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
Login to add comment
Other posts in this group

Article URL: https://arxiv.org/abs/2505.11614
Comments URL: https://news.ycombinator.c
Very small, curated community posting thoughtful political, general, arts, gaming, etc news (specifically not just tech and “nerd” focused) that has some level of commenting functionality?
I LOV
Article URL: https://danq.me/2020/11/11/blink-and-marquee/
Article URL: https://www.folklore.org/Joining_Apple_Computer.html

Article URL: https://arxiv.org/abs/2503.05040
Comments URL: https://news.ycombinator.c

Article URL: https://www.coventry.gov.uk/coventry-light-rail