Enjoyed speaking at the PyTorch conference VC panel keynote. Great exchange on what’s actually moving the needle in AI research and where technical breakthroughs meet venture-scale opportunities. Key takeaways: 1/ RL’s surprising generalization: RL is generalizing better than

Link is live here-

3/ today's models are particularly strong at tool use and can be effectively trained with RL. This technical readiness finally enables production systems for real business problems beyond research benchmarks.

2/ Untapped efficiency gains in pre-training: While the focus is on scaling compute, significant improvements remain. Delta uses linear attention to scale context windows to millions of tokens more efficiently than current quadratic attention mechanisms. Diffusion models for code

Hear our founder @CarinaLHong on the latest @TWIML podcast with @SamCharrington 🎙️on why Axiom is bullish on math 📐🚀 “Math and coding are the two biggest parts of the digital world. Coding is heavily invested in. Math is not. Not because math isn’t trackable, but because math

The “factory as a robot” is becoming real. @SkildAI + @NVIDIA + Foxconn are fusing world-model intelligence with industrial automation. A glimpse of how foundation models will shape physical infrastructure, not just digital ones

💥Jensen featuring @SkildAI during his @nvidia GTC keynote Automating the buildout of Nvidia’s GPU factories The next era of American manufacturing!

Fantastic talk by @achowdhery at #PyTorchCon on how we actually measure agentic intelligence. Loved the focus on long horizon reasoning, coherence, and adaptation to dynamic environments, key gaps as LLMs evolve from text models to true agents

I had a lot of fun presenting at Measuring Intelligence Summit at #PyTorchCon 2025 on what it means for us to push the LLM capabilities for agentic intelligence and RL. In terms of benchmarks to measure agentic intelligence relevant to real world use cases, we discussed how the

More from this star-studded panel moderated by @joespeez, ft @sarahcat21, @saranormous, @brianzhan1, @pxd

So excited for @yijia_liang @peterjliu @yaozhaoai! VC data analysis is now finally automated by AI

Huge congrats to @yusan_lin and the Mirror Mirror AI team on joining a16z speedrun! Exciting to see fashion tech at the frontier of AI, redefining how creativity, data, and personalization intersect.

Full thread here-

Excited to have backed @reflection_ai from the seed. Today’s $2b raise pushes frontier open intelligence, with open models + advanced RL at scale, accessible to all. And most importantly, with @nvidia as their key partner. It’s a big day for open AI. Thanks NYTimes for covering

If you didn’t get a video with the Dynasaur did you really go to CoRL?

I’ve worked with @maxgazor for 3 years, and we backed @reflection_ai and @dynarobotics together. He’s an incredible mentor: intellectual, thoughtful, hands-on, unwavering. While others chase obvious companies at Series A, Max shows up with big checks to partner with founders

Congratulations to @axiommathai for launching. Self-improving reasoning is one of the most ambitious frontiers.

Today, @ekindogus and I are excited to introduce @periodiclabs. Our goal is to create an AI scientist. Science works by conjecturing how the world might be, running experiments, and learning from the results. Intelligence is necessary, but not sufficient. New knowledge is

Today, I am launching @axiommathai At Axiom, we are building a self-improving superintelligent reasoner, starting with an AI mathematician.