Was starting to mess around with the latest LLM models and found that they're not great at counting lines in files.
I gave Gemini 2.5 flash a python script and asked it to tell me what was at line 27 and it consistently got it wrong. I tried repeatedly to prompt it the right way, but had no luck.
https://g.co/gemini/share/0276a6c7ef20
Is this something that LLM bots are still not good at? I thought they had gotten past the "strawberry" counting problems.
Here's the raw file: https://pastebin.com/FBxhZi6G
Comments URL: https://news.ycombinator.com/item?id=44176179
Points: 8
# Comments: 7
Jelentkezéshez jelentkezzen be
EGYÉB POSTS Ebben a csoportban
Made a collaborative, persistent state Mondrian-style painting editor. Feedback welcome!
Comments URL: https://news.ycombinator.

Just a note of warning from personal experience.
Companies don’t really need non-competes anymore. Some companies take an extremely broad interpretation of IP confidentiality, where they consid

Article URL: https://www.lemonde.fr/en/science/article/

Article URL: https://www.space.com/astronomy/astronomers-