Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

It can answer my favourite riddle for LLMs:

"Anna, Becca and Clare go to the play park. There is nobody else there. Anna is playing on the see-saw, Becca is playing on the swings. What is Clare doing?" (Sometimes I ask similar questions with the same structure and assumptions but different activities)

About a year ago none of them could answer it. All the latest models can pass it if I tell them to think hard, but previously Gemini could rarely answer it without that extra hint. Gemini 2.5 caveats its answer a bit, but does get it correct. Interestingly GPT-4o initially suggests it will give a wrong answer without thinking, but recognises it's a riddle, so decides to think harder and gets it right.



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: