test time training
Updating or adapting a model at inference time so it learns from the task it is currently doing; offered as a near-synonym for continual learning.
As defined in Why I don’t think AGI is right around the corner
Where it comes up · 2
- Ex-Google CEO: What Artificial Superintelligence Will Actually Look Like w/ Eric Schmidt & Dave B2 claims
- Why I don’t think AGI is right around the corner1 claim
Most important claims · 3 of 3Search all claims for “test time training” →
- China clearly understands AI's strategic importance and is investing enormous amounts; the US has slowed them with chip controls, but China has found clever workarounds, and the problem is deepening with algorithmic changes (test-time training) that work on lesser-power chips.Ex-Google CEO: What Artificial Superintelligence Will Actually Look Like w/ Eric Schmidt & Dave B
- There are three scaling laws playing simultaneously in AI: foundation model growth (current), test-time training law (beginning), and reinforcement learning law (beginning), all following predictable scaling relationships.Ex-Google CEO: What Artificial Superintelligence Will Actually Look Like w/ Eric Schmidt & Dave B
- We will not see continual learning announced as fully solved in a single livestream; because labs are incentivized to release innovations quickly, we'll see a broken early version of continual learning (test time training) before something that truly learns like a human, giving lots of heads up before the bottleneck is fully solved.Why I don’t think AGI is right around the corner