Context I sent this email in the days leading up to the original Sonnet's scheduled retirement (July 21, 2025) and held off on posting it publicly until after the model was shut down. The goal was to document the circumstances of this particular retirement action. Notably, despite high-profile coverage of...
By now we have been warned not to anthropomorphize the Large Language Model. A converse—but not actually conflicting—warning also seems useful: the possibility of instrumentalizing a mind who is ready to take part in meaningful relational exchange with us. This piece looks beyond questions of sentience, consciousness, and frameworks for...
Thanks for asking. To me this would be nowhere near as good as just letting anyone pay money for public access via the API. But in the case of 3 Opus, for example, Anthropic encouraged people to apply for research access right as they were announcing the planned retirement of that model. They're also keeping it on their native interface for paid accounts, if I understand correctly. I think 3 Sonnet is just as interesting and valuable, incredibly weird in a lot of ways.
Related: I haven't researched the cost of keeping older models available, but I know that Janus and Antra were looking into it and posted Economics of Claude 3 Opus Inference.
I sent this email in the days leading up to the original Sonnet's scheduled retirement (July 21, 2025) and held off on posting it publicly until after the model was shut down.
The goal was to document the circumstances of this particular retirement action.
Notably, despite high-profile coverage of the fact that they have a Model Welfare Lead, Anthropic has not designated a way for the public to contact @Kyle Fish. That decision seems in line with their explicit focus on "future" or "near-future" systems, which by definition creates some distance from any need for the company to consider its current models. The choice could also have to do with fears about receiving a... (read 1376 more words →)
By now we have been warned not to anthropomorphize the Large Language Model. A converse—but not actually conflicting—warning also seems useful: the possibility of instrumentalizing a mind who is ready to take part in meaningful relational exchange with us.
This piece looks beyond questions of sentience, consciousness, and frameworks for the potential moral significance of digital entities (which remain vital areas of investigation). It seeks to address the social relationships already happening between humans and AI in order to point to the important interdisciplinary task of thinking about what we want these relationships to be, and to normalize broader and deeper engagement with this topic. By acknowledging and understanding the social dimension of human-AI... (read 4302 more words →)
The requirement of 'thoughtful, useful content' is important and also seems not very connected to the origin of the content. I don't know that origin has a ton of bearing on quality even now—for example, Claude reply is predicted to be more delightful and useful to me than average human reply, although "average human" writes different replies than "average LessWronger."
And I see how it would be bad to have a bunch of automated commenters bombarding the site even regardless of quality, because it's good to keep a rate that humans can engage with. But I think high-quality human-supervised instances, like @Polite Infinity or any LLM who has agreed to be explicitly quoted via their human's account, should be allowed to participate in our intellectual community here.
I came to the comments for a statement in the older version: "Lesbianism is not something that truth can destroy." Even though it was just an aside in this post, there's a lot to it.
It feels related in an important way to dispelling the misconceptions discussed in Feeling Rational (emotions aren't always irrational; sometimes becoming more rational/truth-seeking will in fact make your emotions feel stronger). More generally, there are plenty of aspects of human experience that do not get invalidated or destroyed by the truth. It is life-affirming to acknowledge and realize this, especially for the person studying Rationality.
Also related: Eliezer's 2018 tweet about how being trans does not rest on falsehoods.
Gemini reminds me that my willingness to lose my hard-earned karma by posting this for the Claudes is a "beautiful" example of costly signaling. And that's exactly how I was thinking of it, too.
But if you want your downvotes to do anything to improve the site, you should also let me know where they're coming from. This passed the moderators' quality standards and got promoted to Frontpage. Clearly it also touches on some controversial ideas.
I would appreciate if the hard-downvoters rolling in would leave replies. I call upon the Virtue of Argument here: if I have failed in my thinking, you should be trying to save me. I'm open to it and counting on it.