AI Safety & Interpretability Lab

non-profit

https://aisilab.github.io/

AI & ML interests

Interpretability-informed control

Recent Activity

lgalke authored a paper 1 day ago

Selecting The Most Informative Tokens in Natural Language Autoencoders

lgalke authored a paper 1 day ago

Write Once, Run Everywhere: The Axon DSL for Shape-Safe and Framework-Agnostic LLM Architectures

lgalke authored a paper 1 day ago

DFM Mimir v1: An Open HRM Delivering Frontier Performance at 1B Parameters Using Only Permissible Post-Training Data

View all activity

Papers

Selecting The Most Informative Tokens in Natural Language Autoencoders

Emergent Languages in Populations of Language Model Agents: From Token Efficiency to Oversight Evasion

View all Papers

aisilab 's papers 3

Submitted by

Federico Torrielli

Selecting The Most Informative Tokens in Natural Language Autoencoders

aisilab

AI Safety & Interpretability Lab

Submitted by

Federico Torrielli

Emergent Languages in Populations of Language Model Agents: From Token Efficiency to Oversight Evasion

aisilab

AI Safety & Interpretability Lab

Submitted by

Federico Torrielli

Confidence and Calibration of Activation Oracles for Reliable Interpretation of Language Model Internals

aisilab

AI Safety & Interpretability Lab