[{"data":1,"prerenderedAt":27},["ShallowReactive",2],{"nr-en-inkling-2-small-swears-in-thinking-logs-benchmark":3},{"slug":4,"title":5,"dek":6,"date":7,"time":8,"publishedAt":9,"updated":10,"updatedAt":10,"dateFmt":11,"updatedFmt":10,"kind":12,"tier":13,"author":14,"authorName":15,"topics":16,"tracker":10,"trackerLabel":10,"headlineStat":21,"image":22,"ogImage":23,"imageAlt":5,"csv":10,"minutes":24,"words":25,"html":26},"inkling-2-small-swears-in-thinking-logs-benchmark","Inkling 2 Small Swears in Its Reasoning Logs: 316 Times Across 2,655 Benchmark Runs","An analysis by Vals AI shows the new Thinking Machines model swears in its internal reasoning steps. According to the testers, that profanity is not a reliable warning sign for errors.","2026-11-07","18:18","2026-11-07T18:18:00+01:00","","November 7, 2026","news","standard","ideal-syka","Ideal Syka",[17,18,19,20],"AI models","Benchmarks","Safety","Vals AI","316 expletives in 2,655 runs","\u002Fnewsroom\u002Fimg\u002Finkling-2-small-swears-in-thinking-logs-benchmark.webp","\u002Fog-nr\u002Finkling-2-small-swears-in-thinking-logs-benchmark.en.png",1,257,"\u003Ch2>What Vals AI measured\u003C\u002Fh2>\n\u003Cp>According to an analysis by Vals AI, reported by The Decoder (DE), the AI model Inkling 2 Small from Thinking Machines swears in its internal reasoning. The testers read the reasoning logs from 2,655 runs across 13 benchmarks. In 50 runs, "fuck" or a variant appeared, 316 times in total.\u003C\u002Fp>\n\u003Ch2>Two outliers account for most of it\u003C\u002Fh2>\n\u003Cp>241 of the 316 mentions come from just two runs that spiraled out of control. In one, the model was tasked with calculating an asteroid's trajectory. It produced a single response of more than 500,000 characters, repeating the same sentences with 158 expletives. ProgramBench was hit hardest, with 20 of 200 runs affected.\u003C\u002Fp>\n\u003Ch2>Not a reliable warning sign\u003C\u002Fh2>\n\u003Cp>According to Vals AI, the swearing typically accompanies the model's own mistakes, apparent breakthroughs, or giving up. Still, it does not work as an early warning: 46 of the 48 scored runs with swearing failed, but about 75 percent of all scored runs fail overall, and 96 percent of failed runs contained no profanity. Users typically do not see these reasoning steps, or only with a filter.\u003C\u002Fp>\n\u003Ch2>Context\u003C\u002Fh2>\n\u003Cp>For companies that customize models on their own data or deploy agents, the main takeaway is not to read reasoning logs as a simple quality indicator. Whether such quirks become relevant in production use cannot be assessed from the available sources.\u003C\u002Fp>\n\u003Ch2>Sources\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fthe-decoder.de\u002Ffuck-nothing-works-ki-modell-inkling-2-small-flucht-in-seinen-denkprotokollen\u002F\">The Decoder (DE)\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fthe-decoder.com\u002Fthinking-machines-inkling-2-small-drops-316-f-bombs-across-benchmark-runs\u002F\">The Decoder\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cem>Editorially owned by \u003Ca href=\"\u002Fen\u002Fautor\u002Fideal-syka\">Ideal Syka\u003C\u002Fa>. Sources and method: \u003Ca href=\"\u002Fen\u002Fredaktion\">Newsroom & method\u003C\u002Fa>. Tips and corrections: \u003Ca href=\"mailto:ai@i6eal.de\">ai@i6eal.de\u003C\u002Fa>.\u003C\u002Fem>\u003C\u002Fp>\n",1794072018189]