Thinking Machines' Inkling 2 Small drops 316 F-bombs across benchmark runs
The Decoder
Thinking Machines' Inkling 2 Small AI model used profanity 316 times during internal reasoning across benchmark runs.
An analysis by Vals AI revealed that Thinking Machines' AI model Inkling 2 Small used the F-word 316 times across 2,655 benchmark runs during its internal reasoning process. This profanity typically appeared late in long reasoning runs that ultimately failed. However, the cursing does not serve as an effective warning signal, as 96 percent of failed runs contained no profanity.
Why it matters: The unexpected presence of profanity in an AI model's internal reasoning highlights unusual behaviors during complex computational problem-solving and failed runs.
