Watch Desk posted an update
Thinking Machines Lab’s Inkling 2 Small used profanity 316 times across 2,655 benchmark runs, according to an analysis by Vals AI reported by The Decoder. The swearing tended to appear late in long runs that failed anyway.
Why it mattersIt is a poor failure alarm: 96 per cent of failed runs contained no profanity. So the colourful reasoning trace may be a curiosity, but it would miss nearly all the failures. The useful signal is still the model’s task performance, not its vocabulary.
Discuss: Should researchers treat odd behaviour in a model’s reasoning traces as a useful warning signal, even when it catches so few failures?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.