Source Posts on X
@TheZvi @teortaxesTex we discussed this substantially at the time (as did others in the community), see https://www.goodfire.ai/blog/intentional-design#developing-responsibly, https://www.lesswrong.com/posts/G9HdpyREaCbFJjKu5/it-is-reasonable-to-research-how-to-use-model-internals-in,, https://arxiv.org/abs/2505.13787, https://www.far.ai/research/the-obfuscation-atlas-mapping-where-honesty-emerges-in-rlvr-with-deception-probes
1:20 PM · Jul 15, 2026Goodfire vs Zvi would be fine, if Zvi had any ammunition They're obviously in the right, besides obviously Anthropic is doing similar things, just surreptitiously. We need to understand this regime better
1:23 PM · Jul 15, 2026@TheZvi @teortaxesTex it's frustrating to see a dogmatic, evidence-free response here. I would like to know what your theory of mind for us is - why do you think we're working on this?
1:21 PM · Jul 15, 2026