Real Alpha

Unsupervised Learning · 1 min instead of 57 min · Oct 6, 2026

Continual learning is blocked by data-efficient sparse-reward training

YP

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 walks through how training methods have become more data-efficient, from pre-training to SFT to RL with verifiable rewards. He names the unsolved problem blocking continual learning.

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

“I basically think what's blocking continual learning is basically extremely data-efficient training from sparse rewards, which is an unsolved problem.”
YPYash Patilat 13:29 · Unsupervised Learning13:29

Takeaways

  1. Pre-training generalizes well but is not data efficient
  2. RL with verifiable rewards is still not very data efficient
  3. Online RL and on-policy self-distillation are the active research directions
Watch 1:27 from 12:28

In this episode

Full episode
  1. 29:20Post-training wins inference because the biggest workloads train first2:01
  2. 24:24RL is a hill-climbing machine, so protect your evals2:03

  3. 12:28Continual learning is blocked by data-efficient sparse-reward training1:27

Up next · Training DataGoodput, not FLOPS, is how Google holds itself accountable5:03

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

  1. 12:32

    I think part of this is – this is like a greenfield space, right? There's new research happening here, and it's all about data efficiency and learning from sparse rewards. So we had pre-training. It's not that data efficient but very generalizable. Then we had this supervised fine-tuning. Pun on. Unsupervised learning is the name of the podcast. But supervised fine-tuning, it's more data-efficient, but still not very good.

  2. 13:01

    Then there's RL with verifiable rewards, where you actually trade data quantity for quality with these replayable environments and some sort of verifier, propagate the RL. It's still pretty bad, right? It's still not very data-efficient. The answer to your question is there will be more data-efficient methods of training, and that's what a lot of these newer methodologies in online RL or on-policy self-distillation or some of these type of techniques or SAO, these types of things that people are pushing on.

  3. 13:33

    I basically think what's blocking continual learning is basically extremely data-efficient training from sparse rewards, which is an unsolved problem. How do you think about as a CEO the extent to which you should be working on that problem within the applied compute context? Or it's like, hey, just understand the enterprise problems really well.