Google DeepMind 'Wet Dream RSI' Paper Release
2 videos · 2 channels · score 5.5k
Google DeepMind and the University of Maryland released the Wet Dream RSI paper on AI self‑improvement via simulated exploration, with Fireship arguing it’s merely a faster search algorithm using Gemini, while Wes Roth claims it makes recursive self‑improvement almost free by testing thousands of policies in a discovery‑tree simulator, sparking debate over the speed of future AI progress
The coverage — 2 videos
Did Google just kickstart the intelligence explosion?
Google DeepMind and University of Maryland's 'Wet Dream RSI' paper claims Gemini improves its own exploration policy via cached discovery logs, but the creator argues it's not true J. Good RSI since the same Gemini writes every policy, making it merely a faster search algorithm with caching.
Google "RSI is here..."
The creator reacts to Google's Dream RSI paper, arguing that replaying thousands of simulated runs through the historical discovery tree makes AI self-improvement nearly free, with exploration deciding whether recursive self-improvement is worth its compute.