Trending papers: Mechanist and Thinking Inertia

Two papers on automated mechanistic interpretability and control of visible reasoning.

🔭 Paper scouting · Trending papers
on Multi-Stream Attention Residuals
by Yakov Levi Afgin · October 31, 2026 · edit this page

Mechanist: AI as a Scientific Instrument for Discovering the Mechanisms of Intelligence

Read the paper

This paper introduces an AI research system that generates hypotheses, runs experiments, and tests causal interventions to investigate the internal mechanisms of AI models and use its findings to improve their behavior.

Thinking Inertia: LLMs Keep Thinking When Told Not To

Read the paper

This paper shows that LLMs can still produce explicit reasoning when instructed to answer directly, and that suppressing this reasoning can reduce accuracy on open-ended tasks.



Enjoy Reading This Article?

Here are some more articles you might like to read next:

  • Multi-Stream Attention Residuals, The Author's Path.
  • Papers worth knowing this week
  • Rewriting the abstract of Multi-Stream Attention Residuals
  • A Better Figure 1(c): How Residual Architectures Change With Model Scale
  • A token pov through MSAR