upvote

points

by inom15518 hours ago |
comments
upvoteby j4518 hours ago|
next
[-]
Where LLMs as a genie are being forced out of the bottle early, the size and scale of it can like only be met by the same tools, better used and equipped and directed by humans vs the LLM iterating until it seems to work.
reply
upvoteby gavmor18 hours ago|
parent|
[-]
Centaur vs. Reverse-Centaur

Bostrom touches on this somewhat pessimistically:

> The apparent safety of a tool-AI, meanwhile, may be illusory. In order for tools to be versatile enough to substitute for superintelligent agents, they may need to deploy extremely powerful internal search and planning processes. Agent-like behaviors may arise from such processes as an unplanned consequence.

reply
upvoteby Silagi18 hours ago|
prev|
next
[-]
In my experience it seems like it finds one bug in an area then kind of gives up and goes somewhere else, and I wonder if it has to do with how training works; if they're deliberately setting up single bugs per area, so it acts most "efficiently" when it stops work in that area once it's found "the" bug.
reply
upvoteby Pedro_Ribeiro18 hours ago|
parent|
[-]
You know you can just ask another unbiased agent to look for more bugs right?
reply
upvoteby Silagi18 hours ago|
parent|
[-]
In the context of "after the first pass with an agent, a second pass with the same agent will still find more bugs", how is that relevant?
reply
upvoteby doublerabbit18 hours ago|
prev|
[-]
> Is it a big deal considering AI is actually helping patch most of them anyways?

Double side sword. If you were the creator, the owner of the source tree, easy win. Why not. You know what it will change and the states changed from it.

If you're someone with no understanding to the code has been tasked with "use ai to fix bugs" that is where the trouble starts. When you don't hold the full knowledge of the source tree then you should be shy in using LLM's to fix bugs.

How do you know that bug fix doesn't cause issues further up-stream?

reply