https://stackresearch.org/tags/interpretability/Recent content in Interpretability on Stack ResearchHugoen-usThu, 02 Apr 2026 00:00:00 +0000https://stackresearch.org/research/trace-topology/Thu, 02 Apr 2026 00:00:00 +0000https://stackresearch.org/research/trace-topology/<p>When a program crashes, the stack trace does not explain the whole bug. It does something narrower and more useful: it shows where execution was, what called what, and which line broke.</p> <p>When a language model&rsquo;s reasoning goes wrong, the failure is usually harder to locate. The final answer may be fluent and wrong. The intermediate trace may drift quietly for a thousand tokens. There is often no structural map of what depended on what, and no obvious place to point and say: this is where the reasoning stopped holding together.</p>https://stackresearch.org/research/the-unaskable-question-machine/Mon, 16 Mar 2026 00:00:00 +0000https://stackresearch.org/research/the-unaskable-question-machine/<p>Ask a language model something it does not know, and it may admit uncertainty or invent an answer. Ask it something a policy forbids, and it may refuse. Those are familiar failure modes. They have names, benchmarks, mitigations, and whole taxonomies around them.</p> <p>There is another category that receives less attention: questions the model cannot engage with because the question contradicts the structure of the system being asked. Not a knowledge gap. Not a safety boundary. A structural impossibility.</p>