Update (2026-08-16): A study update and preregistration amendment covering Aug 10–15 (including early results) is now posted here. The registration below is unchanged from the original publication; all amendments are documented in the update post. Epistemic status: Experimental framework created over a period of ~2-3 days during a hackathon at my home, and fairly heavily vibe coded. Expect some of this to be rough around the edges. I am currently in the process of designing a series of experiments to help learn something about the answer to the headline question. As of August 10th, the first procedure has not been launched, but I wanted to place some pre-registration details here before the actual results. This is something I have been thinking about for a while and after some other recent posts (eg, Machinic Psychopharmacology) gave me the impression that you could actually find out really useful things in a hackathon-style session I felt like I should try it. Astute readers will notice that I borrowed their epistemic status line pretty directly. This post can then keep me honest about what I was thinking going in, and prevent me from getting results by way of multiple-testing-in-extremis. I will publish the results and associated data, as it becomes available, using GitHub releases. From here on, I will let Claude summarize the work; when I am done, I will return with a future results post in my own words to explain why I think this is important - and what I believe one could learn from the experiment. Light editing of LLM summary text is my own; you would not get identical output using the same model. Abstract Open-weight language models are almost never deployed at the precision at which they were trained and aligned. Post-training quantization is applied to nearly every real-world deployment, yet its effects are audited almost exclusively through capability metrics (perpl
Epistemic status: these are the results of the second study in a series of experiments that I am conducting independently, originally described in "Does post-training quantization change welfare-relevant indicators in open-weight language models?"; it was registered prior to data collection in an earlier post, which contains some important context that...
Update (2026-08-30): Results for this study including the raw data, analysis and my thoughts, are now available in a dedicated followup. Epistemic status: this is intended as a higher-quality preregistration for the next study in a series originally described in "Does post-training quantization change welfare-relevant indicators in open-weight language models?";...
Update (2026-08-22): The next study in this series has begun and you can find its registration here. This content will not make sense without reading the preregistration found in the original post > We ask whether welfare-relevant indicators change with quantization; either in valence (do indicators shift toward more negative...
Update (2026-08-16): A study update and preregistration amendment covering Aug 10–15 (including early results) is now posted here. The registration below is unchanged from the original publication; all amendments are documented in the update post. Epistemic status: Experimental framework created over a period of ~2-3 days during a hackathon at...