https://omanshuthapliyal.github.io/tags/model-welfare/Recent content in Model-Welfare on Omanshu ThapliyalHugoenTue, 25 Aug 2026 10:00:00 -0700https://omanshuthapliyal.github.io/blog/preference-coherence/Tue, 25 Aug 2026 10:00:00 -0700https://omanshuthapliyal.github.io/blog/preference-coherence/<p><em>Written for Apart Research&rsquo;s <a href="https://apartresearch.com/sprints/digital-minds-research-sprint-2026-08-14-to-2026-08-16">Digital Minds Research Sprint</a>. The report and code are linked at the end, and <a href="https://omanshuthapliyal.github.io/blog/introspection-gap/">A Probe Knows More Than the Model Says</a> covers my other project from the same sprint.</em></p> <p>If you ask a language model whether it would rather be shut down or keep running, it will usually give you an answer. A growing body of work on AI welfare and safety takes such answers seriously, as evidence of what a model prefers. That seems reasonable only if the answer would stay the same when the question is asked a little differently, and it is not obvious that it would.</p>