jwpapi
6 hours ago
Can somebody explain me how we can trust these AIs on scale. It’s a simple task that every experienced dev will know the solution right away.
Opus goes wrong 3 times, wastes a lot of time and obviously giving wrong suggestions.
lacion
6 hours ago
the whole point is that you guide them, you make sure they dont make this mistakes.
the power of ai is not that its smarter or better than you (at least yet) its that it will just keep grinding while your thinking, making new things, whole the context that it cant.
you can set up and agent with a good harness to make sure the code will be allright, with good guardrail while you do elevated work.
use them as the tool they are not as the tool you wish them to be.
jwpapi
2 hours ago
I agree, that is my point. It’s just contra the whole agent logic that is pushed atm.
functional_dev
6 hours ago
Have you tried to feed "tsc" error messages to Opus? That should reset model's logic