Moksh Jain
@jainmoksh.bsky.social
PhD student at Université de Montréal and Mila
https://mj10.github.io/
https://mj10.github.io/
It's nice to see the elicitation perspective getting discussed! RL on CoT is really just a more reliable way of eliciting latent capabilities of the model than simple prompting. We took this perspective in our work (arxiv.org/abs/2310.04363), which was also one of the first to use RL on CoT.
Amortizing intractable inference in large language models
Autoregressive large language models (LLMs) compress knowledge from their training data through next-token conditional distributions. This limits tractable querying of this knowledge to start-to-end a...
arxiv.org
March 4, 2025 at 2:48 AM
It's nice to see the elicitation perspective getting discussed! RL on CoT is really just a more reliable way of eliciting latent capabilities of the model than simple prompting. We took this perspective in our work (arxiv.org/abs/2310.04363), which was also one of the first to use RL on CoT.