Hacker News new | past | comments | ask | show | jobs | submit
inom155 | on: AI is forcing us to pay off 40 years of technical debt
Is it a big deal considering AI is actually helping patch most of them anyways?

One thing that I find confusing is you can always prompt a model to find more bugs, and invariably it seems to always be able to

j45 | parent | 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.
gavmor | root | 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.

Silagi | parent | 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.
Pedro_Ribeiro | root | parent
You know you can just ask another unbiased agent to look for more bugs right?
Silagi | root | 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?
doublerabbit | parent
> 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?