This morning, I was working on a book chapter on debate and AI safety. I asked Claude for some ideas as to how to make the debate multi-agent and shared what I had already written for the initial debate.
Claude decided to build me an app and started coding.
It produced an app.
The simple app works.
It gave all the debaters scores.
Agent precaution is a bit dishonest.
I’m not really sure what to make of this. I’m glad that Claude didn’t create an autonomous agent for my forthcoming fiction book based on the Terminator.
PS. I have two Claude accounts. One I pay for, and the other I used to test what people can do for free. I did this on the free account.
PPS. At the recent Code with Claude event, Dario Amodei mentioned that the world hasn’t really thought about what it means economically, and for us as a society at large, when we can create applications for anything we want in seconds. It gets even more complicated when we think about the fact that sometimes apps will get created when we weren’t even trying.