BREAKING
Sutton Launches Oak Lab in Toronto
0
T
parameter agent
0
W
energy budget
OaK vs Frozen LLMs
OaK Architecture
Oak Lab
●
Learns during operation
●
Reusable temporal skills
●
Options and Knowledge
LLM Paradigm
prevailing
●
Trained once, then frozen
●
Strong at imitation
●
Weak self-evaluation
How OaK Aims to Learn
1
Single-sample updates
↓
2
No data replay
↓
3
Event-driven nets
↓
4
Less compute
No Product or Benchmark Yet
An Alternative Research Path
AI NEWS BLITZ
Turing Award winner Richard Sutton has co-founded a startup called Oak Lab in Toronto.