Once again, an LLM has a "meltdown". This time GPT-5 Mini started questioning it's own knowledge upon providing a snippet with non existing syntax.
| Discord | Mmesek#2901 |
| Github | https://github.com/Mmesek |
| Gravatar | https://gravatar.com/mmesek |
Mmesek
- 3 Followers
- 38 Following
- 2 Posts
Probably pronouncable version of above: Em-Em-Es-Ek
I'm not exactly sure what happened here but trying that prompt again triggered this strange "meltdown" with Llama 3 LLM (through DuckDuckGo) again.
(In terms of prompt itself... I was experimenting, trust me. Though if someone knows a good rhyme...)
After tinkering with prompt a little it seems that "quitting" makes model answer with "more serious answer" within it's original answer.
Other models seem to behave normally with this prompt.
