Ask HN: Why does LLM struggle in playing 24 puzzle

2 pointsyegle2 comments

I have yet to find an LLM that solves the 24 puzzle with these numbers (6 9 9 10) correctly.

Many seemingly capable LLM will produce incorrect intermediate or final results, e.g. claiming 9 × 9 ÷ (10 - 6) = 24.

Did I miss anything in my prompt? Or if there's a reason LLMs are bad at these questions?

CISO of Anthropic: "No LLM has had an emergent calculator yet." https://news.ycombinator.com/item?id=39591527

Because it's unsafe, according to [1]. I don't fully understand all the caveats listed there, but I won't take such a risk either :-)