The best debugging move with any AI agent is to print the exact context it received before you touch a single line of the prompt.
Most "the model is dumb" moments are really "I fed it garbage and didn't check" moments
The AI pair programmer that lives in your terminal and commits its own work.
Aider edits your real repo, runs on any model, and writes sensible git commits as it goes.
github.com/Aider-AI/aider
devon.eth retweeted
AI is getting good enough that “Can it do this?” is becoming a less interesting question.
For many tasks the answer will eventually be yes.
So I am becoming more interested in another question:
Even if AI can decide something, should it?
> Which career fits me?
> Who should I trust?
> What should I believe?
> Is this risk worth taking?
> Should I leave?
> Should I stay?
> What should matter most now?
I want AI involved in some of those questions. But I don't necessarily want it answering them for me.
Capability and authority are different things.
We have spent years expanding what machines are capable of doing.
The next conversation may need to be about something much more human:
What authority are we willing to give them?