New AI introspection work with Harvey! Came in skeptical the direct access story would hold but found this series of experiments compelling. (Also, for my fellow 2010s-era psycholinguists: come for the AI introspection, stay for the Brysbaert norms.) arxiv.org/abs/2603.05414

Dissociating Direct Access from Inference in AI Introspection

Introspection is a foundational cognitive ability, but its mechanism is not well understood. Recent work has shown that AI models can introspect. We study their mechanism of introspection, first exten...

arxiv.org

Harvey Lederman@harveylederman.bsky.social · 5mo ago

Can large language models *introspect*? In a new paper, @kmahowald.bsky.social and I study the MECHANISM of introspection in big open-source models. tldr: Models detect internal anomalies through DIRECT ACCESS, but don't know what the anomalies are. And they love to guess “apple” 🍎

Can large language models *introspect*? In a new paper, @kmahowald.bsky.social and I study the MECHANISM of introspection in big open-source models. tldr: Models detect internal anomalies through DIRECT ACCESS, but don't know what the anomalies are. And they love to guess “apple” 🍎

This essay is by far the best along its line, but the more I reflect about this stuff the more I think it's hard to hold an image of human experience, human thought, human understanding, human life, and human relationships as meaningful ends while also seeing them as dead-ends

Harvey Lederman@harveylederman.bsky.social · 10mo ago

Essay here: scottaaronson.blog?p=9030

Thanks for the kind words, and thoughtful response @peligrietzer.bsky.social ! I'm not here as much, but I put some responses on the other site: x.com/LedermanHarv...

Peli Grietzer @peligrietzer.bsky.social · 9mo ago

This essay is by far the best along its line, but the more I reflect about this stuff the more I think it's hard to hold an image of human experience, human thought, human understanding, human life, and human relationships as meaningful ends while also seeing them as dead-ends

Enjoyed this nice piece by the great Justin Tiwald on autonomy and morality in Confucianism. Not sure I love the clickbait title, but I love the work Justin is doing uncovering views about moral deference and moral autonomy in (neo)Confucianism... iai.tv/articles/the...

The radical independent thinking in Chinese philosophy | Justin Tiwald

iai.tv

Very excited to be going to Chicago for @agnescallard.bsky.social's famous Night Owls next week! I'll be discussing my essay "ChatGPT and the Meaning of Life". Hope to see you there if you're local!

UT Austin Linguistics is hiring in computational linguistics! Asst or Assoc. We have a thriving group sites.utexas.edu/compling/ and a long proud history in the space. (For instance, fun fact, Jeff Elman was a UT Austin Linguistics Ph.D.) faculty.utexas.edu/career/170793 🤘

The professor I'm currently TAing for is making students use an extension called 'Process Feedback' that tracks key logs and time on the document: processfeedback.org

See how you write or use AI | Process Feedback Every Student’s Work Has a Story |

Process Feedback enables teachers and students to see the writing process and AI usage. It helps students reflect on their writing and the role of AI.

processfeedback.org

If you're doing *any* out of class assessment, you're incentivizing AI use and harming students who do the work themselves. But some day we have to assess writing again. The solution is monitored computer-labs. What Universities are building these? We need to push for them.

Anthropic recently announced that Claude, its AI chatbot, can end conversations with users to protect "AI welfare." Simon Goldstein and @harveylederman.bsky.social argue that this policy commits a moral error by potentially giving AI the capacity to kill itself.

Claude’s Right to Die? The Moral Error in Anthropic’s End-Chat Policy

Anthropic has given its AI the right to end conversations when it is “distressed.” But doing so could be akin to unintended suicide.

lawfaremedia.org

Simon Goldstein and I have an op-ed live in Lawfare today! Anthropic's policy is premised on the idea that AI is a potential welfare subject. We argue that if you take that idea seriously (we don't take a stand on it here), the policy commits a moral mistake on its own terms.

Lawfare@lawfaremedia.org · 10mo ago

Anthropic recently announced that Claude, its AI chatbot, can end conversations with users to protect "AI welfare." Simon Goldstein and @harveylederman.bsky.social argue that this policy commits a moral error by potentially giving AI the capacity to kill itself.

You can read our full review (without a paywall) @ philpapers.org/archive/GOOK.... And you can also check out Harvey's paper that inspired us here: philpapers.org/archive/LEDU...

Now out in @science.org: @chazfirestone.bsky.social and I review Steven Pinker's new book "When Everyone Knows that Everyone Knows...". We learned a ton from it, but think its central thesis—that common knowledge explains coordination—faces a powerful challenge. 🧵 www.science.org/doi/10.1126/...

Knowledge for two

A psychologist explores common knowledge and coordination

science.org

I really enjoyed reading Steven Pinker’s new book and thinking through what it would take to share infinitely iterated knowledge with someone (I know that you know that I know that you know…). In @science.org, my colleague @jeremygoodman.bsky.social & I briefly give our perspective on this issue.

Now out in @science.org: @chazfirestone.bsky.social and I review Steven Pinker's new book "When Everyone Knows that Everyone Knows...". We learned a ton from it, but think its central thesis—that common knowledge explains coordination—faces a powerful challenge. 🧵 www.science.org/doi/10.1126/...

Wow @utaustin.bsky.social maybe @utaustinihs.bsky.social maybe there are other ways to shout out to the department of East Asia Studies and the Department of History, but however that may be, they are doing some amazing stuff with in house games for their JapanLab! Just found even more stuff here.

Simon Goldstein and I have a new paper, “What does ChatGPT want? An interpretationist guide”. The paper argues for three main claims. philpapers.org/rec/GOLWDC-2 1/7

Simon Goldstein & Harvey Lederman, What Does ChatGPT Want? An Interpretationist Guide - PhilPapers

This paper investigates LLMs from the perspective of interpretationism, a theory of belief and desire in the philosophy of mind. We argue for three conclusions. First, the right object of study ...

philpapers.org

Jonathan Lear's Aristotle: the Desire to Understand was pivotal in some of my first encounters with Aristotle. I found Aristotle and Logical Theory later, but it became a key inspiration for how to think about core parts of the corpus...1/2

Max Weber, 1917:

Go grue

emmie malone 🏳️‍⚧️@emmiemalone.bsky.social · 11mo ago

I’ve been training my whole life for this moment

What are the most important new ideas in normative ethics from this century?

📣@futrell.bsky.social and I have a BBS target article with an optimistic take on LLMs + linguistics. Commentary proposals (just need a few hundred words) are OPEN until Oct 8. If we are too optimistic for you (or not optimistic enough!) or you have anything to say: www.cambridge.org/core/journal...

Something I cherish about analytic philosophy is that no matter how famous you are or how profound your ideas sound, it's still your job to answer all the objections. I wish public promoters of philosophy held themselves to the same standard.

Can AI models introspect? What does introspection even mean for AI? We revisit a recent proposal by Comșa & Shanahan, and provide new experiments + an alternate definition of introspection. Check out this new work w/ @siyuansong.bsky.social, @harveylederman.bsky.social, & @kmahowald.bsky.social 👇

Siyuan Song@siyuansong.bsky.social · 11mo ago

How reliable is what an AI says about itself? The answer depends on whether models can introspect. But, if an LLM says its temperature parameter is high (and it is!)….does that mean it’s introspecting? Surprisingly tricky to pin down. Our paper: arxiv.org/abs/2508.14802 (1/n)

exciting new paper from Siyuan! I really enjoyed working with him on this, inspired by important work by Murray Shanahan and Julia Comsa. Hard questions about how to operationalize the notion of “introspection” that’s relevant for practical applications in AI today. Hope you’ll check it out!

Siyuan Song@siyuansong.bsky.social · 11mo ago

How reliable is what an AI says about itself? The answer depends on whether models can introspect. But, if an LLM says its temperature parameter is high (and it is!)….does that mean it’s introspecting? Surprisingly tricky to pin down. Our paper: arxiv.org/abs/2508.14802 (1/n)

Great piece as usual. Particular fun this week to see the shoutout to Adrienne Raphel and the close reading of @rkubala.bsky.social's lovely paper on the aesthetics of crosswords!

Kieran Setiya@ksetiya.bsky.social · 12mo ago

I wrote about crossword puzzles: Under the Net ksetiya.substack.com/p/compositio...

My piece is featured in today's Browser. If you don't know it the Browser is a phenomenal newsletter that puts together fascinating pieces from all over the web. Link in first comment--you should subscribe! @uribram.bsky.social

Harvey Lederman@harveylederman.bsky.social · last yr.

I wrote about automation and the meaning of life, as a guest post on Scott Aaronson's Shtetl-Optimized. (1/5) scottaaronson.blog?p=9030

(1/5) 🤖 Harvey Lederman asks: What if AI doesn’t destroy us—just does everything better? From polar explorers to master mathematicians, he traces our drive for discovery and purpose. If machines take that work, what’s left for us? 🔗 scottaaronson.blog?p=9030&ref=t... @harveylederman.bsky.social

ChatGPT and the Meaning of Life: Guest Post by Harvey Lederman

Scott Aaronson’s Brief Foreword: Harvey Lederman is a distinguished analytic philosopher who moved from Princeton to UT Austin a few years ago. Since his arrival, he’s become one of my …

scottaaronson.blog

Enjoyed this! Question for @mraginsky.bsky.social: even if detailed future scenarios are "big worlds", won't a reasonable decision theory for them be (effectively) similar to one that weights small risks of very bad outcomes very highly?

Maxim Raginsky@mraginsky.bsky.social · 12mo ago

Also, there were interesting connections to my recent essay on superintelligent machines and thought experiments: realizable.substack.com/p/supertzar-...

I appreciated the wide range of references here, from Edith Wharton to the Buddhist Pali Canon. However, this essay resonated most with my recent reading of Richard Wollheim’s _The Thread of Life_, which presented a vision of what it means to lead a life …

Harvey Lederman@harveylederman.bsky.social · last yr.

I wrote about automation and the meaning of life, as a guest post on Scott Aaronson's Shtetl-Optimized. (1/5) scottaaronson.blog?p=9030

My 2 cent: humans might end up looking for meaning in some of the old places. Religion, which has been gradually pushed to the margins of our post-Enlightment society, might make a comeback—not only because it is comforting, but also because it is typically set aside as a result of education. 1/3

ChatGPT and the Meaning of Life: Guest Post by Harvey Lederman

Scott Aaronson’s Brief Foreword: Harvey Lederman is a distinguished analytic philosopher who moved from Princeton to UT Austin a few years ago. Since his arrival, he’s become one of my …

scottaaronson.blog