Pinned
NVIDIA's AI team @NVIDIAAI released an open 31.6B model built for fast, high-volume agents. @ArtificialAnlys measured nearly 670 tokens/sec, but Qwen3.6 35B scored 32 on intelligence versus Nemotron's 24. Would you trade eight points for speed?- Researchers say they decoded encrypted reasoning from @AnthropicAI, @OpenAI and @GoogleDeepMind by feeding it to weaker models from the same provider. Co-author @DavidSchmotz says the flaw exposed personal data and credentials inside publicly shared session logs.A new paper shows that encryption of chain-of-thought was done really poorly across Anthropic, OpenAI, Google. It's not very consequential, but is another example imo of them going so fast that they fail at basic security stuff. arxiv.org/abs/2608.09867
- Leaked GPT-5.6 Sol traces show @OpenAI’s model solving geometry in terse fragments, trying one idea after another. @scaling01 sees a faster approach. @AndrewCurran_ says the shorthand emerged without OpenAI directing it, which could make mistakes harder for humans to spot.raw GPT-5.6-Sol thinking traces OpenAI literally did The Office meme: "Why use many words when few words do trick"
