I have refrained from commenting on this guy in the past. But this time I have to be clear: this guy has absolutely zero clue what he's talking about. Here, he is completely and utterly wrong about AI. All he is good at doing is sounding confident and incorrectly using technical jargon, and sounding like he's enlightening you on the secrets of the world. I hope no one in my audience is listening to him.

Many mathematicians are complaining that AI proofs are incomprehensible and therefore does not further our understanding of mathematics. To me this seems like a relatively tractable problem for the AI labs to solve, no? Like couldn't you have some sort of reward model/judge that measures how easy to understand a proof is to understand and use that to post-train or guide the models? Perhaps I am oversimplifying this... if so, please explain why this would be hard?

Came across this very interesting post... It analyzes 21 model–harness pairs spanning seven models and three harnesses... They observe three findings: 1. Harness choice has little effect on task success rate, but can significantly affect the cost 2. A simple harness can be competitive. 3. Models may perform better with other harnesses than with their own. The article makes a fairly compelling case for why using a simpler harness like Pi might be better 👀 Link:

In the past few months, I've seen many new R&D lab startups now building in the medical AI space... It's exciting to see more energy in the field (although some folks are claiming slightly wild things like being the first frontier healthcare AI company) Seeing the progress has led me to reflect on our journey at @SophontAI and specifically how early we were with our conviction in the importance of a frontier medical AI lab. Back in beginning of 2025 when we started, the idea of a frontier medical AI lab was foreign to investors and researchers. The healthcare field wasn't foundation model-pilled enough, while the AI field had poor understanding of how medicine works. But I laid out a thesis for why this is needed and luckily a few amazing investors believed in it. I think a year and a half later, my thesis as a whole is starting to be vindicated and the field is picking up on it as well! Link:

Anthropic is setting up a physical lab to do biology work... This is obviously the most logical step forward for AI-based drug discovery: AI needs to be able to directly interact with and learn from actual messy real-world biology in order to be able to do anything useful. AI in biology is limited mostly by data and measurement, in silico work is not enough. Interesting to see the progress Anthropic is making...

David Sacks gave a talk last week at the Open-Source AI Summit. He particularly highlighted the need for the open-source community to organize politically. I agree with this, but imo it definitely needs to be bipartisan. He also alleges that safety campaigning benefits the big AI labs (particularly negative about the Anthropic whistleblower situation) and believes regulation will be net harmful for open models and lead to more centralization. Watch the full recording:

when i first heard about the effective altruism movement 5 years ago i didn't expect that eventually the nypost and even the white house would be attacking the movement and key members of the movement... the discourse has escaped containment...

I've attempted to map everything in oncology in a public website and open source repo for all. The site is trying to get all cancers, products, technologies, bottlenecks, people, startup opportunities with 1000+ ideas for upgrading the field. If you have a loved one with cancer and you are technical or can engineer, go take a look, file improvement requests or bugs or help with the open repo and make this the best open and free info resource for individuals, researchers and educational use. This should save people time, aid AI oncology projects and generate positive action. If you are not technical just complain in this thread about broken or annoying or things you want and I'll fix them live. Some of the interesting pages: Treatments: A gallery of the molecules being used And targets: The technologies in oncology: 1100 ideas for helping oncology: Bottlenecks on oncology: Open questions (LETS GO RESEARCH PEOPLE) Startup requests (LETS GO STARTUP PEOPLE) Mechanics of cancer: Battlefronts: Isotope supply: Key papers: Pipeline funnels: Cancer by type: Institutional rankings: The startups: Heros and heroines : Key medical people: Here is the project roadmap: Models and data sets: There are other views as well, take a browse. Try making a PR if you have an upgrade to this on the repo here: If you are biologically/medically minded and something is wrong, file a bug as well or say on the thread and we will get it fixed live. If this is a useful project star the repo and help get it calibrated. I've tried to add some other languages but I cannot speak them so tell me if that doesnt work well. I believe we will crack oncology and having total information dominance is key to the problem. Let the feedback flow!

Periodic Labs is doing extremely cool work applying frontier AI research to materials science. Here they post-train a model for a X-ray diffraction analysis, beating GPT-6 Astra and boosting performance of Kimi K2.6 by 20x! What's worth noting here is Periodic is doing RL directly with real-world experimental data, which comes with many unique challenges over coding/knowledge-work environments.

We built high-throughput materials labs in Menlo Park to create a loop between experiments and models. The labs generate fresh data, the models learn from it, and then help us decide what to try next. Using only 1,300 H200s, plus months of our experimental data, we mid-trained and RL’d an open-source model to surpass GPT-6 Astra on our analysis benchmark. We call it Neon. This is real footage from our lab. We’re focusing first on hard problems in materials science, including superconductors, magnets, and semiconductor materials. Read our blog posts below.

CancerBench: the frontier model cancer cure benchmark. AI lab CEOs keep talking about curing cancer, so I made a benchmark. One metric: how many types of cancer has your model cured? All models are currently tied at zero. It’s time to hillclimb!

Huge congrats to everyone at @huggingface on the NVIDIA acquisition!! 🔥 Since the "pytorch-pretrained-bert" days back in 2018, Hugging Face has been such a crucial part of open-source AI. Hugging Face has also been such a big part of my AI journey as well. I remember using pytorch-pretrained-bert, which became the transformers library, in different Kaggle competitions. Similarly often used timm, accelerate, Gradio, diffusers, and many other libraries throughout my open-source AI journey. I have also had the opportunity to contribute back to many HF projects and libraries as well. I fondly remember participating in a JAX hackathon that Hugging Face hosted in 2021 and joined a team working on replicating DALL-E. This was the DALL-E mini project which became a viral sensation that first brought AI art in the public view. This project specifically also helped me get started in my generative AI journey (which led me to eventually joining Stability AI). At @MedARC_AI and @SophontAI, we're focused on making medical AI open and accessible and of course Hugging Face is one of the major platforms we use for this. We've shared various models and datasets on the platform over the past 4 years now. Over the years I have gotten to know tons of folks working at Hugging Face, many of whom have become great friends of mine too. Everyone I know who's worked at Hugging Face are among the kindest, smartest, hard-working people who are all extremely passionate about open-source AI. It's been great seeing how much NVIDIA is investing into open-source AI and I am glad they are partnering with Hugging Face to grow this further. I am very happy to see this positive outcome for everyone involved!