If you are believing what those things are popping out wholesale without double-checking to see if they’re feeding you fever dreams you are an absolute fool.
I don’t think I’ve seen a single statement come out of an LLM that hasn’t had some element of daydreamy nonsense in it. Even small amounts of false information can cause a lot of damage.
I’m in software and we’re experimenting with using it for certain kinds of development work, especially simpler things like fixing identified vulnerabilities.
We also have a pilot started to see if one can explain and document an old code base no one knows anymore.
you have to be there when the code was written and went through the various iterations.
Well, we don’t have that. We’re mostly dealing with other people’s mistakes and tech debt. We have messy things like nested stored procedures.
If all we get is some high level documentation of how components interact I’m happy. From there we can start splitting off the useful chunks for human review.
Are there any good open-source community-made models that aren’t owned by corporations or at least owned by a Non-Profit/ Public Benefit Corporation?
Hugging face open-r1 and up? It’s an open source deepseek I think.
What exactly is the benefit of using an LLM? Why would I bother using one at all?
I ask it questions all the time and it helps verify facts if I’m looking for more information
If you are believing what those things are popping out wholesale without double-checking to see if they’re feeding you fever dreams you are an absolute fool.
I don’t think I’ve seen a single statement come out of an LLM that hasn’t had some element of daydreamy nonsense in it. Even small amounts of false information can cause a lot of damage.
I’m in software and we’re experimenting with using it for certain kinds of development work, especially simpler things like fixing identified vulnerabilities.
We also have a pilot started to see if one can explain and document an old code base no one knows anymore.
Good code documentation describes why something is done, and no just what or how.
To answer why you have to understand the context, and often, you have to be there when the code was written and went through the various iterations.
LLMs might be able to explain what is done, with some margin of error, but why something is done, I would be very surprised.
Well, we don’t have that. We’re mostly dealing with other people’s mistakes and tech debt. We have messy things like nested stored procedures.
If all we get is some high level documentation of how components interact I’m happy. From there we can start splitting off the useful chunks for human review.
I can honestly see a use case for this. But without backing it up with some form of technical understanding, I think you’re just asking for trouble.
MistralAI look to be something along those lines