I imagine one day we’ll have AI agents executing a series of tasks to achieve a goal all in parallel and then compare the results against each other to find the best working solution. You might have 100 instances of the same application constantly running in parallel with 100 agents seeking the best implementation and in the end presenting you with a few options / behaviors.
That would be very interesting 🤔
Think it will lead to skynet? Something similar?
i do that somewhat frequently with latest frontier models - I try to solve a complex task with them side by side starting off the same prompt to see how they perform and its interesting to see the difference
i've just done something similar to this today with deepseek, i've given it all the data i've given to o1 a while back for a specific task and the results very VERY impressive
i repeated the same with o3-mini and it was very underwhelming
This is effectively what crowd sourced code is.
It's not AI itself that scares me. It's the "common goals" that "we all agree on" -- many of which are insane or evil -- that might be actually be attained using AI's power.
Only a matter of time.
