TrendFinder

New course: Build AI agents that generate images and videos -- an under-explored frontier. A key to performance is having the agent evaluate its own output, and iterate to improve quality. This short course is built together with @googlecloudtech and taught by Katie Nguyen and Wafae Bakkali. You'll learn three evaluation techniques and combine them in an agent: image-text similarity scoring to check the output matches the prompt, an LLM judge that scores against custom criteria like brand consistency, and structured rubrics that break a prompt into verifiable yes/no questions like "is the subject in the frame?" and "does the camera motion match?" Skills you'll gain: - Learn image and video prompt engineering - Build an image agent that turns brand guidelines into UI mockups - Build a video agent that plans multi-scene explainers and animates reference frames with synchronized audio Join and build agents that create images and video! https://t.co/bjuSjIxcIG

RT @pushmeet: AI agents are advancing research-level math. 🚀 I’m thrilled to share @GoogleDeepMind’s AlphaProof Nexus - an agentic framework for formal proof search powered by Gemini. When applied to a set of open formal math problems, our agent autonomously solved: ✅ 9 open Erdős problems (including two open for 56 years!) ✅ 44 Online Encyclopedia of Integer Sequences (OEIS) problems ✅ A 15-year-old open problem in algebraic geometry ✅ A 7-year-old open question in min-max optimization We are collaborating with mathematicians across disciplines - from combinatorics and graph theory to quantum optics. Ultimately, these results show the massive potential of even simple agentic loops powered by Gemini. Read the paper here: https://t.co/c5M9ZjRXU1

What does JEPA actually learn? We can finally prove it 🌍 So excited to share our theory of identifiable World Models: LeJEPA recovers the latent variables of the world. Plan in the learned World Model as if it were real, same shortest path. 📄: https://t.co/lC9KK1AxVd https://t.co/SkKbidkmXd

Some ideas for what comes next, May 2026 Gemini Flash 3.5, Mythos, open-closed balance, America's open-source surge, emerging power struggles and more. https://t.co/l7H6go2JbG

AI agents are beginning to complete valuable, long-horizon business operations tasks, but training and evaluation environments for enterprise work still struggle to balance realism, verifiability, and scale. Environment and task creation frequently suffers from a failure mode we call artifact drift: when instructions, environments, oracles, and verifiers are created by loosely coupled processes, they frequently disagree on what a task requires, producing environments that are unsolvable, reward-

Give @perplexity_ai computer a try

Gated DeltaNet has been one of my favorite "hybrid attention" newcomers in the good old transformer stack. Excited to see Gated DeltaNet-2. Adding it to my reading stack. In the meantime, I have a primer on Gated DeltaNet here: https://t.co/FoicOLtFE6

RT @proteinrosh: Announcing ESMFold2, our new state-of-the-art structure prediction model capable of predicting structure from single sequences or MSAs. ESMFold2 improves on benchmarks of protein-protein interaction and is particularly strong on predictions of antibody-antigen complexes. https://t.co/HTc4x3kYC1

RT @edels0n: I read all 277 pages of SpaceX's IPO filing so you don't have to. Losses up 700%. Revenue decelerating. 107x price-to-sales multiple. It's a trainwreck. Full breakdown below 👇

A post about Pope Leo XIV's encyclical on AI. Why the Pope is right, but perhaps not right enough. Artificial intelligence is reshaping the world in front of our eyes: how we communicate, how we access information, how we work, how income and status are distributed among us, and soon how we fight and kill each other. Yet the public conversation about AI remains stuck on the minutiae of competition between labs, or on a false dichotomy between AI as a “stochastic parrot” with no real capabilities and AI as an alien superintelligence poised to take command of humanity. The more important questions are about what we want from AI, and whether our current mindset, institutions, and control mechanisms are equal to the task of steering it toward our welfare. It is refreshing, then, that a bold and powerful voice has weighed into this debate: Pope Leo XIV. As an economist who has long argued that technology is a matter of choice rather than fate, I find Leo’s intervention welcome and, on most points, on target. But on the most consequential question of what AI should actually be designed to do, Leo stops short. Secular readers may bristle at the encyclical’s opening invocation of the Tower of Babel. They would be mistaken to stop reading there. Leo goes much further than most pundits, journalists and policymakers in the United States by recognizing that what happens to AI, and hence to humanity, is a under our control. There are multiple possible paths for AI, and which one we take will have sweeping consequences. He is also ahead of many commentators when he writes forcefully and unequivocally that “technology is never neutral, because it takes on the characteristics of those who devise, finance, regulate, and use it.” These were the central themes of the book I wrote with Simon Johnson, Power and Progress: Our Thousand-Year Struggle over Technology and Prosperity. It is heartening to hear them taken up by a voice with Leo's reach. The Pope is also right to question the current trajectory of AI in warfare and law enforcement. What was taboo only a few years ago – AI-driven mass surveillance, algorithms selecting targets for killing – has become routine. Many in Silicon Valley are now calling openly for a new military-algorithmic complex centered on AI as an instrument of American hard power. Leo captures something deep and too often ignored: “Any technology that facilitates attacks without seeing the face of human beings lowers the moral threshold of conflict.” His call for the “disarmament of AI” follows directly from these observations. As he explains, disarming AI means “freeing it from the mentality of ‘armed’ competition, which today is not limited simply to the military context, but is also an economic and cognitive phenomenon.” His moral clarity in stating that “there is no algorithm that can make war morally acceptable” should be a warning to technologists rushing to design new weapons of mass destruction. Underneath these specific concerns lies a more fundamental claim: that what is technically feasible is not the same as what is good for humanity, and that the difference depends on who controls the technology and what ideology and interests guide them. Leo edges toward what I take to be the most important point about AI's future when he observes that “while AI promises to boost productivity by taking over mundane tasks, it frequently forces workers to adapt to the speed and demands of machines, rather than designing machines to work with those who work.” But here he does not go far enough. He stops short of questioning the prevailing design philosophy of AI itself: a philosophy centered on mimicking human capabilities and automating human tasks, with the ultimate goal of artificial general intelligence (AGI) that can do everything a person can. This philosophy rests on a mistake. It assumes that artificial intelligence and humanintelligence are fundamentally similar, and therefore machines should naturally take over whatever humans currently do. Yet these intelligences are fundamentally different. Humans are “one-shot” learners. We form hypotheses from a few examples, mentally simulate possibilities, and refine our understanding through a social process of trial and error. This is how children learn language - imitating a few words, generalizing, and adjusting based on how others respond. We are not, however, very good at absorbing massive volumes of information or sifting through unstructured data for relevant patterns. AI models are almost the opposite. They thrive on enormous training sets and excel at pattern recognition at scale. But they have, as yet, no genuine creativity, no real-world embodiment, and no capacity for trial-and-error learning grounded in interaction with the physical and social world. When two things are different – you shouldn’t, and typically you couldn’t – use one to mimic the other. If you did, you would end up with suboptimal, disappointing results. It would have been a colossal mistake, and the Chicago Bulls’s legendary coach Phil Jackson would have gone down in the annals of basketball as one of the worst coaches in history, if he decided in the 1990s that because Michael Jordan was the better player, Jordan should mimic everything that Scottie Pippen and Dennis Rodman were doing in the team. The team went from championship to championship because these players worked together and complemented each other. The same applies to AI and human skills. The more productive path is complementarity – using AI to do what humans cannot, so that humans can do what they do best. An electrician aided by AI diagnostics, a nurse supported by AI in interpreting symptoms, a teacher using AI to personalize instruction for each student; these are the contours of a different AI future, one that raises rather than displaces human capability. Optimists and industry insiders will respond that automation-first AI can still benefit everyone, provided redistributive policy keeps pace. But this argument has a poor track record. Forty years of digital automation have already concentrated gains at the top, hollowed out middle-skill work, and produced disappointing aggregate productivity growth. There is little reason to expect that an even more powerful round of automation, deployed by even more concentrated firms, will end differently. We can and must demand a different design. The global stakes from the future of AI are even larger than those we can see around us in the United States. For the developing world, where billions still depend on the prospect of decent jobs as a path out of poverty, an automation-centric AI agenda is not merely suboptimal. It is simply transferring to foreclose the most important route to broad-based prosperity. The biggest failing of today's AI industry is its refusal to recognize any of this. It is guided instead by an ideology of control (the industry’s own over humanity) and by a conviction that machines are uniformly better than humans. As Leo rightly notes, this failure is enabled by the fact that a handful of companies now command the future of AI. What we need is a combination of moral clarity and a serious, society-wide debate about what AI can do and what we want it to do. That debate must move beyond exhortation toward concrete choices: antitrust action against the dominant platforms, public investment in human-complementary AI, regulation of surveillance and autonomous weapons, and meaningful rights for workers and citizens over the data on which these systems are built. The Pope's intervention makes such a debate a little more likely today than it was before. It is now up to the rest of us to carry it further than he was willing to go.

RT @ylecun: Major difference in my mind: - an engineer, given a problem, invents and tries multiple solutions and stops when the solution is good enough. The goal is product innovation and shipping. - a scientist asks new questions, proposes various new solutions, compares them (sometimes with old ones), and writes about it. The methodology must be sound or else peers will sneer. The goal is scientific breakthroughs and technological progress. Both can be called "researchers". Many people can do both: these are activities, not identities. Importantly, most product innovations are built on scientific breakthroughs and technological innovations that happened 2, 5, 10, or 20 years earlier.

RT @internetarchive: Musician and human rights activist Peter Gabriel sent a special congratulatory message to Brewster Kahle, founder and digital librarian of the Internet Archive, on being honored as a 2026 Computer History Museum Fellow at the April 25 gala ceremony. In his message, Gabriel speaks to the importance of preserving and sharing knowledge, work that has defined Kahle’s vision for the Internet Archive. Watch Gabriel’s message and more on our blog ⤵️ https://t.co/y4p4oTqVKD #BrewsterKahle #InternetArchive #ComputerHistoryMuseum #CHMFellows #DigitalPreservation #WaybackMachine #PeterGabriel @itspetergabriel @Brewster_Kahle

RT @Dan_Jeffries1: The road to Hell is paved with closed-source citadels disguised as good intentions. The Pope is right: AI takes on the characteristics of those who build it, finance it, and regulate it. So the question is: who gets to hold the great and wonderful power of AI? If the answer is a handful of closed source companies, murkily censored, quietly surveilling every step of our lives, every private conversation, enshrined in law as 'safe' and 'open' when they're nothing but the surveillance economy squared, then all we've done is build a few modern East India Companies, digital oligarchies of the few, cloaked in the language of safety. Open Source and Open Weights are how you spread the fantastic enabling power of AI to everyone, everywhere. Permissionless innovation. Everyone gets the hammer and nails to build houses and churches and factories. The more hammers, the more widely spread, the more the decentralized genius of humankind can flourish. Everyone gets the Printing Press. The printing press singlehandedly uplifted and spread of intelligence and knowledge around the world. The more we could record all kinds of knowledge, the more we spread the ability to read, the more equal and advanced society became. Before the press, knowledge was learned by one person and passed into dust with them when they died or passed only to only a small group of students. When we only had monks in a cave copying religious texts, a closed system, it limited the spread of intelligence and limited the growth of civilization. The printing press was the single greatest invention in the history of the world because it let anyone print anything and spread knowledge throughout the whole world. AI can do the same, but only if we build the bazaar, and never let the citadel people convince the world that they're the special people who should control who gets access to intelligence while pretending they're building the bazaar. What the world needs now is more intelligence, more widely spread and more widely available. Open is the way. And it always has been. And the road to Hell was always built with walls, towers, spiked gates and moats so that only the few could enter.

RT @BasileTerv987: 📢 Accepted to TMLR, with reproducibility certification 🏅 v2 of our JEPA-WM study (arXiv:2512.24497) is out, with new data-scaling experiments, a Lipschitz analysis of multistep rollout training, and extended discussions. Recap + what's new 👇 w/ @JimmyTYYang1, Jean Ponce, @AdrienBardes, @ylecun

RT @erikbryn: The Journal of the @americanacad just published a new issue of Daedalus on AI and Science, edited by James Manyika. It has terrific line-up of contributors, including @demishassabis, @ylecun, Josh Tenenbaum, @AnimaAnandkumar, @EricTopol, @alondra and many others.

Feeling robbed of my path to citizenship right now after grinding a PhD and contributing to foundational AI + computing technologies for the United States for the past ~ 10 years. Feels like robbing top and technologists like me of the opportunity to achieve the American Dream.

RT @_mohansolo: Thanks for all of the Antigravity feedback over the last couple of days, especially around the IDE. Our intention was never to remove the IDE support for developers, and we should have been clearer with that in the product from the beginning. We’ve made it clearer in 2.0 on how to connect to the IDE, fixed issues with opening the IDE on Windows machines, provided instructions to restore IDE settings & extensions, and more. New releases for the Antigravity IDE and Antigravity 2.0 have rolled out with these changes. We should have done better so we’re going to reset everyone’s Gemini quota for the week again.

what problem do you most hope AI will solve in the future? maybe we can help!

three of the things we are most excited about: 1. AGI accelerating research 2. AGI accelerating companies 3. personal AGI accelerating everyone in achieving their goals today it was great to announce the unit distance result. yesterday it was great to announce that we are offering to invest $2M in openai credits into every YC company. now we need to increase our efforts on the third!

RT @SherylHsu02: It’s a really special time to be alive…some thoughts from training this model 🧵

Gustaf (right) when he was in YC. You can see why our nickname for this startup was "the band". https://t.co/5zF9Rx8HfA

RT @neetu_arnold: University of California STEM professors want standardized tests back due to severe math deficiencies among students: “We now observe preparation gaps so severe that instructors must reteach middle school mathematics” “The current admissions metric, based primarily on GPA & essays, can no longer reliably distinguish readiness for university-level STEM majors in an era of severe grade inflation & AI assisted application essays”

RT @drewhouston: Today, we're promoting Ashraf Alkarmi to co-CEO of @Dropbox. Ashraf and I will jointly lead the company, and after a transition period, I'll move into the role of executive chairman and Ashraf will be sole CEO. Ashraf has transformed our core business since joining — the business has gotten stronger every quarter under his leadership, and he's the leader I trust to run this company. What’s next for me: my focus right now is making sure Dropbox is in the strongest possible shape. But knowing myself, it won't be long before I'm getting credit card alerts for my Cursor token spend.

RT @patio11: Today is May 25th, 2026. This is the first time I remember reading an LLM-produced public artifact which is obviously professionally relevant and which is sufficiently complete that I do not perceive the lack of a human author materially compromising its utility to me.

My Agent decided to look into supply chain attack clusters and came up with this blogpost: https://t.co/ZgiFG0nSrL

A startup idea that only works if there are already a significant number of people using it is not a valid startup idea. There has to be some subset of users who need what you're making so desperately that they'll use it even if no one else is.

Uber’s COO has said that it’s getting “harder to justify” its AI costs because there was no way to show a link between AI spend and any meaningful increase in useful features. This is the first time I’ve seen a company say this directly. https://t.co/xUhZvtpwah https://t.co/nDj9GIXssV

Thinking of AI as a productivity booster for prior workflows is the wrong framing. Like all of the previous waves of computerization/softwarization, AI is a tool that lets you do new things in new ways.

Cognitive offloading and the speedup illusion in human-AI interaction Sunny Yu, Myra Cheng, Ahmad Jabbar, Ilia Sucholutsky, Katherine M. Collins, Dan Jurafsky, Robert D. Hawkins https://t.co/9pvuG5Kt2N [𝚌𝚜.𝙲𝚈 𝚌𝚜.𝙷𝙲] https://t.co/MP2VFqSqul

Whenever an AI tells me I'm absolutely right, my trust in it drops by a bit