georgia_berg

Sand­bag­ging: How Models Use Re­ward-Hack­ing to Down­play Their True Capabilities

Pre­dict­ing Shifts in AI-driven Se­cu­rity Risks

In­tro­duc­tion to Corrigibility

What the Lud­dites Can Teach Us About So­cietal Re­sponse to AI

Agen­tic prop­erty-based test­ing: find­ing bugs across the Python ecosystem

AI Policy Tues­day: Pre­dict­ing Shifts in AI-driven Se­cu­rity Risks

AI Safety Thurs­day: Mon­i­tor­ing LLMs for de­cep­tive be­havi­our us­ing probes

AI Policy Tues­day: Open Global In­vest­ment as a Gover­nance Model for AGI

AI Safety Thurs­day: Model­ing and De­tect­ing De­cep­tive Alignment

AI Safety Thurs­day: The Limi­ta­tions of Re­in­force­ment Learn­ing for LLMs in Achiev­ing AI for Science

AI Policy Tues­day: Will the spec­tre of trans­for­ma­tive AI be more challeng­ing than the real thing?

AI Policy Tues­day: De­bunk­ing the US-Chi­nese AGI Race

AI Policy Tues­day: Redlines for AI

“If Any­one Builds It, Every­one Dies” Toronto Read­ing Group

AI Safety Thurs­day: Chain-of-Thought Mon­i­tor­ing for AI Control

AI Policy Tues­day: The Con­cept of Poli­ti­cal Space and AI Safety

AI Safety Thurs­day: Su­per­in­tel­li­gence Endgames

AI Safety Thurs­day: Tech­ni­cal AI Gover­nance—Mo­ti­va­tions, Challenges, and Advice

AI Policy Tues­day: The Case for Reg­u­lat­ing AI Com­pa­nies, Not AI Models

AI Safety Thurs­day: At­tempts and Suc­cesses of LLMs Per­suad­ing on Harm­ful Topics