Paul retweeted
Nice paper on optimizing agents and their graders together.
The key finding is that you want to update your LLM grader alongside the agent instead of keeping it fixed, because the agent's failure modes change as it improves.
DUET alternates between the two.
Each round picks training tasks, runs the solver, has the grader score and critique the trajectories, then a tool-using updater revises either the grader or the solver while the other stays fixed.
On the Finance Agent Benchmark, the solver reaches 96.5%, above GEPA (94.4%) and SkillOpt (94.0%). It also beats both when they optimize against the benchmark's own reference evaluator.
Grader agreement rises from 68.4% to 78.5% on GDPval. Training the grader with the solver frozen moves agreement only from 67.9% to 68.8%.
Paper: academy.dair.ai/papers/duet-…
Paul retweeted
Color consistency is a persistent friction point in production. Our interactive research team just returned from The 19th European Conference on Computer Vision (@eccvconf), where they presented new work focused on solving just this: ensuring colors match reliably across shots and AI-generated reference photos.
Paul retweeted
If you explained agriculture to folks in SF, you'd blow so many minds. It's... self-replicating, recursively self-improving nanotech for the physical world. And it's all open-source!
One instance can turn into hundreds of seeds, that turn into hundreds of instances. Next thing you know you have trillions of them. It's super-exponential
Paul retweeted
Congratulations to our CEO @JensenHuang on receiving the National Medal of Science for advancing GPU computing to power scientific breakthroughs.
And to fellow honorees Elon Musk, Lisa Su, Michael Dell, Satya Nadella, and Sergey Brin.
.@POTUS presents the National Medal of Science to Sergey Brin, @JensenHuang, @elonmusk, and @LisaSu, and the National Medal of Technology and Innovation to @MichaelDell and @satyanadella.
Paul retweeted
not as clear-cut, but also yes
Tired: Two-pizza teams
Wired: Two-slice teams
@danshipper on the best team structure in the AI-era
Paul retweeted
John Urschel is an incredible guy. Had the pleasure of co-presenting an award with him at the Breakthrough Prize a few years ago - wonderful person and an inspiration!
GENIUS: Former #Ravens offensive lineman John Urschel retired from the NFL at just 26 years old to pursue mathematics full-time.
Now, years later, Urschel has reportedly helped RESOLVE A LONG-STANDING MATHEMATICAL CONJECTURE involving Gaussian elimination.
Urschel earned a 4.0 GPA, got bachelor’s and master’s degrees in mathematics from Penn State, and pursued his Ph.D. at MIT while playing in the NFL.
From blocking NFL defensive linemen to solving problems that have challenged mathematicians for years.
One of the smartest athletes EVER.
Paul retweeted
i wrote approximately 2000 tweets and articles (like Deep Learning is Hitting A Wall) about the importance of supplementing deep learning/ LLMs with neurosymbolic AI over a span of 25 years and the tech bros are circulating the one time i forget to explicitly mention this, completely ignoring the rest of my work.
good grief.
Paul retweeted
Runway 🤝 Claude
Bring your ideas to life with @claudeai Motion + Runway.
Use Claude Motion to animate charts, create customer walkthroughs or create short explainers, then bring your work into Runway to generate videos and images.
Learn more at the link below.
Paul retweeted
got 722 problems but the hodge conjecture ain’t one
We’re releasing a broad range of new mathematical results produced by an internal frontier model.
We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results.
github.com/openai/math
Paul retweeted
Tired of winning
Trump's Beef Catastrophe: Incompetence, Blithering Idiocy, and Blatant Lies Have Obliterated America's Beef Industry
In yet another humiliating display of incompetence, the authoritarian Trump regime has turned America's once-mighty beef industry into a national punchline. After months of touting "record" export numbers as proof that his chaotic trade policies were somehow working, the U.S. Department of Agriculture has been forced to admit it inflated beef export sales by a staggering 90 percent.
According to a bombshell Reuters report, the USDA bragged in early July that U.S. exporters had sold more than 126,000 metric tons of beef to foreign buyers for a sales period ending in late June. The actual number? Just 12,000 metric tons. That's not a rounding error. That's a catastrophic fabrication that would embarrass a high-school bookkeeper.
The fake numbers were so absurd they should have raised red flags immediately. The USDA claimed record sales of more than 38,000 metric tons to Chile and over 32,000 metric tons to Italy — countries that aren’t even major markets for American beef. Reality? A pathetic 367 tons to Chile and 350 tons to Italy. Fourteen other countries saw their numbers slashed as well.
This isn't just bureaucratic sloppiness. This is the Trump regime's signature brand of hype, lies, spin, and propaganda colliding with economic reality — and the American cattle industry is paying the price.
U.S. beef prices have hit record highs this year thanks to tight cattle supplies and strong domestic demand. Exports have been declining since 2022. Higher prices have priced American beef out of the global market, as commodity analyst Austin Schroeder bluntly told Reuters: "We're priced out of the world market to a certain extent. It wouldn't make a lot of sense for that big of an export number."
Meanwhile, domestic supplies are so low the U.S. has had to ramp up beef imports. Earlier this year, a screwworm outbreak forced Texas to declare a state of disaster, further hammering producers already squeezed by Trump-era chaos.
Instead of acknowledging the mess his administration created — from trade wars that disrupted markets to regulatory neglect that left ranchers vulnerable — Trump and his sycophants simply let the USDA pump out fantasy numbers and call it a win.
This is the same pattern we've seen across the board: brag about successes that don't exist, then quietly revise the data when caught. Only this time it's not some abstract economic indicator. It's the backbone of rural America — cattle ranchers, feedlot operators, processors, and exporters — watching their industry get trashed by incompetence and covered up with lies.
The revised numbers aren't just embarrassing, they're devastating. They confirm what ranchers have been screaming for months: America's beef sector is in serious trouble, and the people running the show are too busy tweeting victory to notice or care.
The only question left is how many more "record" numbers will have to be walked back due to Trump being a serial lying sociopath who is too psychologically fragile to handle the truth.
Paul retweeted
You guys are so worried about web-based GEO that you forget to take advantage of EEO: email engine optimization.
I'm draftin' side notes to AI left and right 😉
Paul retweeted
In the past few months, we had a lot of fun (and stress 😅) to produce 12 versions (+ many subversions) and 137 pages in our training run log book.
Turns out human-human collaboration is important to improving human-AI collaboration. 😊
People talk, listen, watch, think, and collaborate at the same time, in real time. We've designed an AI that works with people the same way.
We share our approach, early results, and a quick look at our model in action.
thinkingmachines.ai/blog/int…
Paul retweeted
The Adolescence of Technology: an essay on the risks posed by powerful AI to national security, economies and democracy—and how we can defend against them: darioamodei.com/essay/the-ad…
Paul retweeted
Grok Voice Transcribe 2.0 is twice as accurate as its predecessor in customer-support calls, spoken credentials, and short voice commands.
Paul retweeted
Workera is being acquired by Pearson! It's been a privilege to support CEO @kiankatan and the whole Workera team through this journey. Back in 2019, Kian had the insight that rigorous skills measurements would be important. With recent advances in AI, this is now more true than ever. Workera's technologies for rigorously measuring people's skills in different tasks and job roles are helping many businesses understand where their employees are strong and where there are areas for development.
With Workera and Pearson combining forces, I'm delighted that under @omarabbosh and Kian's leadership. Workera now has the potential to support an even larger number of people and businesses. Thank you, Kian and the Workera team for having me serve as the company's chairman, for launching Workera out of DeepLearning.AI and AI Fund, for your friendship over the years, and for your important work advancing assessments.
I have some big news to share. Workera is being acquired by Pearson!
Over six years ago, I was teaching at Stanford and thinking about a simple question: what if we could understand everyone's skills as precisely as the best teachers understand their students?
I believed it could lead to a more meritocratic world. People could be recognized for what they can actually do, not just their credentials or network. They could understand their strengths and gaps, and rapidly develop the skills they need next. Organizations could discover talent they might otherwise overlook and manage their workforce with trusted skills data.
What felt like a dream at the time is now a reality. Workera brought together experts in AI, psychometrics, and enterprise execution to build AI systems that reinvent how skills are measured. Our team pioneered AI-native skills intelligence, agent-led multimodal assessments, and even ambient skill measurement. We've established skills benchmarks across organizations, industries, and roles.
And this mission feels more important today than ever! AI is changing work as we speak. Some roles are disappearing, new ones are emerging, and we need to help billions of people develop new skills and navigate what comes next.
When I first spoke with @omarabbosh, it became clear that our companies shared the same mission. Pearson has helped generations of people learn and prove what they know. If you're reading this, there's a good chance you've taken a Pearson assessment, learned from their educational materials, earned a professional credential through them, read their psychometrics research, or benefited from their enterprise products in many other ways.
Bringing together Workera's technology and AI talent with Pearson’s global scale and deep expertise in learning and assessment means we can pursue our mission at a scale we could only imagine on our own.
To our customers and partners, thank you for believing in us. Expect even more innovations coming out of Workera and Pearson.
To the Workera team, I’m incredibly proud of what you've built, and your continued dedication to our beautiful mission. To our board and our chairman @AndrewYNg, thank you for your belief, support, and mentorship.
To everyone, we have big plans for this next chapter, so please stay tuned.
We're just getting started! 😊
Paul retweeted
OpenAI just dropped solutions to a whole bunch of hard Math problems
The AI solved more problems than what humans could solve in a decade
Expect AI to ace all science problems very soon!
Paul retweeted
If the target market is real time robotics, the Jetson Thor system, with 128 GB of memory, but only 273 GB/s of bandwidth, seems over-provisioned with expensive memory. You want the model evaluating at tens of fps, so it can't use more than 10 GB of weights at most.
More memory could be used if the model is a wide mixture of experts, or a large strategic planning model is operating in small time slices over a longer period of time, and more memory always makes development life easier, but for cost optimized systems, you should be able to get by with much less.