PodcastIntel
Sign in Get Started Free
Neural intel Pod
Neural intel Pod

Why "Hidden Reasoning" in Filler Tokens Changes AI Safety Forever

Jul 26, 2026 · 00:36:15
AI Summary
  • Frontier models decouple computation from surface tokens, proving "hidden reasoning" in filler tokens.
  • DeepSeek V3 and Kimi K2 use filler tokens for multi-hop tasks like 2-fact addition.
  • Mechanistic Interpretability reveals attention shifts, impacting AI safety research and future models.

More from Neural intel Pod

Neural intel Pod
Aug 20, 2026 · 00:32:21
View all episodes →

Get AI Summaries for Every New Episode

Subscribe to Neural intel Pod and get AI summaries, guest tracking, and email digests delivered automatically.

Sign Up Free →