Unsupervised Learning · 2 min instead of 57 min · Oct 6, 2026
RL is a hill-climbing machine, so protect your evals
Yash Patil
CEO, Applied Compute
Jacob EffronHost · AI Investor, Redpoint Ventures
Growth investor in Abridge, Ramp and Physical Intelligence; ex-Flatiron Health.
About this segment
Yash says non-verifiable domains can be hill-climbed by turning them into proxy-verifiable ones with rubrics and expert answers. Defining the hill is the hard part, so private evals are a competitive asset.
From “Applied Compute CEO on the Limits of RL, the New AI Hyperscaler & Why Post-Training Wins Inference”, Oct 6, 2026. Watch the full episode
“it is surprisingly easy to hill climb on non-verifiable domains if you turn them into some proxy verifiable domain.”
Takeaways
- Rubric-based RL against expert answers works well as a proxy
- Different experts at different companies produce different models
- Public benchmarks get gamed, so keep business-specific evals private
In this episode
Full episode- 29:20Post-training wins inference because the biggest workloads train first2:01
24:24RL is a hill-climbing machine, so protect your evals2:03
- 12:28Continual learning is blocked by data-efficient sparse-reward training1:27Up next
Transcript
Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.
- 24:29
the way these projects started and the internal models that were being trained were on highly verifiable domains like math and coding. Partially because they're highly verifiable, partially because everyone at the lab loves math. But it was more of a form. I think it is surprisingly easy to hill climb on non-verifiable domains if you turn them into some proxy verifiable domain. So rubric-based RL and this sort of stuff actually works quite well.
- 25:03
So I would say the state where we're at right now is if it's not something that has one definitive answer, basically giving it an expert answer and grading it against that serves as quite a good proxy. So that's actually why different companies have different experts. And so optimizing to two different groups and two different companies can actually give you very different models. So to keep all your ex-employees off of the data labeling companies.
- 25:28
Exactly. Yeah, yeah. No, I think it's actually what we have with RL is we essentially have a hill climbing machine. The hardest part is actually defining the hill to climb, which is why evals, I think, is a really important thing for companies to, A, focus on building, and B, actually safeguard pretty carefully. If you are thinking about competition and stuff like that, if you look at every public benchmark, it's going to get benchmarked. If you have things that are very particular to how your business operates and you want to make your models really good at that, it's actually to your advantage not to tell everybody else what good and bad looks like so people can go and make their things really good at that. So that's why I think, yeah, it is really important to actually keep your evals and employees and all that sort of stuff. Your employees are not