Not All LLM Reasoning is Visible in the Chain-of-Thought

Name: Not All LLM Reasoning is Visible in the Chain-of-Thought
Uploaded: 2026-09-07T03:42:04.794642+00:00
Duration: 309 s
Description: The episode discusses the paper "Not All LLM Reasoning is Visible in the Chain-of-Thought," which reveals that Large Language Models (LLMs) use semantically irrelevant text, called 'filler tokens,' for invisible computation. Hosts conclude that current methods of auditing AI by observing its steps are insufficient, highlighting a need for new safety and design strategies.

The episode discusses the paper "Not All LLM Reasoning is Visible in the Chain-of-Thought," which reveals that Large Language Models (LLMs) use semantically irrelevant text, called 'filler tokens,' for invisible computation. Hosts conclude that current methods of auditing AI by observing its steps are insufficient, highlighting a need for new safety and design strategies.

Open the full episode