Dingal AI Research — Notes and reports https://dingal-ai-research.co.uk/notes/ Dingal AI Research is a one-person research practice studying the evaluation and behaviour of AI systems, and publishing the methods openly. en Sun, 23 Aug 2026 22:04:44 GMT A minimum reporting standard for AI evaluations https://dingal-ai-research.co.uk/notes/reporting-standard-for-evaluations/ https://dingal-ai-research.co.uk/notes/reporting-standard-for-evaluations/ Tue, 18 Aug 2026 00:00:00 GMT Note A short proposal for the least an evaluation result has to disclose before it can be compared with another result or repeated by anyone else. Failure compounds faster than per-step accuracy suggests https://dingal-ai-research.co.uk/notes/long-horizon-agent-reliability/ https://dingal-ai-research.co.uk/notes/long-horizon-agent-reliability/ Wed, 29 Jul 2026 00:00:00 GMT Report Measuring how language-model agents fail over long tasks. Per-step accuracy barely separates the systems; end-to-end success separates them a lot. The evaluation harness is now public https://dingal-ai-research.co.uk/notes/evaluation-harness-released/ https://dingal-ai-research.co.uk/notes/evaluation-harness-released/ Thu, 11 Jun 2026 00:00:00 GMT Release The code I use to run and record my own evaluations, released so the results here can be reproduced. A year of interpretability work, and what held up https://dingal-ai-research.co.uk/notes/a-year-of-interpretability-work/ https://dingal-ai-research.co.uk/notes/a-year-of-interpretability-work/ Thu, 15 Jan 2026 00:00:00 GMT Note Which methods for looking inside a model produced findings I would still defend, and which did not survive being checked. Why this exists https://dingal-ai-research.co.uk/notes/why-this-exists/ https://dingal-ai-research.co.uk/notes/why-this-exists/ Mon, 02 Sep 2024 00:00:00 GMT Note What this practice is for, what it will publish, and the things it will not do.