I was comparing LLMs with some common test phrases in an attempt to improve my usage. I asked Qwen3-235B (reasoning) "is there a seahorse emoji?
Beware false memories"
Which, after a minute of thinking, eventually gets stuck as shown:
Expected the stop button to stop thinking, but it only seems to work during responding. It turns gray but the model keeps thinking.
The model eventually stops because it hits the limit, but it would've been nice to stop it earlier!
I'm not too worried about Qwen 235B but wouldn't want someone to have this happen with a more expensive model and no way to stop it.
