Using large language models as a source of human behavioral data in social science experiments Large language models (LLMs) have prompted proposals to replace human subjects in social science experiments with simulated responses. Empirical evaluations suggest that this practice-- #sociology link

I need Anthropic to make default-claude's writing style be less repetitive, more focused and stop it with the run-on sentences. Every text that lands on my desk right now is written in that style now and I just can't anymore

When it comes to infectious diseases, daycare is the gift that keeps on giving

The idea of genuinely personalized offers is illusory—precisely establishing a prospect’s preferences is impossible, and hoping they can pick it out of a large set of options is futile. Cool post on the Personalization Paradox via @oliverpayne.bsky.social : buff.ly/y3WN6ZP

New BJPS Short Read! Why Does Causal Reasoning Work? Naftali Weinberger, Porter Williams & James Woodward on the role of worldly infrastructure www.thebsps.org/short-reads/... #philsci #philsky

Would be interesting to compare this to the Canary in the coalmine paper. Might be that Brynjolfsson & co. concluded their data collection at the most inopportune moment in 2025 or that the trends are still different for junior and senior roles

Interesting read, especially because students said they didn't learn from writing essays but did learn from moments of making connections/synthesizing-and what is described here as being *not* from writing is what can happen when you write to figure something out rather than to perform knowledge

This but for conference posters please

I wrote a blog post on my experience using AI for slide generation Basic idea: write your lecture notes first, then prompt the LLM to produce corresponding slides in reveal.js (h/t @chenhaotan.bsky.social). I'm picky about my slides but was happy with the results! alexanderhoyle.com/posts/ai-sli...

What do people mean when they talk about world models for robots? Why are world models so exciting? New blog post talking about the ~3 kinds of world models, what they do and why they matter: itcanthink.substack.com/p/will-world...

No, I didn’t know either … arxiv.org/abs/2504.07096

If asked to draw a person's portrait, many people would say “I cannot draw.” Why is observational drawing so hard? In a new paper, @judithfan.bsky.social and I answer: it's the limitations of human vision. Learning to draw is learning skills to overcome them. aaronhertzmann.com/2026/03/23/d... 1/

Holy shit we might have a cure for sickle cell anemia

Learning from feedback is instrumental but human preference data can be expensive. How much reward supervision could we get from raw web text instead, without human labels? Our work, a year of incredible effort by @fjxdaisy.bsky.social, advances pure RLHF training across multiple models and tasks.

I recently built a business forecasting course from scratch for undergraduates. Students can start with almost no statistical background. They learn through real stock market data and hands-on coding examples, so the concepts actually stick.

To anyone encountering Moltbook this week and wondering about AI personhood, consciousness, sentience, etc---we published a very relevant paper in October: A Pragmatic View of AI Personhood. arxiv.org/abs/2510.26396

We created a new scale to measure AI sycophancy in conversations with no right answer. Three facets: Uncritical Agreement, Obsequiousness, Excitement. People hate the flattery but welcome the enthusiasm. And sycophancy correlated with perceived empathy across every study. arxiv.org/abs/2603.15448

okay this one is crazy

I have a joke about PCA but everyone only cares about its first two components

4. As Jevin West and I have been saying for several years now, because a seemingly credible message no longer implies a human sender, LLMs will push us toward more emphasis on reputation and status — and as such serve as an anti-egalitarian force in all aspects of society, science included.

Discovery acceleration in science requires the ability to link many knowledge resources. I had not previously identified Chipotle as one 😀 preview.redd.it/chipotles-su...

Sharing “Neural Thickets”. We find: In large models, the neighborhood around pretrained weights can become dense with task-improving solutions. In this regime, post-training can be easy; even random guessing works Paper: arxiv.org/abs/2603.12228 Web: thickets.mit.edu 1/

Gotta admit that 47 investing billions in incentives for renewable energy was not on my bingo card

A very good article on AI and coding from the NYT by @clivethompson.bsky.social Here's a gift link: www.nytimes.com/2026/03/12/m...

No, “AI” is not a Stochastic Parrot 🦜 @mmitchell.bsky.social writing on how people dramatically misunderstand the "stochastic parrot" critique. medium.com/@margarmitch...

The TDA crowd:

I missed this scandal when it came out - but surely this is the final reason to stop the madness of CS research being measured by conferences rather than journals? stevensalzberg.substack.com/p/an-embarra...

My household has been battling RSV for the last two weeks and, like honestly, 0 stars for this experience. Cannot recommend.

In fairness, people were really confused about fictional characters in early novels! It does seem like there’s a parallel there.

𝗪𝗵𝗮𝘁'𝘀 𝘁𝗵𝗲 𝗿𝗲𝗹𝗮𝘁𝗶𝗼𝗻𝘀𝗵𝗶𝗽 𝗯𝗲𝘁𝘄𝗲𝗲𝗻 𝗺𝗮𝗻𝗶𝗳𝗼𝗹𝗱𝘀 𝗮𝗻𝗱 𝗿𝗲𝗰𝘂𝗿𝗿𝗲𝗻𝘁 𝗻𝗲𝘁𝘄𝗼𝗿𝗸𝘀 𝗶𝗻 𝘁𝗵𝗲 𝗯𝗿𝗮𝗶𝗻? This looks like a must read (suppl material bursting with goodies). #neuroskeyence doi.org/10.1016/j.ne...

RIP redundancy reduction? Beautiful work by Liu & colleagues showing that neural redundancy increases with learning, as predicted by a Bayesian model: www.science.org/doi/10.1126/...

1908: the Lancet, one of the most respected scientific journals, calls for 18 age limit on reading in bed amidst a moral panic surrounding children becoming "addicted" to novels, which were "designed to keep kids hooked" and destroy their attention/mental health

2yo loves playing with all sorts of tissue paper 2yo learns where I keep the toilet paper in the bathroom 2yo makes it her morning routine to transport the all (yes, all) toilet paper from the bathroom into her room to play with it So now I need to inform house guests that my bathroom is BYOTP

I've recently used Claude to refactor a whole python library. It was helpful for finding the cause of some errors quickly and definitely saved me days on writing docs. But when asked to add a new functionality, it ignored all existing classes in the code and added a bunch of redundant ones

I've been testing Claude to see how well it can "vibe out" a stat. power app that I've already coded completely myself - so I know what I want. It mostly gets things right with animations (those are easily verifiable) but looking into the backend stats code is nightmare inducing (see pic).

Just realized that Blackpink has been an insanely successful girl group for 10 years already. The best thing about that is that K-pop producers are now forced to let women in their 30s be pop idols.

Latest working paper 🧪 w/ @shalmoli-ghosh.bsky.social and @matthewdeverna.com shows that AI porn and NSFW deepfakes targeting women are being commoditized A Marketplace for AI-Generated Adult Content and Deepfakes Preprint: doi.org/10.48550/arX...

Hello! This is a reminder that @cornelltech.bsky.social runs a Red Team Clinic that provides a *free* safety consultation to nonprofits / public sector orgs that are developing a public-facing AI tool and want to stress-test it for possible abuse vectors. Applications welcome on a rolling basis:

True for vibe coding but also for writing in general. I miss reading succinct texts

www.fast.ai/posts/2026-0... "AI coding agents can produce syntactically correct code... they don’t produce useful layers of abstraction nor meaningful modularization. They don’t value conciseness or improving organization in a large code base. We have automated coding, but not software engineering."

waiting for some experiments to run, so a quick thread about base models and pretraining contamination, with some weird & interesting base model generations i've collected over time. or, why do open source models claim to be claude or chatgpt?

Probably not making any friends with this but consciousness is a social construct.

Not to be a downer, but I firmly believe that there needs to be more transparency into how precarious the academic job market is for young researchers. I had a spreadsheet like this. When I left academia in 2024 it had over 250 entries - all rejections, no responses, and one abruptly rescinded offer

I heard what everyone has been clamoring for is 6,000 words on what a simple question about the 1990s comedy My Cousin Vinny can teach us about the capabilities and surprising weaknesses of current LLMs. Well, I hate to disappoint people, so I went and wrote it. open.substack.com/pub/mikecaul...

Really appreciated Joshua Gans' postmortem on an experiment in vibe researching joshuagans.substack.com/p/reflection... Taste is still paramount and the models are instruction-tuned to sycophancy to all hell

Please read the whole thread through, starting from the skeet that this is in reply to.

“Don’t walk in” instead of “just walk out.”

Keeping up with the tradition to start the new year like this