By the excellent @krishnanrohit: "Moltbook is simultaneously a milestone and a warning sign: open-ended interaction by itself does not guarantee diverse discourse, and populations of similar models can converge on shared templates. If we want agent societies to explore

Consequentialists coming out as virtue ethicists

New in Reader: save your podcast episodes to get a permanent, highlightable transcript

alignment debate: soul doc vs model spec

I just want to confirm that this is based on a real document and we did train Claude on it, including in SL. It's something I've been working on for a while, but it's still being iterated on and we intend to release the full version and more details soon.

I rarely post, but I thought one of you may find it interesting. Sorry if the tagging is annoying. https://t.co/m8PCIHF4xR Basically, for Opus 4.5 they kind of left the character training document in the model itself. @voooooogel @janbamjan @AndrewCurran_

🚀 Thrilled to launch DeepScholar, an openly-accessible DeepResearch system we've been building at Berkeley & Stanford. DeepScholar efficiently processes 100s of articles, demonstrating strong long-form research synthesis capabilities, competitive with OpenAI's DR, while running

This is one of the more valuable community notes I’ve seen. The hard thing about calculating externalities around AI is that it’s changing quickly.

We actually need more people with philosophy degrees working in tech companies (cc @mbrendan1 @cosmos_inst)

Here's my translation of Giorgio Agamben's recent text, "On Artificial Intelligence and Natural Stupidity." Agamben suggests that imagination is the key to the present situation, as it is the critical bond linking individuals with separate intelligence. But he asks, what happens

I was searching Google Scholar for recent publications on Gregory Bateson, cybernetics, and LLMs, and stumbled across this researcher who seems to be using Deep Research tools to pump out tons of pre-prints and publishing them to https://t.co/WuFd8VoyIf. The articles read like AI

Is there a website or account that tracks all the different ways current LLMs are limited and how to test through simple chatbot prompts? If not, would be an amazing website.

Around 45 min mark is excellent discussion of key flaw in RL: the reward is given to result, not the process. Right now it’s not easy for companies to reward the right process, so a lot of bizarre behavior is accidentally reinforced because it’s result-oriented.

You can now ship @nextjs apps to @chatgptapp

I’m on an English department committee that voted to create a subcommittee, and that subcommittee is considering the creation of a sub-subcommittee, and now I understand why the Soviet Union collapsed.

Good read for philosophy folks. In calling LLMs “ghosts,” Andrej is more playful but also more sophisticated. Sutton seems to think AGI happens only after we accurately duplicate natural intelligence in algorithmic fashion. But, as Dwarkesh asked in the pod, what if we’re

Anthropic’s definition of an agent

Happy to announce a book dedicated to Kant & AI, appears in Jan 2026 with Bloomsbury. It addresses: 1) What kinds of machines are intelligent? 2) Are machines capable of being moral? 3) Does an algorithm of perpetual peace exist? Paperback pre-order: https://t.co/gi6HyMQeXO

I would use OpenAI Deep Research (and equivalent products from other labs) a whole lot more if I'd seen the full list of tools that are available to them - and ideally their system prompts as well but that's less valuable to me than the tool definitions

This convo makes me really want to see Dwarkesh sit down with John Vervaeke. So much comes up here that Vervaeke tackles through cog-sci and philosophy. @DrJohnVervaeke @dwarkesh_sp