A big thanks to the The Index Podcast for having me on and discussing all things evals, benchmarking, and decentralized AI!

Some thoughts on why mech interp is the clearest path to allignment: lesswrong.com/posts/DBn83cvA… Please leave feedback if you are into AI Safety, allignment, and more!

In SF for a week, reply or dm HERE Technologies if you want to meet!

Announcing that I just became an a16z scout for a16z speedrun 🧊 🧢 It took us 1.5 years to go from nothing to an apartment in SF with a view of the Golden Gate Bridge, shipping code every week, with O1 VISAs and world class investors like General Catalyst and a16z speedrun 🧊. Every

Atlas V1 is now live and fully open to the public. Get up to date analytics on the performance of the latest AI models, such as OpenAI GPT 5, Anthropic's Claude 4.5 Sonnet, and more. Check it out here: app.layerlens.ai Read the official announcement here:

Check it out and give us feedback! Also, despite the name, we are not building an LLM browser.

Introducing LLM Eval Battlegrounds. Learn more here: medium.com/layerlens/atla…

Final Post in our series on the public release: medium.com/layerlens/atla… Check out all eval spaces here: app.layerlens.ai/evaluation-spa…

Most “open-source” LLMs today are not really open-source at all – they’re open-weights: you get a giant tensor file, but not the full story of how it came to be. Olmo 3 allenai.org/blog/olmo3 might be the most actually open open model we’ve seen so far. Most “open” LMs today

💡 Model launches create excitement, but most of the real cost sits behind the scenes. The MLOps market was $1.58B in 2024 and is projected to reach $19.55B by 2032. That growth reflects the operational burden of making models work in production. Organizations that invest in

LayerLens just launched a super cool evaluation dashboard about the China and US open source AI models. Check it out app.layerlens.ai/evaluation-spa…. Some perspectives : 👇1️⃣ The top tier is now a 4-model race The upper band is basically: • 🇨🇳 Qwen / DeepSeek • 🇺🇸 Llama / GPT-OSS