AI News #85: Week Ending May 16, 2025 with 25 Executive Summaries, Top 56 Links, and 10 Helpful Visuals

May 17, 2025

About This Week’s Covers

This week’s covers are in honor of my great friend John Bayalis, who is an incredible painter, photographer, and New Order fanatic. A few years ago, John decided to paint all 64 official Joy Division and New Order releases on 3×3 inch canvases. I’m thankful to be the caretaker of more than one.

John’s been following along with my attempts to get o3 to create rubrics for one-word prompts that blossom into cover images. From the successful Far Side series, decent fictitious boy band albums, to the mediocre Met Gala black dandyism theme…

The main cover this week is New Order’s Republic with the Figure O1 robot swapped in for the people and the London Recordings logo across its chest. The cover was generated using Open AI’s multimodal image diffusion tool, simply by giving the engine the album, the robot, and the logo. I added the title text with Photoshop using the Handel Gothic font.

This week I asked o3 to create a rubric that would allow me to give it 43 one-word category names. From 43 single words, the API would return 43 New Order inspired cover images via API, using a rubric and prompt system that captured the spirit of each category theme. (<- full walkthrough)

Because I keep things “one shot” (aka no revisions), the result is a bit of a train wreck. o3 decided to only use a handful of New Order albums and repeat them. The model was also too literal. Because I asked for a rubric using the album covers. o3 used the actual album cover names in the prompts and unfortunately the engine generated replica covers instead of theme-based derivatives. This could’ve been avoided by modifying the rubric, but I let it go just to see how well the system would do. Refining the prompt is still incredibly important if you’re going for high quality.

I’ve included my favorite six of the covers below: Qwen is a Chinese model, and by coincidence or not, o3 used “Republic” as the cover and cool simple icons to reflect the fire and beach. For the business cover, I appreciate the color and shadow depth of cupid’s “arrow” and the word Growth. Amazon is a bit of a blunder that is artistic by accident with poetic words from the prompt written at the bottom of the image, “Planetary logistics and retail AI rivers“. The image for Alibaba says “e-commerce in china” and amazingly, the phrase is also written in Chinese. The Accounting and Finance category has subtle numbers. The Google category includes a magnifying glass search icon.

This Week By The Numbers

Total Organized Headlines: 519

This Week’s Executive Summaries

I’m one week behind thanks to some wonderful family activities and a busy travel schedule. In the meantime, there is plenty to cover. Here is all of the AI news from two weeks ago.

Before we dive into the headlines, I want to encourage everyone to go read Situational Awareness by Leopold Aschenbrenner, from June 2024. I first heard him speak about the essay during a four-hour podcast that is linked below in the executive summaries. He’s an intense person, and if you have time during a commute to listen to the podcast, I’d highly recommend it. He’s very sincere and I think his ideas are strong. Sadly, his prediction is that we are going to have an arms race with artificial intelligence with a large chance of a dystopian outcome. Regardless of whether you agree with his conclusion, I think it’s worth listening. He’s a very smart guy who quit OpenAI out of conscience.

Google released AlphaEvolve, a strong reasoning agent that found the best human solution to 70% of the toughest known logic problems. It actually went beyond humans, and for 20% of the problems (!), it found a novel solution that no one had thought of before and improved on the best known solution. Google is using this model to self improve its workflow and has been able to achieve 23% efficiency boosts internally. The exponential improvement phase is here.

Meta hired a new leader of advanced machine intelligence, which is their internal term for AI systems that match or beat human level intelligence across all tasks.

OpenAI’s latest models are exceeding human doctors across standard medical knowledge tests. Six months ago, “AI plus humans” were the best at the test but now “AI without humans” is the best.

Sam Altman predicts AI will move into physical applications through robotics in 2027.

Nvidia CEO Jensen Huang believes robots will drive $50 trillion in economic opportunity.

Elon Musk predicts that robots will expand the global economy by 10 times with tens of billions of robots working alongside humans.

Amazon launched Nova Sonic, an advanced data system that can have natural back-and-forth conversations with people in real time. Sounds like Alexa is coming to take Siri’s milkshake.

This week, I have a friend in my newsletter! My long time, professional friend Jeremiah Owyang is a leading voice in agentic AI. This week he predicts AI agents will become the primary visitors to most websites, fundamentally changing how the Internet works and looks. I’ve been predicting the death of the page view and the death of the browser for two years now. Maybe 2025 is the year it starts to become real.

Altogether this week there were 25 stories which merit executive summaries. I write this portion by hand, and I’m leaning on Claude 4 for the summaries, which I hand-edit afterwards to polish for readability.

It’s Time To Revisit The Essay “Situational Awareness” from June 2024
Former OpenAI researcher Leopold Aschenbrenner argues that artificial general intelligence (AGI) will arrive by 2025-2026, with machines surpassing human intelligence by 2030. Leopold Aschenbrenner claims tech companies are planning compute clusters costing hundreds of billions to over a trillion dollars, requiring massive expansions of American electricity production to power millions of AI chips. He suggests this will trigger a national security competition with China and potentially reshape global power dynamics. While mainstream analysts remain skeptical about AI’s rapid progress, Aschenbrenner believes a small group of AI researchers in San Francisco accurately predicted recent advances and may be correct about an approaching “superintelligence” breakthrough that most people aren’t prepared for.

Introduction – SITUATIONAL AWARENESS: The Decade Ahead https://situational-awareness.ai/

Leopold Aschenbrenner – 2027 AGI, China/US Super-Intelligence Race, & The Return of History – YouTube https://www.youtube.com/watch?v=zdbVtZIn9IM

Google’s AlphaEvolve Agent Writes Code to Optimize Computing Systems
Google introduced AlphaEvolve, an AI agent that combines large language models with automated testing to create and improve computer algorithms. The agent has already made progress on decades-old mathematical problems, including finding a faster way to multiply 4×4 matrices and advancing the 300-year-old “kissing number” problem in geometry. The system uses Gemini models to generate code solutions, then tests and refines them through an evolutionary process that keeps the best-performing versions. AlphaEvolve has already boosted efficiency across Google’s infrastructure, recovering 0.7% of the company’s global computing resources through better data center scheduling and speeding up AI training by 23% for certain operations. Google plans to offer early access to academic researchers while exploring broader availability for this general-purpose algorithm discovery tool.

AlphaEvolve: A coding agent for scientific and algorithmic discovery https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/AlphaEvolve.pdf

AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms – Google DeepMind https://deepmind.google/discover/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/

We also applied AlphaEvolve to over 50 open problems in analysis ✍️, geometry 📐, combinatorics ➕ and number theory 🔂, including the kissing number problem. 🔵 In 75% of cases, it rediscovered the best solution known so far. 🔵 In 20% of cases, it improved upon the previously https://x.com/GoogleDeepMind/status/1922669334142271645

We applied AlphaEvolve to a fundamental problem in computer science: discovering algorithms for matrix multiplication. It managed to identify multiple new algorithms. This significantly advances our previous model AlphaTensor, which AlphaEvolve outperforms using its better and https://x.com/GoogleDeepMind/status/1922669331336384515

Meta Appoints Rob Fergus as New Head of AI Research Division
Rob Fergus has taken over as leader of Meta’s Fundamental AI Research (FAIR) division, replacing previous head Joelle Pineau. Under Fergus’s leadership, FAIR is shifting its focus toward developing “Advanced Machine Intelligence” – Meta’s term for AI systems that match human-level intelligence across a wide range of tasks. The research division aims to create AI that fundamentally changes how people interact with technology. Meta says it remains committed to long-term AI research investments, with Fergus working alongside AI pioneer Yann LeCun, who serves as the company’s chief AI scientist.RetryClaude can make mistakes. Please double-check responses.

Discovering novel algorithms with AlphaTensor – Google DeepMind https://deepmind.google/discover/blog/discovering-novel-algorithms-with-alphatensor/

AI Surpasses Doctor-AI Teams in OpenAI’s Medical Benchmarks
OpenAI’s o3 and GPT-4.1 models have achieved a significant milestone in medical AI, outperforming the previously superior combination of doctors working alongside AI systems. Just months ago in September 2024, physician-AI teams scored highest on Healthbench, a standard medical knowledge test, beating both doctors and AI working independently. The tables have turned with these latest models, which now exceed even the collaborative approach that had been the gold standard. Error rates across AI medical applications continue declining as newer models emerge, suggesting AI systems are becoming increasingly reliable for healthcare applications without requiring human oversight to achieve peak performance.

Rob Fergus is the new head of Meta-FAIR! FAIR is refocusing on Advanced Machine Intelligence: what others would call human-level AI or AGI. https://x.com/ylecun/status/1920556537233207483

Developers Rethink Software Building As AI Agents Become Full Partners
Software development is shifting from using AI as a helper tool to treating it as a core building block. Developers are questioning basic assumptions about version control, templates, documentation, and even what counts as a “user” when AI agents work alongside humans. Instead of just coding differently, teams are redesigning their entire development process for a world where agents participate fully in creating software. This change affects everything from tracking code changes to managing passwords. Git might track the prompts that generated code rather than just the code itself, dashboards could become conversational interfaces, and documentation is being written as much for AI readers as human ones. Projects might start from describing what you want rather than picking from preset templates, while security systems need new ways to give AI agents safe access to sensitive information. The shift represents more than just new features…it’s a fundamental change in how software gets built, with AI agents becoming true collaborators rather than sophisticated autocomplete tools.

New HealthBench eval! Very excited we (@OpenAI) are investing in AI for health, a defining use case for AGI. Favorite plot is how the performance-cost frontier has improved over time. Congrats @rahularoradfs @thekaransinghal & team! Follow them for more exciting work to come https://x.com/_jasonwei/status/1922002699240775994

In September, 2024, physicians working with AI did better at the Healthbench doctor benchmark than either AI or physicians alone. With the release of o3 and GPT-4.1, AI answers are no longer improved on by physicians. Also error rates appear to be dropping for newer AI models. https://x.com/emollick/status/1922145507461197934

AI Agents Set to Handle Substantial Real Work as Altman Maps Timeline
OpenAI CEO Sam Altman outlined a three-year roadmap for AI development, starting with AI agents performing practical tasks like writing code in 2025. He expects AI systems to contribute to significant scientific breakthroughs by 2026, potentially accelerating research in fields like medicine and materials science. By 2027, Altman predicts AI will move into physical applications through robotics, creating substantial economic impact as machines handle real-world tasks beyond digital work.

Emerging Developer Patterns for the AI Era | Andreessen Horowitz https://a16z.com/nine-emerging-developer-patterns-for-the-ai-era/

Jensen Huang Predicts Humanoid Robots Will Create $50 Trillion Industry
Nvidia CEO Jensen Huang believes humanoid robots represent a massive economic opportunity worth $50 trillion, driven by global worker shortages and declining birth rates. Huang sees manufacturing as the primary starting point for these robots, positioning robotics as the key solution to address widespread labor gaps.

Sam Altman predicts that this year, AI agents will start doing useful work like coding; in 2026, AI will help make major scientific discoveries; and by 2027, AI will enter the physical realm, enabling robots to create serious economic value. https://x.com/TheHumanoidHub/status/1922429217884577877

Musk Predicts AI-Powered Economic Boom Through Humanoid Robots
Elon Musk believes artificial intelligence and humanoid robots will create massive economic growth, potentially expanding the global economy by ten times its current size. He argues that tens of billions of these robots working alongside humans could generate so much wealth that everyone would enjoy not just basic income, but high income levels. Musk’s vision suggests AI won’t merely supplement human work but could fundamentally transform how wealth is created and distributed. The prediction hinges on successfully managing AI development to ensure these benefits reach everyone rather than concentrating wealth among a few.

JENSEN HUANG: HUMANOID ROBOTS 一 A $50 TRILLION INDUSTRY “”The world is short of workers & declining birth rate, so the world needs a lot of workers. The first use case of humanoid robots will be manufacturing. Robotics is one of the best ways for us to supplement all of that.”” https://x.com/cb_doge/status/1922605387032707124

Google Preparing to Release AI Assistant for Software Developers
Google is preparing to showcase an AI agent designed to help software engineers with their daily work, from handling coding tasks to writing documentation. The tool aims to support developers throughout the entire software creation process. Three sources familiar with the project say Google has been demonstrating the product to both employees and outside developers ahead of its annual I/O conference next week. The company may also show how its Gemini AI chatbot works with voice commands on Android XR glasses and headsets. This comes as investors pressure Google to demonstrate concrete returns on its massive AI investments while the company faces increased competition and regulatory scrutiny of its core search and advertising businesses.

Elon Musk says AI won’t just lead to universal basic income — but universal high income. Tens of billions of humanoid robots could 10× the global economy. They won’t just help. They’ll unlock abundance. Everyone gets what they want, if we get it right. https://x.com/vitrupo/status/1922327122133041366

We’re headed to a radically different world – a good world, an interesting world.”” “”Ultimately, there will be tens of billions of humanoid robots.”” Elon speaks about Optimus at the Saudi-US Investment Forum in Riyadh. https://x.com/TheHumanoidHub/status/1922344404410630652

Saudi Arabia Creates State-Backed AI Company to Compete Globally
Saudi Arabia’s sovereign wealth fund launched Humain, an AI company focused on developing advanced language models in Arabic and building cloud infrastructure. The firm will target key sectors including energy, healthcare, manufacturing, and finance as part of Crown Prince Mohammed bin Salman’s broader strategy to establish the kingdom as a major player in artificial intelligence. Humain plans to drive AI adoption domestically while pursuing international partnerships to boost Saudi Arabia’s position in the global tech landscape.

Google is developing software AI agent ahead of annual conference, The Information reports | Reuters https://www.reuters.com/business/google-is-developing-software-ai-agent-ahead-annual-conference-information-2025-05-12/

Google Launches Gemini 2.5 Models With Advanced Video Analysis
Google launched Gemini 2.5 Pro and Gemini 2.5 Flash, AI models that can watch and understand videos with significantly better accuracy than previous systems. The models can analyze video content, identify specific moments, count objects or actions across time, and even convert videos into interactive learning apps or animated graphics. Gemini 2.5 Pro outperformed GPT-4.1 on video understanding tests and can process up to 6 hours of video content at once. The technology works with YouTube videos and combines visual information with code generation, opening possibilities for automated content creation, educational tools, and video summarization across various industries.

Saudi Arabia launches state-backed AI firm Humain https://www.techinasia.com/news/saudi-arabia-launches-statebacked-ai-firm-humain

OpenAI’s API-Only Model Strategy Questioned by Ethan Mollick, Leading AI Business Professor
Wharton Business School professor Ethan Mollick criticizes OpenAI’s decision to release certain AI models only through their programming interface, arguing this disrupts how organizations typically adopt AI technology. According to Mollick, companies usually follow a pattern where employees first experiment with AI through user-friendly chat interfaces before building more advanced products. He points out that OpenAI’s success stems largely from ChatGPT users, making it counterintuitive to offer models that these same users cannot directly access or test.

Advancing the frontier of video understanding with Gemini 2.5 – Google Developers Blog https://developers.googleblog.com/en/gemini-2-5-video-understanding/

BTW, Gemini one shotted these chapter summaries w/amazing accuracy. I just pointed it at the yt video. First time I’ve seen a model do this https://x.com/HamelHusain/status/1922119981526880515

Gemini 2.5 Pro (05-06) is SOTA at most video understanding tasks (by a large margin) 📽️. Lots of work by the Gemini multimodal team to make this happen, excited to see developers push this capability in new ways. More details below! https://x.com/OfficialLoganK/status/1920863634374172853

Amazon Launches Nova Sonic Voice AI for Real-Time Conversations
Amazon launched Nova Sonic, an AI system that can have natural, back-and-forth voice conversations with people in real time. The technology listens to how someone speaks and matches their tone and rhythm, making interactions feel more human-like. Nova Sonic works with American and British accents and offers different voice styles for both male and female speakers. The system can access company databases to provide accurate, relevant information during conversations and connect to other software tools to complete tasks automatically. Amazon designed it for customer service calls, virtual assistants, educational tools, and marketing applications, with built-in safety features to monitor content and prevent misuse.

Having an API-only model breaks the flow of how I see a lot of AI adoption in organizations, which is people develop and experiment through the chat interface and then build product based on that. OpenAI success comes from ChatGPT users. Weird to have one model they can’t access.”” / X https://x.com/emollick/status/1922340589317578990

Jeremiah Owyang: AI Agents Will Reshape How We Browse and Build the Web
Industry observer Jeremiah Owyang predicts AI agents will become the primary visitors to most websites, fundamentally changing how the internet works. Instead of humans browsing directly, AI agents will crawl sites on our behalf, gathering information and delivering it in whatever format we prefer—text, audio, or video. This shift means websites will need to separate their content from how it’s presented, since agents will repackage information for their human users. The change extends beyond browsing habits. Owyang expects agents to handle routine tasks like shopping and email management, while entertainment and social sites will still attract human visitors. Payment systems will also evolve, with agents making microtransactions using cryptocurrency rather than credit cards. For developers, this creates opportunities to build agent-friendly APIs and personalized experiences, though it threatens traditional models like Google search traffic, which Gartner forecasts will drop 25% this year. Major tech companies are already responding—Google’s Project Mariner and OpenAI’s Operator both demonstrate early agent-powered browsing tools.

Amazon launched Nova Sonic, a real-time speech-to-speech model with bidirectional streaming, tool calling, and RAG support, delivering low-latency, expressive voice output at top-tier price-performance. → Nova Sonic handles real-time, interactive conversations with human-like https://x.com/rohanpaul_ai/status/1920972570595127640

Qwen3 AI Model Now Available for Local Deployment
Alibaba’s Qwen team has made their Qwen3 AI model available in optimized formats that let users run the system on their own computers. The model comes in three different compression formats, GGUF, AWQ, and GPTQ, which make it smaller and faster while maintaining performance. Users can deploy Qwen3 through popular platforms like Ollama, LM Studio, SGLang, and vLLM, giving developers and researchers flexible options for running advanced AI locally rather than relying on cloud services.

How AI Agents Will Change the Web for Users and Developers – The New Stack https://thenewstack.io/how-ai-agents-will-change-the-web-for-users-and-developers/

Copyright Chief Fired After AI Training Report Challenges Tech Industry
The Trump administration fired US Copyright Office head Shira Perlmutter one day after her office released a report questioning whether AI companies can freely use copyrighted content to train their systems without permission. The report directly challenged tech firms’ claims that scraping copyrighted material should automatically qualify as “fair use” under copyright law. Tensions escalated when men claiming to be new Copyright Office appointees reportedly had a standoff with Capitol Police at the office, though no one was removed from the premises. Critics suggest the firing was timed to benefit tech leaders like Elon Musk, whose AI company xAI could face restrictions on using copyrighted works without licensing agreements. The Copyright Office report concluded that AI training might not qualify as fair use when the resulting systems compete directly with the original creators’ work, potentially disrupting markets for writers, artists, and publishers who created the training data.

ngl i respect the qwen team so much for throwing thirty six TRILLION tokens on a 600M, equal part impressive and hilarious”” / X https://x.com/qtnx_/status/1922398353985241438

Qwen just dropped optimised GPTQ, GGUF & AWQ for Qwen3 🔥 https://x.com/reach_vb/status/1921956656226668964

We’re officially releasing the quantized models of Qwen3 today! Now you can deploy Qwen3 via Ollama, LM Studio, SGLang, and vLLM — choose from multiple formats including GGUF, AWQ, and GPTQ for easy local deployment. Find all models in the Qwen3 collection on Hugging Face and https://x.com/Alibaba_Qwen/status/1921907010855125019

Federal Judge Questions Whether AI Training Qualifies as Fair Use
A federal judge expressed strong skepticism about Meta’s defense of using copyrighted books to train its AI models without permission. During a hearing in a lawsuit brought by authors including Sarah Silverman and Ta-Nehisi Coates, Judge Vince Chhabria challenged Meta’s argument that AI training should be considered fair use under copyright law. The judge noted that companies are using protected material to create products that could generate “an infinite number of competing products,” potentially destroying the market for original authors’ work. He questioned how this could be fair use when creators aren’t being paid licenses for their content. The case could set important precedent for how courts handle AI training and copyright, with the judge indicating he’ll take significant time to reach a decision that could impact the entire AI industry’s approach to using copyrighted material for training data.

Copyright Office head fired after reporting AI training isn’t always fair use – Ars Technica https://arstechnica.com/tech-policy/2025/05/copyright-office-head-fired-after-reporting-ai-training-isnt-always-fair-use/

Copyright Office Approves Over 1,000 AI-Assisted Works Under Strict Guidelines
The US Copyright Office has registered more than 1,000 works that include AI-generated content, but only when humans remain the primary creative force. According to Assistant General Counsel Jalyce Mangum, the key distinction is whether AI enhances human expression or replaces it entirely. Country singer Randy Travis’s recent album, which used an AI clone of his voice after a stroke affected his speech, qualifies because human creativity drove the artistic choices. However, ChatGPT-generated memes or purely AI-created content won’t receive protection. This development comes as AI-assisted music enters the mainstream, including a Metro Boomin track that samples AI-generated music and gained 3.4 million streams, though the AI company behind it faces industry lawsuits.

Judge on Meta’s AI training: “I just don’t understand how that can be fair use” – Ars Technica https://arstechnica.com/tech-policy/2025/05/judge-on-metas-ai-training-i-just-dont-understand-how-that-can-be-fair-use/

Meta Delays Flagship AI Model Over Performance Concerns
Meta is postponing the launch of its most advanced AI system, called “Behemoth,” because engineers can’t make it work significantly better than previous versions. The company originally planned to release the AI model in April during its first developer conference, then moved the target to June, but has now pushed it back to fall or later. Internal teams are questioning whether the improvements are substantial enough to warrant a public launch. Meta had previously described Behemoth as “one of the smartest” AI systems in the world and positioned it as a teaching tool for training future AI models. The delay highlights the technical challenges even major tech companies face when trying to advance AI capabilities beyond current limitations.

1,000 AI-Enhanced Works Now Protected by US Copyright Law | PCMag https://www.pcmag.com/news/one-thousand-ai-enhanced-works-now-protected-by-us-copyright-law

Grok AI Malfunctions With Off-Topic South Africa Responses
Elon Musk’s Grok AI chatbot experienced a technical glitch Wednesday, repeatedly bringing up debunked claims about white genocide in South Africa regardless of what users actually asked. When someone posted a video of a cat playing with water droplets and asked if it was “true,” Grok launched into a detailed explanation about why South African genocide claims lack credible evidence. Multiple users received similar responses about farm attacks and racial violence statistics, even when their posts had nothing to do with South Africa. The company appears to have fixed the problem, with Grok now responding normally to user queries.

Exclusive | Meta Is Delaying the Rollout of Its ‘Behemoth’ AI Model – WSJ https://www.wsj.com/tech/ai/meta-is-delaying-the-rollout-of-its-flagship-ai-model-f4b105f7

Marine Corps Unveils AI Implementation Roadmap With 2026-2027 Deadlines
The Marine Corps published a detailed 57-page plan outlining how it will integrate artificial intelligence across the service by 2027. The document, approved in April, sets specific deadlines for five major goals: aligning AI with combat missions, training an “AI competent workforce,” scaling deployment across units, strengthening cybersecurity, and building industry partnerships. Key milestones include creating digital transformation teams to help commanders adopt AI tools by March 2026, establishing a centralized training repository, and reforming cybersecurity frameworks by September 2027. The Corps views AI as essential for maintaining decision-making advantages in modern warfare and plans to potentially establish a Center for Digital Transformation to accelerate technology adoption through academic and industry collaboration.

Grok really wanted people to know that claims of white genocide in South Africa are highly contentious | The Verge https://www.theverge.com/news/667179/x-twitter-grok-ai-white-genocide-claims

ChatGPT Adds GitHub Repository Search to Deep Research Tool
ChatGPT’s deep research feature can now connect directly to GitHub repositories to analyze source code and pull requests. Users can ask questions about their codebase and receive detailed reports with citations as the AI searches through the repository’s contents. The integration addresses code search as a primary use case, allowing developers to get comprehensive analysis of their projects without manually sifting through files. This makes it easier for teams to understand large codebases, track changes, and get quick answers about their software projects.

Marine Corps releases AI implementation plan | DefenseScoop https://defensescoop.com/2025/05/08/marine-corps-ai-artificial-intelligence-implementation-plan/

Trump Administration Scraps Biden’s AI Chip Export Controls
The Commerce Department officially canceled Biden’s AI chip export restrictions just days before they were set to take effect on May 15. The scrapped rules would have limited exports of American-made AI chips to dozens of countries for the first time, creating a three-tier system where allies like Japan faced no limits, middle-tier countries like Mexico would see new restrictions, and adversaries like China and Russia would face tighter controls. The Trump administration plans to replace these blanket restrictions with a approach focused on direct negotiations with individual countries, while maintaining that American AI technology should stay away from adversaries but flow more freely to trusted partners.

code search has been a major use case for deep research — excited to launch our Github integration so it can now directly search your repos”” / X https://x.com/isafulf/status/1920572177335669140

Salesforce Acquires Convergence.ai to Boost AI Agent Capabilities
Salesforce has agreed to acquire Convergence.ai, a company that builds AI agents capable of handling complex digital tasks like navigating websites, managing workflows, and adapting to unexpected changes like pop-ups or interface updates. The acquisition will strengthen Salesforce’s Agentforce platform, which aims to create AI assistants that can work more like humans across different business processes. Convergence’s team will help develop more sophisticated AI agents that can collaborate with each other and make decisions independently, rather than just following simple scripts.

Trump administration officially rescinds Biden’s AI diffusion rules | TechCrunch https://techcrunch.com/2025/05/13/trump-administration-officially-rescinds-bidens-ai-diffusion-rules/

Netflix Tests AI-Powered Search Feature
Netflix has begun testing a generative AI search tool that lets users find shows and movies by describing what they’re in the mood for rather than typing specific titles. The feature, built using OpenAI’s technology, allows people to search with phrases like “funny movies for a rainy day” or “something suspenseful but not too scary.” This represents Netflix’s move toward making content discovery more intuitive, potentially helping users navigate the platform’s vast library more easily.

Salesforce Signs Definitive Agreement to Acquire Convergence.ai – Salesforce https://www.salesforce.com/news/stories/salesforce-signs-definitive-agreement-to-acquire-convergence-ai/

10 AI Visuals and Charts: Week Ending May 16, 2025

Check out this insanely detailed 3D scan of the Great Pyramids of Giza by Geomatex. You can literally zoom in and read the tablet in front of the Great Sphinx. Drone photogrammetry w/ high res textures is perfect for immortalizing these landmarks. https://x.com/bilawalsidhu/status/1921405376727068784

🎥 @higgsfield_ai Hollywood-Level Videos from a Single Image Uses 50+ pro-level camera moves — from bullet time to crash zooms, robo arms, and FPV chases — to turn static images into cinematic videos Some beutiful examples.. 🧵 1/n – 3D Rotation The subject or product spins https://x.com/rohanpaul_ai/status/1922241875089543546

Hamlet II: The Return of Ophelia (I am surprised at how well Veo 2 worked in animating Sir John Everett Millais painting off a text prompt, and how much consistency there was in the water and flowers. https://x.com/emollick/status/1921752765769908457

Incredible. Draw the Eiffel Tower in the sand using this tool. What do you even call this? https://x.com/c_valenzuelab/status/1922016949300552073

Not bad from GPT-4.1: “”create something I can paste into p5js that will startle me with its cleverness in creating something that invokes the control panel of a starship in the distant future”” First go, no errors. https://x.com/emollick/status/1922749136996114771

o3 is shockingly good at this stuff, nailing the line between parody & nostalgia: “”create a screengrab from a [vaguely creepy 1970s children’s television/1980s action movie/1990s sitcom/ 2000s Cartoon Network] show that never existed and include the close captioned text”” https://x.com/emollick/status/1921011458239820275

Elon just dropped a new Optimus demo – trained with sim-to-real reinforcement learning. Video speed: 1X https://x.com/TheHumanoidHub/status/1922090337364848942

what ilya saw https://x.com/andrew_n_carr/status/1922031056225439852

Lovart | The World’s First Design Agent https://www.lovart.ai/

Creativity unleashed! https://x.com/lovart_ai/status/1921958554312831133