Decoding intermediate activations in llama-2-7b
Date : 2023-07-21
Introduction
In line with previous research, Nina Rimsky found that the decoded block outputs at most layers, except a few early ones, were interpretable. She also found that the other intermediate outputs were interpretable and provided some intuition on what different layers were responsible for. Some very interesting insights in the post.
Read article here
Recently on :
Artificial Intelligence
Information Processing | Computing
PITTI - 2026-07-14
What Happens When AI Agents Have to Find Their Own Market?
Implementation of the Scaling Trust project : early lessons from simulating discovery, negotiation, and trust between autonomou...
PITTI - 2026-03-05
Scaling Trust : a Missing Piece in Multi-Agent Worlds
Humanity’s ability to build complex civilizations relies on an "invisible infrastructure" - the shared culture, institutions, a...
PITTI - 2026-01-14
Cultural, Ideological and Political Bias in LLMs
Transcription of a talk given during the work sessions organized by Technoréalisme on December 9, 2025, in Paris. The talk pres...
WEB - 2025-11-13
Measuring political bias in Claude
Anthropic gives insights into their evaluation methods to measure political bias in models.
WEB - 2025-10-09
Defining and evaluating political bias in LLMs
OpenAI created a political bias evaluation that mirrors real-world usage to stress-test their models’ ability to remain objecti...