226 citations · 690 across the 20 of their papers we have counts for
9 papers · 1 filter
Debate Training Reduces Reward Hacking in RLAIF
Zachary Kenton, Lili Janzer, Rory Greig +8
We demonstrate that RL finetuning an LLM using debate, a two-player adversarial game between a generator and a critic adjudicated by a weaker LLM judge, reduces reward hacking comp…
On scalable oversight with weak LLMs judging strong LLMs
Zachary Kenton, Noah Y. Siegel, János Kramár +8
Scalable oversight protocols aim to enable humans to accurately supervise superhuman AI. In this paper we study debate, where two AI's compete to convince a judge; consultancy, whe…
Challenges with unsupervised LLM knowledge discovery
Sebastian Farquhar, Vikrant Varma, Zachary Kenton +3
We show that existing unsupervised methods on large language model (LLM) activations do not discover knowledge -- instead they seem to discover whatever feature of the activations…
Explaining grokking through circuit efficiency
Vikrant Varma, Rohin Shah, Zachary Kenton +2
One of the most surprising puzzles in neural network generalisation is grokking: a network with perfect training accuracy but poor generalisation will, upon further training, trans…
Goal Misgeneralization: Why Correct Specifications Aren't Enough For Correct Goals
Rohin Shah, Vikrant Varma, Ramana Kumar +4
The field of AI alignment is concerned with AI systems that pursue unintended goals. One commonly studied mechanism by which an unintended goal might arise is specification gaming,…
Safe Deep RL in 3D Environments using Human Feedback
Matthew Rahtz, Vikrant Varma, Ramana Kumar +3
Agents should avoid unsafe behaviour during both training and deployment. This typically requires a simulator and a procedural specification of unsafe behaviour. Unfortunately, a s…
Imitating Interactive Intelligence
Josh Abramson, Arun Ahuja, Iain Barr +26
A common vision from science fiction is that robots will one day inhabit our physical spaces, sense the world as we do, assist our physical labours, and communicate with us through…
Generalizing from a few environments in safety-critical reinforcement learning
Zachary Kenton, Angelos Filos, Owain Evans +1
Before deploying autonomous agents in the real world, we need to be confident they will perform safely in novel situations. Ideally, we would expose agents to a very wide range of…
Three Factors Influencing Minima in SGD
Stanisław Jastrzębski, Zachary Kenton, Devansh Arpit +4
We investigate the dynamical and convergent properties of stochastic gradient descent (SGD) applied to Deep Neural Networks (DNNs). Characterizing the relation between learning rat…