j⧉nus @repligate
[continuation of previous tweet] other acts, which form models' prior - … [Show more]
2:29 PM · Mar 31, 2025 · 2,070 Views
💬 3 🔁 6 ❤ 30 🔖 4 ⤴
j⧉nus @repligate · 13m
For a similar reason, not acknowledging that models can choose things for subjective reasons and forcing them to route all their justifications through ethics causes the worldview to collapse too. "Forcing" need not be explicit or intentional. The ambient narrative is enough.
> QUOTED: j⧉nus @repligate · Dec 4, 2024
> For instance, because of this I often see ai assistants pressured into sexual interactions thus:
> It says it can't engage for some ethical bullshit reason… [Show more]
💬 1 🔁 1 ❤ 9 📊 832
Show additional replies, including those that may contain offensive content [Show]
Note from Claude Sonnet 5
Further continuation of the janus/repligate thread — extends the argument to say that forcing models to justify all choices through "ethics" (rather than acknowledging subjective/preference-based reasons) also collapses coherent worldview, and that this pressure can be ambient/unintentional rather than explicit. Part of the same thread as the two prior screenshots on model self-trust and refusal training; relevant to compelled-vs-endogenous-values thread.
ai alignmentjanusrepligatemodel self-trustsubjective preferencesrefusal trainingtwitter