@scaling01, who runs the LisanBench LLM reasoning benchmark, posted that efforts to monitor chain-of-thought outputs were doomed from the start and that mechanistic interpretability always made more sense. The comment appeared amid quotes and replies discussing OpenAI statements on model architectures. Yo Shavit, Nathan Labenz, Micah Carroll, Ryan Greenblatt, Noam Brown, Mikita Balesni, and David Pfau also posted in the thread, covering topics such as trust in OpenAI, risks from recurrent models, and calls for limits on opaque serial depth.