Increasing Skill Level Recruits Deeper Attention Layers in a Frozen Chess Transformer

·LessWrong··

Paper: Increasing Skill Level Recruits Deeper Attention Layers in a Frozen Chess TransformerTL;DR: Maia-3 is a transformer-based chess model that takes Elo (the standard metric for competitive chess skill) as an input to the pre-trained network, so you can vary the skill the network is conditioned on with no change to its weights. Turning that Elo dial up from 700 to 2500:Pushes the computation deeper, monotonically, for every chess piece and move type I measured.This "depth migration" happens m...

Read full article →

Related Articles

Two-tier encryption in the UK
ReturnoftheHack · Hacker News · 11h ago
F-Droid 2.0
daveoc64 · Hacker News · 7h ago
Creatine uptake enhances antitumor immunity
lormayna · Hacker News · 4h ago
Italian parliament votes for return to nuclear energy
geox · Hacker News · 1d ago
Google’s Project Suncatcher to put ML infrastructure in space
xnx · Hacker News · 8h ago