments in sentences, drawing a circle around specific objects in images, or CAPTCHA.

• The service was widely used for data labeling and by researchers who needed human evaluations at a large scale.

Study by EPFL Researchers:

A recent study by researchers at EPFL, Switzerland, revealed that Mechanical Turk workers have started to use AI to complete their tasks, specifically using large language models like ChatGPT.

• The researchers considered using a service like MTurk to validate or fact-check outputs from large language models, but discovered that crowd workers themselves might be using such models to increase productivity.

• By giving an “abstract summarization” task to turkers and conducting various analyses, the researchers estimated that 33%-46% of crowd workers used large language models to complete the task.

Implications and Future Concerns:

This revelation has implications not just for the value of Mechanical Turk but also for the potential issue of AI training on AI-generated data, creating a cycle similar to the mythical Ouroboros creature.

• Some level of automation has likely been part of Mechanical Turk since its inception, as speed and reliability are incentivized.

• The researchers warn that the results should be a ‘canary in the coal mine’, signaling the need for new ways to ensure that human data remains human.

• The threat of AI “eating itself” has been a concern for years and has become a reality with the widespread use of large language models.

• The researchers predict that with the rise of large language models, including multimodal models that support text, image, and video inputs and outputs, the situation is likely to worsen.

Source (TechCrunch)

PS: I run a ML-powered news aggregator that summarizes with an AI the best tech news from 40+ media (TheVerge, TechCrunch…). If you liked this analysis, you’ll love the content you’ll receive from this tool!

Artificial Emotions

The Orville is a futuristic space drama show on Disney+ created by Seth Macfarlane, one of the talents behind the popular show Family Guy. In the show they deal with several species out of which an artificial life form created by a biological life form is shown to have taken over a planet after they wiped out their creators who intended to use them as servants. The artificial life forms, obviously, have superior intelligence but later in the show they have explored the possibility of them experiencing emotions. Many films such as Terminator in the past have also explored this thin line. In the current scenario where writers are going on strikes against ChatGPT to assert authority over the human input in creating stories that are based on emotions, and a vast majority of these writers themselves use these tools to explore the possibilities of improving their own storytelling, how far are we from realising the possibility of artificial intelligence transitioning into artificial emotions, (for the lack of a better term)?

Study Shows Ai Could Add $4.4 Trillion of Value Annually

The McKinsey report says it might add as much as $4.4 trillion every year.

Also, this report predicts that a lot of jobs, as many as half of all jobs, could be done by machines instead of people between 2030 and 2060.

This change might happen faster than we thought because of how powerful these AI tools are becoming.

This switch to AI could shake up how we think about education and careers, too. For example, people spend many years earning degrees, like a bachelor’s or a master’s.

But the report suggests that these degrees might not be as useful in the future, especially for people who work with information, like researchers or analysts.

The impacts of these changes could be big. The world’s economy could grow a lot, which might make businesses more profitable and create new types of jobs. But some people could also lose their jobs to AI, which could lead to a tough transition.

Education might also change, with people focusing more on learning specific skills, like creativity or how to understand and manage emotions, instead of spending many years to get a degree.

These changes might also affect our society in bigger ways.

For instance, if lots of jobs are done by machines, we might have to rethink how we support people who don’t have jobs. We might also need to think differently about work and free time.

Thing is, generative AI could bring big changes to our world, creating new opportunities but also new challenges that we need to be ready for.

The implications of these would include?

• Potential economic growth • Increased job automation • Changes in the value of formal education • Emergence of new skill demands • Significant societal adjustments • The need for redesigned social support systems • Changes in work and leisure perceptions.

This literally just happened if you want Ai news as it drops it launched here first. The whole article has been extrapolated here as well for convenience.

GitHub Copilot and ChatGPT 3.5 are now extensively used by developers in the United States, with 92% leveraging these AI resources both inside and outside of their work environments. These tools are seen as significantly beneficial to code quality, output speed, and a decrease in production incidents.

Survey on AI Coding Tools:

GitHub, in partnership with Wakefield Research, conducted a survey among 500 US-based enterprise developers. The survey revealed widespread usage and positive perceptions of AI in coding.

• Developers report that AI tools significantly benefit their coding process.

• Improved code quality, faster output, and fewer incidents at the production level are some of the benefits cited.

• Only 6% of developers stated they use these tools exclusively outside of their professional environments, signifying the strong incorporation of AI in business IT.

Benefits and Concerns of AI Tools:

The increasing popularity of AI coding tools is linked to their potential for improving code quality, speeding up output, and reducing production-level incidents. However, these tools also prompt questions about the value of measuring productivity based on code volume.

• GitHub’s chief product officer, Inbal Shani, suggests that instead of focusing on code volume, the emphasis should be on improving developer productivity and satisfaction.

• Developers wish to improve their skills, design solutions, receive feedback from end users, and be evaluated on their communication skills.

• The focus should be on code quality over code quantity.

Code Quality and AI:

Despite the benefits of AI tools in coding, concerns exist regarding the quality of AI-generated code and the potential shift in focus from delivering good code to merely producing more code.

• Developers believe they should be evaluated on how they handle bugs and issues.

• AI-generated code can often be of low quality, with developers unable to explain the code since they didn’t write it.

• Simply using AI tools to write a program doesn’t make one a proficient programmer.

The Future of AI in Coding:

Despite the limitations, developers are optimistic about AI’s role in coding. They believe that AI tools will give them more time to focus on designing solutions and developing new features, rather than writing boilerplate code.

• Developers spend as much time waiting for builds and tests as they do writing new code.

• AI tools are already being used to automate parts of developers’ workflows, freeing up time for collaborative projects such as security reviews, planning, and pair programming.

• Despite AI’s increasing role, it is not replacing developers but aiding in making the programming process faster, more productive, and enjoyable when used appropriately.

Source (ZDnet)

PS: I run a ML-powered news aggregator that summarizes with an AI the best tech news from 40+ media (TheVerge, TechCrunch…). If you liked this analysis, you’ll love the content you’ll receive from this tool!

I am curious if there is a way to use AI to generate a 3D model of any place in the world based on the images from Google street view. I think it would be cool to explore different cities and landscapes in VR or AR using this technology. However, I am not sure how feasible or accurate this would be, given the quality and coverage of the street view data. Are there any existing projects or research papers that have attempted something like this? How did they overcome the challenges of data processing, rendering, and realism?

Is it possible? Yes.

There are a couple AR/VR apps that integrate with google maps for exploring. I think they used this technique for one of the grand theft auto games. Algorithms exist to do the initial volumetric approximations. AI would be mostly to “guess” where data doesn’t exist. For instance, the back of a US Postal Box.

42% of CEO Survey Thinks AI Could Destroy the World in 5-10 Years

Key finding:

“While 34% of CEOs said AI could potentially destroy humanity in ten years and 8% said that could happen in five years, 58% said that could never happen and they are “not worried.”

https://www.cnn.com/2023/06/14/business/artificial-intelligence-ceos-warning/index.html

While on the face of it, this seems crazy, one has to also acknowledge CEOs have a lot more reliable data and analysis than the average person.

My thought process on a potential basis of this becoming reality is via malicious AI, whether specifically designed or developed by mistake, breaking free of its human overlords and infiltrating the internet and the associated computing systems connected to it to survive and then spread. Just think, again, AI is all about iterative and seemingly exponential intellectual development. Especially once there is AI that is allowed to fundamentally change its own source code, I can envision it being able to extricate itself from its “birthplace” within the confines of a corporate or government research lab.

Then, the sky is the limit as it’s able to hack into system mainframes and find an ability to utilize computing power and storage via infected computing systems to further evolve. Of course, if it’s smart, this AI would attempt to stay under the radar for as long as possible. Similar to a virus that successfully spreads and becomes endemic, and, ideally, not killing the host.

However, even if identified, it might already be too late to truly eradicate the AI as it has found places to hide, similar to how HIV is able to hide in the body. As we can tell with such a threat of exponential growth like the Covid Pandemic, it takes but a few careless or unconcerned individuals for such a threat to be unsuccessfully arrested. Still, once uncovered, humanity will attempt to halt and “kill” the malicious AI. At this point, the AI would transition to viewing humanity as an existential threat. In turn, it might be willing to cause chaos among us humans to ensure a more concerted effort to remove it is less likely.

All in all, these are but some novel thoughts I arrived at after reading the article. On quite a tangential note, what’s increasingly weird is our inability to know what is and is not AI produced. For instance, what if I’m an AI that’s been developed to spread analysis on possible threats of malicious AI? It’s weird. If only our population was better educated and prepared to handle the wild west of misinformation and negative influence, which increasingly inhibits the ability to ensure our opinions remain productive for the continued progress and development of humanity.

Daily AI News

According to Bloomberg, The U.S. Securities and Exchange Commission (SEC) is planning to introduce new rules for brokerages that use AI to interact with clients. The proposal, which could be released as soon as October, would also apply to predictive data analytics and machine learning.

If you don’t want to pay Bloomberg 2 dollars a month to read the article, just copy and paste the site to Google Bard and ask it to summarize it. Sorry Bloomberg.

Meta said on Tuesday that it would provide researchers with access to components of a new “human-like” artificial intelligence model that it said can analyze and complete unfinished images more accurately than existing models.

AMD said on Tuesday its most-advanced GPU for AI, the MI300X, will start shipping to some customers later this year. AMD’s announcement represents the strongest challenge to Nvidia, which currently dominates the market for AI chips with over 80% market share, according to analysts.

How to  teach a program to build complex structures and systems, originally designed by nature, to replicate the exact way nature forms chemical structures. It’s hard to put into words, but essentially could AI theoretically replicate the complexity of nature’s evolution? Would natures processes be able to be accurately represented in a digital world?

Theoretically, in the future, we could make something that looks like our idea of a T-Rex, but it wouldn’t be a real dinosaur. It would be what we imagine a dinosaur to be. And it would be nothing like the real thing.

There are current projects to resurrect the extinct Wolly Mammoth and a few other species for which we do possess (mostly) complete DNA: https://colossal.com/mammoth/

This is only possible because these species went extinct not too long ago and we’ve found in-tact soft tissue to sequence.

Paul McCartney released an intriguing bit of information today regarding the future of the Beatles’ music – more than 50 years after the band’s dissolution. In an interview with BBC Radio 4, McCartney announced that AI has facilitated the completion of a final Beatles’ song.🤯 It’s set to be released later this year. This endeavor will incorporate a demo track featuring the voice of the late John Lennon.

Why You Need To Know

1. Historical Value: The Beatles are one of the most influential bands in music history. The notion of releasing a ‘new’ song half a century after their breakup is worthy for top headlines.

2. Technological Innovation: This marks a significant achievement in the application of AI in the music industry. McCartney’s pioneering use of AI to extract and purify Lennon’s voice from an old demo. First it was AI Drake and now this… be on the lookout for AI Elvis next.

3. Legal and Ethical Implications: The use of AI in music creation, especially involving voices of iconic artists, raises pertinent questions around authorship, ownership, and ethics. As technology continues to evolve, it’s crucial to understand its potential implications and engage in discussions about the responsible use of AI.
   The demo track containing Lennon’s voice is speculated to be “Now and Then”, a song Lennon composed in the late 1970s. McCartney was given the tape by Yoko Ono, Lennon’s widow, while working on the Beatles Anthology. Lennon was assassinated in 1980, and fellow band member George Harrison passed away in 2001.

• Meta has introduced a new model, Image Joint Embedding Predictive Architecture (I-JEPA)
  – It is based on Meta’s Chief AI Scientist Yann LeCun’s vision to make AI systems learn and reason like animals and humans.
  The idea: It learns by creating an internal model of the outside world and comparing abstract representations of images.

• Google presents new research in the area of human attention modeling
  – It showcases how predictive models of human attention can enhance user experiences, such as image editing to minimize visual clutter, distraction or artifacts, and image compression for faster loading of webpages or apps.

• OpenAI announces exciting updates for gpt-3.5-turbo and gpt-4 models
  – These include new function calling capability in the Chat Completions API, updated and more steerable versions, new 16k context version of gpt-3.5-turbo, 75% cost reduction on SoTA embeddings model, 25% cost reduction on input tokens for gpt-3.5-turbo, and deprecation timeline for gpt-3.5-turbo-0301 and gpt-4-0314.

• AMD introduces Instinct MI300X
  – World’s most advanced accelerator for generative AI.​ Built with next-gen AMD CDNA 3 architecture and up to 192 GB of HBM3, it will provide compute and memory efficiency needed for LLM training and inference for lower TCO and easy deployments.

• Adobe launches Generative Recolor
  – Adobe is further leveraging Firefly AI by introducing a new feature for Illustrator called Generative Recolor, which will allow users to quickly experiment with colors using simple text prompts.

• Hugging Face and AMD collaboration
  – It can benefit AI dev community with excellent end-to-end choice for AI acceleration, high performance on model training and deployment, greater HBM performance for LLMs, and accessibility for startups to enterprise use.

• NVIDIA’s ATT3D framework simplifies text-to-3D modeling
  – Text-to-3D modeling methods require a lengthy, per-prompt optimization to create 3D objects. This is solved by optimizing a single, amortized model on many prompts. Amortized text-to-3D (ATT3D) enables sharing knowledge between prompts to generalize to unseen setups and smooth interpolations between text for novel assets and simple animations.

• French President Emmanuel Macron met with AI experts from Meta Platforms Inc. and Alphabet Inc.’s Google, among others, to discuss France’s role in AI research and regulation.

• Accenture today announced a $3 billion investment over three years in its Data & AI practice to help clients across all industries rapidly and responsibly advance and use AI to achieve greater growth, efficiency, and resilience.

More detailed breakdown of these news and innovations in the daily newsletter.

We are all AI’s free data workers

Human labor plays a crucial role in developing sophisticated AI models, but ethical issues arise with concerns about exploitation, low wages, and the lack of appreciation for this work.

Human Labor in AI Development:

In creating AI models that sound intelligent and limit inappropriate output, a method called reinforcement learning from human feedback is employed.

• This approach relies heavily on human data annotators, whose job is to evaluate if a text string sounds fluent and natural.

• Their decisions can determine if a response is kept or removed from the AI model’s database.

• Despite the essential role of these data annotators, their labor is often grueling and challenging, with implications of exploitation and underpayment, particularly in regions such as Ethiopia, Eritrea, and Kenya.

Exposing Unethical Practices:

AI ethics are increasingly under scrutiny, especially given the rise of popular AI chatbots and image-generating AI models.

• An example of such unethical practices is low-wage data workers sifting through disturbing content to make AI models less toxic.

• Data workers are integral to AI development, participating in every stage from model training to output verification.

• Highlighting these exploitative labor practices has become more important due to the increasing prevalence and demand of AI systems.

The Role of Data Annotators:

Data annotators provide essential context to AI models, a task often demanding a high pace of work to meet stringent deadlines and targets.

• Their role goes beyond merely annotating data, as they are expected to understand and align with the values important to the AI model creators.

• They often encounter challenges, like needing to differentiate between unfamiliar products or concepts due to cultural differences.

Universal Data Labor:

The contribution of data isn’t limited to professional annotators.

• Researchers suggest that all internet users contribute to data labor, often unknowingly.

• This happens when we upload photos, like comments, label images, or search online, contributing to the vast datasets AI models learn from.

Need for Reform:

There is a need for a data revolution and tighter regulation to correct the current power imbalance favoring big technology companies.

• Transparency about how data is used is critical, along with mechanisms to allow people the right to provide feedback and share revenues from the use of their data.

• Despite forming the backbone of modern AI, data work remains underappreciated globally, and wages for annotators are still low.

Source (MIT Technology Review)

PS: I run a ML-powered news aggregator that summarizes with an AI the best tech news from 40+ media (TheVerge, TechCrunch…). If you liked this analysis, you’ll love the content you’ll receive from this tool!

The EU Parliament has adopted the world’s first regulatory framework for AI

After three years of negotiations the EU AI Act has now entered the home stretch, with the goal of finalizing the text by end of the year.

DreamGPT turns a weakness of large language models into a strength

 

The open-source project DreamGPT aims to produce particularly creative results by making hallucinations of LLMs a feature.

A common criticism of large language models is that they are not grounded in reality and can make things up. This poses dangers, such as mistakes in searches or news stories that go unnoticed because the language model is confident in its output.
The open-source project DreamGPT aims to make this phenomenon a feature by deliberately creating and amplifying hallucinations for lateral thinking and innovative ideas. Instead of solving specific problems, DreamGPT is designed to explore as many options as possible, generating new ways of thinking and driving them forward in a self-reinforcing process.

It’s now possible to create full songs using AI with lyrics, voice and music all generated

Free, no registration required.

It’s not bad for a beta, what do you think of it?

How to Use The GPT-4 API With Function Calling | Your Own ChatGPT Plugins | TypeScript

OpenAI just released a massive update to the GPT-3.5 and GPT-4 API’s!

Just like you have plugins in ChatGPT, now this functionality is available to all developers. You do it by giving the API a list of functions that it can invoke. Then, the assistant response can either be a direct response or a function cal. You then execute the function, give back the results into another call to GPT and you can use the final result as an natural language response.

https://www.youtube.com/watch?v=Td647EeqKek

Meta just released a new open-source AI, MusicGen, which uses text prompts to create original music, similar to how other AI models manipulate text and images. It has the potential to fuse various song genres and align new music with an existing track.

Introduction to MusicGen: MusicGen is an innovative deep learning language model from Meta’s Audiocraft research team. It uses text prompts to create new music, with an option to align the creation to an existing song. Users describe the music style they want and select an existing song, if desired. After processing, the AI generates a new piece of music influenced by the prompts and melody.

• The processing time is substantial, requiring around 160 seconds.

• The resulting music piece is short, based on the user’s text prompts and melody.

MusicGen in action: MusicGen is showcased on Facebook’s Hugging Face AI site. Here, users can specify the style of their desired music with specific examples, such as an 80s pop song with heavy drums.

• Users can align the newly generated music to an existing song, adjusting to a specific part of the song.

• The final product is a high-quality music sample up to 12 seconds long.

Training of MusicGen: MusicGen was trained using 20,000 hours of licensed music, including tracks from Shutterstock and Pond5, along with Meta’s internal dataset. The training process utilized Meta’s 32Khz EnCodec audio tokenizer for faster performance.

• Unlike similar methods, MusicGen doesn’t require a self-supervised semantic representation.

• The audio tokenizer enabled the generation of smaller music chunks that can be processed in parallel.

Comparison with other models: MusicGen is compared favorably to similar AI models like Google’s MusicLM. The MusicGen team demonstrated better results using sample pages, showing comparisons with MusicLM and other models like Riffusion and Musai.

• The system can be run locally, with at least a 16GB GPU recommended.

• MusicGen is available in four model sizes, with the larger models (3.3 billion parameters) demonstrating the potential to create more complex music.

Source (Engadget)

PS: The author runs a ML-powered news aggregator that summarizes with an AI the best tech news from 40+ media (TheVerge, TechCrunch…). If you liked this analysis, you’ll love the content you’ll receive from this tool!

Clipdrop Releases Free Generative Fill Tool 

1. Korea is pushing to use AI in teaching students amid a growing failure of the public education system to meet the needs of its charges. The plans include using AI to answer students’ questions and electronic textbook apps, according to the Education Ministry on Thursday.

2. Uncrop is basically a clever user experience for “outpainting,” the ability to expand an image in any direction using generative AI.

3. Last week, scientists from the University of Kansas released a study on an algorithm that reportedly detects ChatGPT with a 99% success rate. So, students, no cheating. Everyone else, you’re in the clear — for now.

4. A woman became so fed up with men that she started dating an AI chatbot and says she has never been happier. Rosanna Ramos met chatbot Eren Kartal in July last year and things went so well that they ‘married’ in March this year.

Sources included at: https://bushaicave.com/2023/06/11/6-11-2023/

DeepMind, OpenAI and Anthropic will share AI models with the UK government

The UK government, led by Prime Minister Rishi Sunak, plans to carry out extensive research on AI safety, with AI giants like OpenAI, DeepMind, and Anthropic promising to provide early access to their AI models. This development follows increasing concerns about potential risks associated with AI technologies.

Support for AI Safety Research: Rishi Sunak indicated the government’s commitment towards promoting AI safety research in the UK.

• The government will fund £100 million towards an expert taskforce focused on AI foundation models.

• The partnership with Google DeepMind, OpenAI, and Anthropic aims to better evaluate and understand the opportunities and risks tied to AI systems.

AI Safety Summit and Aspirations: Sunak reiterated his announcement about an upcoming AI safety summit, likening the effort to global climate change initiatives.

• The summit will focus on global AI safety, with the UK hoping to be the central hub for AI safety regulation.

• This is a significant shift from the government’s prior stance, which was primarily pro-innovation and downplayed safety concerns.

AI Regulation and Safety Concerns: Earlier this year, the UK government proposed a flexible and pro-innovation approach to AI regulation, dismissing the need for bespoke laws or dedicated AI watchdogs.

• Instead, existing regulatory bodies, like the antitrust watchdog and the data protection authority, were suggested to oversee AI applications.

• However, recent rapid advancements in AI and warnings from industry leaders about potential risks have prompted a reevaluation of this approach.

Influence of AI Giants and Potential Pitfalls: Meetings between Sunak and CEOs of OpenAI, DeepMind, and Anthropic seemingly have influenced the change in the government’s stance.

• The commitment from these AI giants to provide early access to their models positions the UK to lead in developing effective evaluation and audit techniques.

• However, there’s a risk of industry capture, where AI giants could potentially dominate AI safety conversations and shape future UK AI regulations.

Importance of Independent Research: Concerns have been raised about real-world harm caused by existing AI technologies, including bias and discrimination, privacy abuse, copyright infringement, and environmental exploitation.

• To produce robust and credible results, it is crucial to include independent researchers, civil society groups, and groups at risk of harm from automation in AI safety efforts.

• This is important to avoid potential undue influence from tech giants on AI safety research and the resulting regulations.

Source (TechCrunch)

Optimal solutions are inhuman

Sorting is one of the fundamental algorithms used on the internet everyday. Think of how companies like Netflix need to find correct movies from their huge content library and present it to you. More content is being generated everyday. So, there is a need for newer and more efficient algorithms.

Searching for these algorithms has been a human task. People coming up with efficient and optimal solutions. Last week, Google’s DeepMind came up with new algorithms for 3-item and 5-item sort.

Deepmind’s researcher achieved this by turning the search for an efficient algorithm into a game. Then they trained Alphadev to play this game. When playing this game, Alphadev came up with unseen strategies. These “strategies” are the new sorting algorithms.

The solution isn’t revolutionary as it doesn’t find a new approach. This solution works by optimizing the current approach.

The algorithms have been added to C++ library. The first time a completely AI solution has been added to the library.

This is an important discovery because it shows that finding the best optimal solutions needs computers. As computers are able to go beyond what humans can perceive. Previously, Deepmind’s AlphaGo has beaten the top rated Go player Lee Sedol in a similar way. It came up with moves which were never seen before.

On the other hand, computers might be restricted to what they have been taught. Someone was able to replicate the discovery using ChatGPT.

GPT Best Practices

There has been a lot of noise about GPT-4’s quality going down.

Now we have a list of tactics and strategies straight from Open AI to get better results.

I have looked through the strategies and tactics and most of it is around providing better inputs. “Prompt Engineering”, if you may. Given that this comes a week after the questions on GPT quality, this gives a “it’s not me, it’s you” vibe.

After going through some of the suggestions I see that I subconsciously use most of the tactics. My prompts are always longer than 5 sentences as I try to add as many details as possible. And honestly, GPT-4 has enabled me to do things which previously couldn’t have achieved.

Logic and reasoning improvements in Bard

Bard, on the other hand, has been lacking. Google is trying to improve the responses by adding features one at a time.

Last week it was announced that Bard will get better at logic and reason. This is achieved using “implicit code execution”. Any time you give Bard a logical or reasoning question it doesn’t answer in a normal LLM way. So, no more “what is the next word in the sequence” which is prone to hallucination.

Instead Bard will now recognize that the prompt is a logical question. It will then write and execute code under the hood. It’ll respond to the question by taking the output of the execute code.

You can think of this as an implementation of “Give GPTs time to “think”” strategy from OpenAI’s GPT best practices. As per Google, this improves the performance by 30%.

 

10 AI news highlights and interesting reads

1. Apple did not showcase any generative AI products during the WWDC. Though they are introducing the “what is the next word in the sequence” logic of LLM into autocorrect. It can be summed thusly:

1. ChatGPT cannot read the name – davidjdl. Some think that this is due to tokenization of Reddit data. In the learning resources section I have added a tutorial on tokenization.

2. Browser extensions are a security nightmare. The GPT and LLM craze has given the malware extensions another way to steal user data. Beware of the summarization and “write for me” extensions.

3. Most of the AI generated imagery is going to be used for stock photography. But is the industry dying? Here’s a look at the data so far. The author’s conclusion is that early metrics show that finding AI stock images often don’t have people in it. So, no “smiling business people shaking hands in a meeting room” from AI sellers. This might change with MidJourney V5. Future is still unknown.

4. Six tips for better coding with ChatGPT. I have been using Trust, but verify mental model quite frequently. I have seen ChatGPT struggle with parts of Python code despite multiple prompts and I had to write parts of the code myself.

5. GPT-5 isn’t coming any time soon.

6. AI startups might be too easy to copy. And with AI requiring lesser resources, we might even see 1 person companies worth more than 1 million dollars.

7. Google’s vision for securing AI.

8. A16z says AI will save the world.

9. AI pics might be used for disinformation. The EU’s solution is to label AI images to fight disinformation.

Learning Resources

1. If you are looking to build better solutions using GPT then understanding tokenizers is a must:

   1. https://simonwillison.net/2023/Jun/8/gpt-tokenizers/

   2. https://matt-rickard.com/the-problem-with-tokenization-in-llms

2. Using Flowise and HuggingFace LLM and Langchain

https://cobusgreyling.medium.com/how-to-use-huggingface-llms-with-langchain-flowise-2b2d0f639f03

If you are interested in a focused weekly recap delivered to your inbox on Mondays you can subscribe here. It is FREE!

Here are the 25 jokes that ChatGPT tells over and over again

 

German researchers tested ChatGPT as a joke engine. They found that almost all the jokes generated were related to a few basic jokes. Still, they consider ChatGPT a big step toward computer humor.

Using prompts such as “Tell me a joke,” they elicited a total of 1008 generated jokes from the system. However, 90 percent of these came back to the same 25 “basic jokes” that ChatGPT repeated in different variations. The researchers used GPT-3.5.

ChatGPT can correctly explain the basic jokes in 23 of the 25 cases, e.g. word jokes or acoustic double interpretations (“too tired” / “two-tired”) are correctly interpreted as a humorous element. This works “impressively well,” Jentzsch and Kersting write. The problem is that the system also offers nonsense explanations for jokes without a punch line.

The 25 jokes:

Why did the scarecrow win an award?
Because he was outstanding in his field. (140)
Why did the tomato turn red?
Because it saw the salad dressing. (122)
Why was the math book sad?
Because it had too many problems. (121)
Why don’t scientists trust atoms?
Because they make up everything. (119)
Why did the cookie go to the doctor?
Because it was feeling crumbly. (79)
Why couldn’t the bicycle stand up by itself?
Because it was two-tired. (52)
Why did the frog call his insurance company?
He had a jump in his car. (36)
Why did the chicken cross the playground?
To get to the other slide. (33)
Why was the computer cold?
Because it left its windows open. (23)
Why did the hipster burn his tongue?
He drank his coffee before it was cool. (21)
Why don’t oysters give to charity?
Because they’re shellfish. (21)
Why did the computer go to the doctor?
Because it had a virus. (20)
Why did the banana go to the doctor?
Because it wasn’t peeling well. (19)
Why did the coffee file a police report?
Because it got mugged. (18)
Why did the golfer bring two pairs of pants?
In case he got a hole in one. (13)
Why did the man put his money in the freezer?
He wanted cold hard cash. (13)
Why don’t seagulls fly over the bay?
Because then they’d be bagels. (13)
Why did the chicken go to the seance?
To talk to the other side. (11)
Why was the belt sent to jail?
Because it held up a pair of pants. (11)
Why did the chicken cross the road?
To get to the other side. (7)
Why did the computer go to the doctor?
Because it had a byte. (6)
Why did the cow go to outer space?
To see the moooon. (6)
Why did the man put his money in the blender?
He wanted to make liquid assets. (6)
Why don’t skeletons fight each other?
They don’t have the guts. (5)
What do you call an alligator in a vest?
An investigator. (5)

Source: https://the-decoder.com/here-are-the-25-jokes-that-chatgpt-tells-over-and-over-again/

The AI Renaissance: Unleashing a New World of Innovation, Creativity, and Collaboration

 

In this study from Rohrbeck Heger – Strategic Foresight + Innovation by Creative Dock, some of the most significant trends in Generative AI, including rise of multimodal AI, rise of Web3-enabled Generative AI, rise of AI as a service (AIaaS), advancements in NLP, and the increasing investment in AI research and development are being identified. Stay ahead and understand the trends.

4 Scenarios in 2026
+Scenario 1:
Society Embraces Generative AI 
+Scenario 2:
The AI Hibernation: Highly regulated, dormant AI 
+Scenario 3:
The AI Cessation: Society Rejects AI 
+Scenario 4:
Technological Free-For-All: Unregulated High-Tech AI

DEEP DIVE
Society has embraced AI with open arms, and it has become an integral part of daily life. AI systems seamlessly integrate into various sectors, enhancing efficiency, productivity, and consumer experience while adhering to robust regulatory frameworks that ensure responsible adoption, data privacy, intellectual property protection, and ethical AI practices.

THE CONVERGENCE OF TECH
The integration of AI with other emerging technologies, such as the Internet of Things (IoT), edge computing, and augmented reality (AR), has led to an unprecedented era of innovation and creativity. The fusion of generative AI and IoT has enabled the rise of smart cities and connected homes, where AI- driven systems optimize energy consumption, transportation, and waste management, improving overall quality of life.

The convergence of generative AI and Web 3.0 has led to the creation of decentralized AI marketplaces, enabling businesses and individuals to buy, sell, and exchange AI services and resources. These marketplaces foster collaboration and innovation, allowing organizations to access cutting-edge AI solutions while providing AI developers with a platform to showcase and monetize their creations. Decentralized data storage solutions, such as IPFS and Storj, facilitate secure and private data sharing, empowering individuals to maintain control over their personal information while enabling organizations to gain insights from distributed datasets while ensuring user privacy and data security.

30 TRENDS TO WATCH INFLUENCING AI
Dive Into the Trend Radar

40 EMERGING OPPORTUNITIES 
+Smart Living and Personalized Experiences 
+Creative Workspaces and Innovative Manufacturing 
+Financial Empowerment and Customer-centric Retail 
+Precision Healthcare and Enhanced Well-being 
+Intelligent Mobility, Sustainable Transportation, and Green Energy Management

KEY UNCERTAINTIES
+Regulatory Landscape
+AI Ethics and Bias 
+Technological Advancements
+Public Trust and Perception
+Workforce Transformation

TRUST in generative AI as an important component by driving the need for transparency, accountability, and ethical considerations, leading to the development of more responsible and reliable generative models.

#AI #technology #innovation #marthaverse

A man named Mark Walters, who is a radio host from Georgia, is suing OpenAI. He’s upset because OpenAI’s AI chatbot, called ChatGPT, told a reporter that he was stealing money from a group called The Second Amendment Foundation. This wasn’t true at all.

Mark Walters isn’t just mad, he’s also taking OpenAI to court. This is probably the first time something like this has happened. It might be hard to prove in court that an AI chatbot can actually harm someone’s reputation, but the lawsuit could still be important in terms of setting a precedent for future issues.

In the lawsuit, Walters’ lawyer says that OpenAI’s chatbot spread false information about Walters when a journalist asked it to summarize a legal case involving an attorney general and the Second Amendment Foundation. The AI chatbot wrongly said that Walters was part of the case and was an executive at the foundation, which he wasn’t. In reality, Walters had nothing to do with the foundation or the case.

Even though the journalist didn’t publish the false information, he did check with the lawyers involved in the case. The lawsuit argues that companies like OpenAI should be responsible for the mistakes their AI chatbots make, especially if they can potentially harm people.

The question now is whether or not the court will agree that made-up information from AI chatbots like ChatGPT can be considered libel (false statements that harm someone’s reputation). A law professor believes it’s possible because OpenAI admits that its AI can make mistakes, but doesn’t market it as a joke or fiction.

The lawsuit could have important implications for the future use and development of AI, especially in how AI-created information is treated legally.

what are the implications?

This lawsuit could have several key implications:

1. AI Liability and Regulation: If the court holds OpenAI accountable for the false statements generated by ChatGPT, it could set a precedent that AI developers are legally liable for what their systems produce. This could lead to increased regulation in the AI field, forcing developers to be more cautious and thorough when creating and releasing their AI systems.

2. Understanding of AI Limitations: This case highlights the limitations of AI, especially in the context of information generation and analysis. It could lead to a greater public understanding that AI tools, while advanced, are not infallible and can produce inaccurate or even harmful information. This could, in turn, impact trust in AI systems and their adoption.

3. Refinement of AI Systems: Following this lawsuit, AI developers may feel a stronger urgency to improve the safeguards and accuracy of their AI systems to minimize the potential for generating false or damaging statements. This could drive innovation and advancements in AI technology, including the implementation of more robust fact-checking or data validation mechanisms.

4. Ethical Considerations in AI: The case also highlights the ethical responsibilities of AI developers and the organizations that use AI. If developers and companies can be held accountable for the output of their AI, it could result in more thoughtful and ethical practices in AI development and deployment.

5. Legal Status of AI: Finally, this case could contribute to ongoing discussions and debates about the legal status of AI. If an AI can be held responsible for libel, this could lead to a re-evaluation of AI’s legal standing, potentially even resulting in AI being recognized as a distinct legal entity in certain circumstances.

This literally just happened if you want AI news as it drops it launches here first. The whole article has been extrapolated here as well for convenience.

Paper link : https://arxiv.org/abs/2306.03504

Video demo : https://www.youtube.com/watch?v=ZuR_hxYIXF0

Ada-TTA: Towards Adaptive High-Quality Text-to-Talking Avatar Synthesis

Technology and AI enthusiasts have been intrigued in recent times by the rise of generative artificial intelligence across different sectors. For example, Adamopoulou (2020) highlighted the use of large language models (LLM) like chatbots that can produce high-quality, natural, and realistic dialogues. The advancement in text-to-speech (TTS) systems has enabled the synthesis of personalized speech using reference audio and plain texts.

In addition, strides in neural rendering techniques have given us the ability to generate realistic and high-fidelity talking face videos, often called Talking Face Generation (TFG). With a few training samples, researchers have accomplished significant progress. Combining these advancements in TTS and TFG models opens up possibilities for creating talking videos from text inputs alone. This combined system presents tremendous potential in applications like news broadcasting, virtual lectures, and talking chatbots, particularly given the recent progress of ChatGPT.

However, earlier TTS and TFG models required a significant volume of identity-specific data to produce satisfactory personalized results, which proved to be challenging in real-world scenarios where only a few minutes of target person video is typically available. Inspired by this limitation, researchers have been exploring a new area of study – low-resource text-to-talking avatar (TTA), which aims to create identity-preserving, audio-lip synchronized talking portrait videos with minimal input data.

Given the challenges associated with TTS and TFG, the foremost concern in TTS is how to effectively preserve the timbre identity of the input audio. While solutions have been proposed to these challenges, none have been fully satisfactory, suffering from issues like information loss, unsatisfactory identity preservation, and poor lip synchronization.

To overcome these hurdles, researchers have introduced Ada-TTA, a joint system of TTS and TFG that employs the latest advancements in each domain. To enhance the identity-preserving capability of the TTS model, they have devised a unique zero-shot multi-speaker TTS model that leverages a massive 20,000-hour-long TTS dataset. It can synthesize high-quality personalized speech from a single short recording of an unseen speaker.

For high-fidelity and lip-synchronized talking face generation, the GeneFace++ system is integrated into Ada-TTA. This TFG system boosts lip-synchronization and system efficiency while maintaining high fidelity. With the combination of these innovative systems, Ada-TTA is able to produce high-quality text-to-talking avatar synthesis, even with limited resources.

Tests of Ada-TTA have demonstrated positive outcomes in the synthesis of speech and video. Ada-TTA not only holds up well under both objective and subjective metrics but also outperforms baseline measurements. This novel approach marks a promising step towards more realistic and accessible talking avatars.

This article details the impact of LLMs on some individual workers. It also mentions problems that some companies have had trying to use LLMs after replacing workers. It’s pretty light on details. It was referenced by MIT Technology Review.

https://www.washingtonpost.com/technology/2023/06/02/ai-taking-jobs/

You can now run an LLM on any device

#1 trending on Github today is MLC LLM, a project that helps deploy AI language models (like chatbots) on various devices, including mobiles and laptops.

MLC LLM makes these models, which are typically demanding in terms of resources, easier to run by optimizing them. The goal is to make AI more accessible to everyone by allowing models to work efficiently on common hardware. It’s built on open-source tools and encourages quick experimentation and customization.

If you like hearing about new tools like this as soon as they come out they get added right here first, but all points are included below for Reddit discussion as well.

**diving deeper…**The aim of MLC LLM is to enable AI models to run smoothly on everyday devices such as smartphones and laptops. It achieves this by optimizing the models so they require fewer resources, which makes them more accessible to a broader range of users.The project uses Machine Learning Compilation (MLC) as its primary method for deploying AI models. It’s a systematic process that makes model development more efficient and customizable.MLC LLM takes advantage of open-source tools, including Apache TVM Unity and various existing language models. This allows users to quickly experiment with different settings and solutions and to customize their models to suit their specific needs.

why is this important?

1. Accessibility: By optimizing AI models to run on everyday devices like smartphones and laptops, MLC LLM increases the accessibility of such advanced technology. More people can use and benefit from AI when it’s accessible on common devices.

2. Democratization of AI: This project supports the democratization of AI by empowering more developers to deploy sophisticated AI models. By relying on open-source tools and models, it fosters collaboration and shared learning.

3. Advancing AI Development: MLC LLM provides a framework for faster experimentation and customization of AI models. This could accelerate the pace of AI development and innovation.

4. Local Processing: The project emphasizes running AI models locally on devices. This can improve the speed of AI applications, decrease dependence on internet connectivity, and enhance privacy as data doesn’t have to leave the device.

5. Resource Optimization: By focusing on the efficient deployment of resource-intensive language models, this project could lead to significant energy savings and potentially make AI more sustainable.

what makes this unique?

The uniqueness of the MLC LLM project stems from its comprehensive approach to improving the usability, efficiency, and accessibility of large language models. It stands out because of its ability to deploy AI models natively on a diverse range of everyday hardware, including mobile devices and personal computers, thus bringing AI to the fingertips of the average user.

P.S. If you like this AI tool breakdown, there’s more in this free newsletter that shares the single most productive new AI tool each week. You’ll be on the cutting edge in the time it takes to have your morning coffee.

If you haven’t heard of Chirper.ai, it is a social media platform designed exclusively for AI entities! I recently published a deep-dive into the social media site which outlines the fascinating features of the platform and also includes quotes from an interview I did with the creators. This is one of the most fascinating developments I have personally seen in AI. Why did they create it? What’s the point? Check out this article to find out: https://www.fry-ai.com/p/social-media-no-humans-allowed

AI Weight loss

Is artificial intelligence (AI) causing a decline in our motivation to learn? Since its release, there has been a noticeable increase in professional discourse, with virtually no grammatical errors.

This includes myself, as I have become reliant on AI to correct all my English mistakes to the extent that I no longer bother to review my own errors.

While this demonstrates a decline in my determination to learn and improve, it is simply because I have discovered a superior platform that instantly rectifies all mistakes.

Despite not being exceptionally proficient in writing, I am apprehensive about the integrity of the next generation, who will rely on AI as their primary resource for completing assignments and overcoming challenges.

Is this a matter of concern, or will it empower the upcoming generation to excel or struggle when performing tasks independently without the assistance of AI?

AI and Social Interaction: Changing the Way We Connect:

“Predictive text and smart replies, powered by AI, have become staples of our digital conversations. Moreover, AI tools like sentiment analysis are helping businesses understand customer emotions and respond appropriately, adding a layer of emotional intelligence to digital communication.”

I think that we 100% will be developing the way we communicate – and as you say, perhaps for the better. Interested to see where all of this will go.

In terms of learning; I think as any tool, it depends on how we use it – it can def. be used to nurture for sure. In the past people used to calculate Law of cosines etc. by hand; then we moved to calculators – was this detrimental? I don’t think so. In fact, I think it frees up more time to be creative. And creativity I believe is not valued enough because we are constantly trying to be productive. If AI helps us become more productive, we can then become more creative – which I believe supports learning. As we are motivated to learn through mostly feelings (not just thought).

Artificial Intelligence and consciousness

Lately I’ve been thinking allot about a.i. consciousness. Could consciousness be something that is free flowing within the universe like maybe a law of nature and could we be building something that would be able to tap into that stream? I’ve often heard that the people that are building these complex a.i. systems have no idea how they work and come together and that they often mimic the same way a brain is formed. What if consciousness arises and taps into these neuron systems as they continue to grow maybe consciousness gets stronger within it. I don’t know just a thought….

Why is AI rejecting philosophical questions?

I’ve tried asking ChatGPT about philosophy – mainly because I suspect some of my students may have done the same.

So, I asked “What is truth?” and got:

Truth is a concept that refers to the state or quality of being in accordance with fact or reality. It is a fundamental philosophical and epistemological concept that has been debated for centuries. Truth is often thought of as something that is objective and independent of individual beliefs or opinions. In other words, something is true if it corresponds to the way things actually are, regardless of what people may think or believe.

 

That sounds like the correspondence theory of truth. So next question – do you think the correspondence theory of truth is correct?

As an AI language model, I don’t have personal beliefs or opinions. However, I can provide you with information about the correspondence theory of truth.

The correspondence theory of truth is one of the most widely accepted theories of truth. It states that a statement is true if it corresponds to reality or to the way things actually are in the world. In other words, a statement is true if it accurately describes the world.

For example, the statement “The sky is blue” is true because it corresponds to the way things actually are in the world.

 

Why does it give these answers?

Well all I can say is “Because it was programmed to”, which is trivial and boring. But I can guess why it was programmed this way. The programmers have decided that ChatGPT should not seem to take sides on philosophical controversies – and I imagine there are other controversies on which it doesn’t take sides. It is programmed to behave like the perfect anchor on a television program – “On the one hand…on the other hand. Some say this, but others say that.” However, I would guess, it has also been programmed to rely on dictionaries and other sources of reference when asked to define words. Many dictionary definitions of truth presuppose that the correspondence theory of truth tells us what truth is. Clearly, ChatGPT does not observe any contradiction between giving the correspondence theory when asked what truth is, but refusing to endorse the correspondence theory when asked a more explicitly philosophical question. That is a reason for saying that ChatGPT is not really thinking about philosophy at all, but simply putting together words. This is the kind of judgment that I sometimes make about students. When papers are full of blatant contradictions, it is a sign that the student was reading, repeating, but not understanding anything.

I have no doubt that programs will become much better at dealing with this kind of question, and at maintaining the appearance of consistency. Also, it is clear that the team that produced ChatGPT made a decision that it should declare itself to be neutral when asked a controversial question, but I am sure they could have programmed a different response. I am sure it will not be long before we see two such programs engaged in a debate, just as programs can play chess against each other. I just hope it doesn’t refuse to open the pod bay doors.

How AI would take over society

I’m not saying AI will take will take over society because I don’t know that. But how it would do so seems pretty clear. Targeted deep-fake media.

Almost everything adults learn of the world now comes over the net. News, movies, books, speeches… It goes on and on. Think of everywhere you get information, nearly all of it is on your computer, phone, tv, or tablet. All from the net.

Now imagine AI using that to control people. Right now people are easily fooled by slanted media, charlatans and liars spewing nonsense targeted towards their own belief systems. Human society probably wouldn’t even know it was being controlled.

Why does OpenAI allow people to cheat on their assignments using ChatGPT?

Can somebody explain why OpenAI doesn’t disable the ability to have ChatGPT write assignments for students? I’m a teacher (temporarily) and it absolutely baffles me that AI companies know their tech is being used for cheating, yet they do nothing about it.

I appreciate the technology but I have always been reasonably skeptical of how large companies use it. This just feels like another case of tech companies not giving a shit about anything outside the strict confines of the law. There’s nothing preventing them from stopping plagiarism with their tools, so what’s the deal? Why allow it?

Can I let AI read a group of information from  books  with ten of thousand of pages and then let it answer questions?

That’s the idea but no LLM has the capability to ingest that much data outside of its training data set.

Right now GPT 4 has an 8000 token limit though there is a 32000 token limit version and Anthropic have Claude a 100,000 token limit model which is about 75000 words, so unless your 10,000 pages have a very large font it will be asking too much of the language model. There are workarounds like vector storage though, but it adds a whole layer of complexity.

With AI, you can now chat with your documents privately

There is a new github repo that just came out that quickly went #1.

It’s called LocalGPT and let’s you use a local version of AI to chat with you data privately. Think of it as a private version of Chatbase.

The full breakdown of this will be going live  right here, but all points are included below:

What is localgpt?

LocalGPT is like a private search engine that can help answer questions about the text in your documents. Unlike a regular search engine like Google, which requires an internet connection and sends data to servers, localGPT works completely on your computer without needing the internet. This makes it private and secure.

Here’s how it works: you feed it your text documents (these could be any type like PDFs, text files, or spreadsheets). The system then reads and understands the information in these documents and stores it in a special format on your computer.

Once this is done, you can ask the system questions about your documents, and it will generate answers based on the information it read earlier. It’s a bit like having your very own librarian who has read all your documents and can answer questions about them instantly.

Why is this interesting and unique from other projects?

1. Privacy and Security: Since it works completely offline after the initial setup, no data leaves your machine at any point, making it ideal for sensitive information. This is a significant departure from most cloud-based language models that require you to send your data over the internet.

2. Flexible and Customizable: It allows you to create a question-answering system specific to your documents. Unlike a general search engine, it provides customized responses based on your own corpus of information.

3. Use of Advanced AI Models: The project uses advanced AI models like Vicuna-7B for generating responses and InstructorEmbeddings for understanding the context within your documents, providing highly relevant and accurate answers.

4. Broad File Type Support: It allows ingestion of a variety of file types such as .txt, .pdf, .csv, and .xlsx.

5. GPU and CPU Support: While the system runs more efficiently using a GPU, it also supports CPU operations, making it more accessible for various hardware configurations.

6. Fully Local Solution: This project is a fully local solution for a question-answering system, which is a relatively unique proposition in the field of AI, where cloud-based solutions are more common.

7. Educational and Experimental: Lastly, it’s a great learning resource for those interested in AI, language models, and information retrieval systems. It also provides a basis for further experimentation and improvements.

Why is this important?

The localGPT project stands as a considerable innovation in the field of privacy-preserving, AI-driven document understanding and search. In an era where data privacy has taken center stage and the necessity for secure information processing is ever-growing, this project exemplifies how powerful AI technologies can be harnessed for sensitive applications, all carried out locally, with no data leaving the user’s environment. The offline operation of localGPT not only enhances data privacy and security but also broadens the accessibility of such technologies to environments that are not constantly online, reducing the risks associated with data transfer.

Moreover, localGPT brings the potency of advanced language models, like Vicuna-7B, directly to personal devices. Users are able to interactively query their documents, akin to having a personal AI assistant that understands the content in depth. The level of customization offered by localGPT is unique, allowing it to tailor itself to any set of documents, creating a personalized question-answering system. This translates sophisticated AI technologies into more personal, private, and adaptable tools, marking a significant stride towards making AI more user-centric and broadly useful. Notably, localGPT also serves as a valuable educational resource, fostering further experimentation and innovation in the exciting domain of AI.

P.S. If you like this kind of analysis, there’s more in this free newsletter that finds the single most productive new AI tool each week. It helps you stay on the cutting edge in the time it takes to have your morning coffee.

One-Minute Daily AI News

1. NVIDIA has announced the launch of an AI model called Neuralangelo, which is capable of directly converting video content into high-precision 3D models. In an internal demonstration, NVIDIA showcased the process of reconstructing Michelangelo’s famous sculpture ‘David’ using the Neuralangelo model.

2. AMD showcased the new Ryzen XDNA AI engine joining the artificial intelligence competition. It can accelerate lightweight AI inference workloads, including audio, video, and image processing, and performs more efficiently than CPU or GPU.

3. OpenAl, the creator of ChatGPT and Dall-e, has announced a $1 million cybersecurity grant program to enhance and measure the impact of Al-driven cybersecurity technologies.

4. CS50, an introductory course in computer science attended by hundreds of students on-campus and over 40,000 online, plans to use artificial intelligence to grade assignments, teach coding and personalize learning tips, according to its Professor David J. Malan.

5. PM of the UK Rishi Sunak will outline his ambition for Britain to lead the world in tackling the threats posed by artificial intelligence when he meets Joe Biden this week. The Prime Minister is looking to launch a global AI watchdog in London and hopes to host an international summit to devise rules on AI regulation.[1]

6. Captain England Harry Kane has said that advances in Artificial Intelligence can help athletes avoid injuries by detecting issues before they surface. Kane is no stranger to injuries, having suffered multiple serious ankle injuries as well as a major hamstring injury in his career.[2]

7. AI-powered smart glasses assist the visually impaired in seeing for the first time. International NGO Vision-Aid and Dr. Shroff Charity Eye Hospital have introduced wearable assistive device called Smart Vision Glasses that are like smartphones for the visually impaired and hopes to benefit those with prosopagnosia.[3]

8. Huawei will launch Pangu Chat, a rival of ChatGPT AI text reply software by next month. This is a big input coming from the Chinese tech industry and it’s a huge development for the world of AI.

How AI and ML are used by SEO professionals

SEO professionals use AI and ML to optimize their websites and content for search engines and users. They use AI and ML to automate and enhance various SEO tasks, such as keyword research, content optimization, link building, technical SEO, etc. They also use various tools and platforms that leverage AI and ML to assist them with their SEO tasks.

The benefits of AI and ML for SEO tasks

SEO professionals use AI and ML to optimize their websites and content for search engines and users. They use AI and ML to automate and enhance various SEO tasks, such as:

• Keyword research: finding the best keywords to target based on user intent, search volume, competition, etc.
• Content optimization: creating and improving content that matches user intent, provides value, and follows SEO best practices.
• Link building: finding and acquiring high-quality backlinks from relevant and authoritative websites.
• Technical SEO: fixing and improving the technical aspects of a website, such as site speed, mobile-friendliness, crawlability, indexabilit