Comment by koakuma-chan
Comment by koakuma-chan 17 hours ago
Not really. For example we still can’t get coding agents to work reliably, and I think it’s a memory problem, not a capabilities problem.
Comment by koakuma-chan 17 hours ago
Not really. For example we still can’t get coding agents to work reliably, and I think it’s a memory problem, not a capabilities problem.
On the other hand, test-time weight updates would make model interpretability much harder.