I see @3blue1brown, I click.
I also enjoy hearing opinions on AI from people with related theoretical expertise but without the incentive to hype a SaaS product.
Yes. Failed attempts, backtracks, and error messages aren’t neutral. They become distractors. The winning strategy is “need-to-know basis” only. Don’t fight the agent. Transformers have no way to ignore errors once they’re in context, so errors easily turn into recurring patterns.
Here's a way to get a visceral sense of the dumb zone (starts at ~120K tokens on SOTA models):
1. Agent does something dumb
2. /rewind to before the stupid thing
3. /compact (assuming you're in a spot where compacting makes sense)
4. Run it again
Watch it get smarter
Failed attempts, backtracks, and error messages aren’t neutral. They become distractors. The winning strategy is “need-to-know basis” only. Don’t fight the agent. Transformers have no way to ignore errors once they’re in context, so errors easily turn into recurring patterns.
Hence, rewind.