gwern

It Looks Like You're Trying To Take Over The World

This story was originally posted as a response to this thread. It might help to imagine a hard takeoff scenario using only known sorts of NN & scaling effects... > In A.D. 20XX. Work was beginning. "How are you gentlemen !!"... (Work. Work never changes; work is always hell.) >...

Mar 9, 2022419

Capability Phase Transition Examples

Feb 8, 202239

"Summarizing Books with Human Feedback" (recursive GPT-3)

Nov 15, 202124

EfficientZero: human ALE sample-efficiency w/MuZero+self-supervised

"Mastering Atari Games with Limited Data", Ye et al 2021: > Reinforcement learning has achieved great success in many applications. However, sample efficiency remains a key challenge, with prominent methods requiring millions (or even billions) of environment steps to train. Recently, there has been significant progress in sample efficient image-based...

Nov 2, 2021137

AI ALIGNMENT FORUM
AF

AI ALIGNMENT FORUM
AF

gwern

gwern

It Looks Like You're Trying To Take Over The World

EfficientZero: human ALE sample-efficiency w/MuZero+self-supervised

OpenAI announces GPT-3

Capability Phase Transition Examples

gwern

It Looks Like You're Trying To Take Over The World

EfficientZero: human ALE sample-efficiency w/MuZero+self-supervised

OpenAI announces GPT-3

Capability Phase Transition Examples

It Looks Like You're Trying To Take Over The World

Capability Phase Transition Examples

"Summarizing Books with Human Feedback" (recursive GPT-3)

EfficientZero: human ALE sample-efficiency w/MuZero+self-supervised

"Decision Transformer" (Tool AIs are secret Agent AIs)

gwern's Shortform

"Scaling Laws for Autoregressive Generative Modeling", Henighan et al 2020 {OA}