Nathan Lambert

I regularly hear from people that my post-training book/writing has helped them get a job at labs by focusing on material that develops intuitions for how training methods work. It's very validating, because this type of "idea/math heavy" writing doesn't really sell many books. It's rewarding the right people who put the effort into learning. I'm happy I made the time for it.

With the data center slowdown coming after the election, I find this post increasingly likely. There’s way more incentive in a path that seems eminently doable. I consider this in line with the acceleration in AECI from the recent METR report. Coding agents and inference compute changes how we work — giving an acceleration in our battle against scaling laws’ exponentials. It doesn’t mean more accelerations are inevitable.

open.substack.com/pub/robotic/p/…

I’m building a neo non profit not a neo lab.

All the inference startups are in a very hard place. They have the capital today to try and secure a more lasting solution — many players coming for their scalps. Need build models, their own supply, or some magic post-training solutions.

Was the topic of my recent piece.

Looks like the smartest western open weight model by a margin. While much smaller than muse spark open or mistal large. While with a better license (apache). Love it!