←back to thread

Some thoughts on LLMs and software development

(martinfowler.com)
416 points floverfelt | 1 comments | 28 Aug 25 18:52 UTC | HN request time: 0.207s | source
Show context
sebnukem2 ◴[28 Aug 25 19:28 UTC] No.45056066[source]▶
> hallucinations aren’t a bug of LLMs, they are a feature. Indeed they are the feature. All an LLM does is produce hallucinations, it’s just that we find some of them useful.

Nice.

replies(7): >>45056284 #>>45056352 #>>45057115 #>>45057234 #>>45057503 #>>45057942 #>>45061686 #
tptacek ◴[28 Aug 25 19:48 UTC] No.45056284[source]▶
In that framing, you can look at an agent as simply a filter on those hallucinations.
replies(4): >>45056346 #>>45056552 #>>45056728 #>>45058056 #
Lionga ◴[28 Aug 25 20:14 UTC] No.45056552[source]▶
Isn't an "agent" not just hallucinations layered on top of other random hallucinations to create new hallucinations?
replies(1): >>45056610 #
tptacek ◴[28 Aug 25 20:20 UTC] No.45056610[source]▶
No, that's exactly what an agent isn't. What makes an agent an agent is all the not-LLM code. When an agent generates Golang code, it runs the Go compiler, which is in the agent's architecture an extension of the agent. The Go compiler does not hallucinate.
replies(1): >>45056862 #
Lionga ◴[28 Aug 25 20:48 UTC] No.45056862[source]▶
The most common "agent" is an letting an LLM run a while loop (“multi-step agent”) [1]

[1] https://huggingface.co/docs/smolagents/conceptual_guides/int...

replies(1): >>45057232 #
1. tptacek ◴[28 Aug 25 21:27 UTC] No.45057232[source]▶
That's not how Claude Code works (or Gemini, Cursor, or Codex).