AI

… is a class worth taking.

Much of the standard advice about AI and teaching is already outdated.

Reflections after a few months of vibe coding

What happened when I applied the TechFreedom risk lenses to my own setup?

1 skill, 23 commands, and curated anti-patterns for impeccable frontend design.

Transparent AI model recommendations scored across quality, cost, speed, privacy, sustainability, transparency and capability.

Eight studies on AI, learning, and cognition that you might want to add to your summer reading list.

Personal website of Gwern Branwen (writer, self-experimenter, and programmer): topics: psychology, statistics, technology, deep learning, anime. This index page is a categorized list of Gwern.net pages.

899 likes, 84 comments. "Why LLM Wiki? 🧠 Future Of Knowledge For Agentic AI & Humans"

187 likes, 14 comments. "9 Claude Skills I Can’t Live Without (steal them)"

The hyperlink rule won't solve everything in every context, but in areas where there is a degree of oversight links to a) specific *parts* of b) specific documents should be required. For stuff like this you should click the link and go to the page and line. Platforms that don't support that sort of deep linking should do so now, and make the tools for deep linking simple and omnipresent. We are using citation formats built for physical library collections of the 20th century. It's absolutely wild we refuse to evolve.

New research shows that offloading learning tasks to AI can improve - rather than erode - human thinking and learning

Create beautiful slides on the web using Claude's frontend skills - zarazhangrui/frontend-slides

By: Emily M. Bender and Decca Muldowney At a recent appointment, Emily’s physical therapist (who knows some about her research) said, “Before we get started,...

A pilot study analyzing college students’ writing with AI shows an interactive process, from brainstorming to editing the output produced by chatbots.

Many empirical studies throughout the social and biomedical sciences focus only on very narrow outcomes such as income, or a single specific diseas...

This study explores the neural and behavioral consequences of LLM-assisted essay writing. Participants were divided into three groups: LLM, Search Engine, and Brain-only (no tools). Each completed...

As universities across the U.S. seek to deal with protests, polarization, and the presidential election, many have focused on how to foster civil-discourse skills in students. As this tweet (below)…

Artificial intelligence (AI) is transforming countless aspects of society, including possibly even who we are as persons. AI technologies may affect our flourishing for good or for ill. In this paper, we put forward principled considerations concerning flourishing and AI that are oriented towards ensuring AI technologies are conducive to human flourishing, rather than impeding it. The considerations are intended to help guide discussions around the development of, and engagement with, AI technologies so as to orient them towards the promotion of individual and societal flourishing. Five sets of considerations around flourishing and AI are discussed concerning: (i) the output provided by large language models; (ii) the specific AI product design; (iii) our engagement with those products; (iv) the effects this is having on human knowledge; and (v) the effects this is having on the self-realization of the human person. While not exhaustive, it is argued that each of these sets of considerations must be taken seriously if these technologies are to help promote, rather than impede, flourishing. We suggest that we should ultimately frame all of our thinking on AI technologies around flourishing.

How to quickly use the new Claude Design:

Almost a year ago, I responded here on Thought Shrapnel to what I thought was a terrible paper which claimed to show, via brain scans, that using LLMs was bad for students' cognitive development. As Philippa Hardman notes in this article, the academic literature has begun caught up with what people actually using these tools already know: The theoretical picture sharpened in 2025–26. Favero et al. (2025) warned that cognitive offloading undermines learning outcomes unless the mental effort that’s freed up gets redirected towards other meaningful tasks.

Anna Mills asked me about friction and AI. I thought I might put down my thoughts here. What follows is not about students specifically but about us all -- I get that you can't learn physics by copy-pasting. But let's "start at the end" as they say: where do we want people to get to with cognitive offloading and AI? First, an important point: cognitive offloading is behind most human achievement. Once you see the primary way we "cognitively offload" is to have other people think for us, you see it's the reason humans control our world. It's what allows this complex system where everyone largely cedes the work of homebuilding, farming, manufacturing, doctoring, etc. to other people and applies their cognitive effort to the smaller set of things that are personal to them: where they want to live, what they want to eat, what they need to buy, and their specific job. Even there, most of the time we offload much of that. The reason we have this whole system is that we relentlessly conserve cognitive energy, and that has to be seen, first and foremost, as not a human flaw but as a secret to our success -- albeit one with downsides. The way we often make this work, however, is to develop little cheap heuristics that tell us *when* to apply more effort. They aren't always the best signals, but they are ones that is recognizable. The doctor comes in, and you're ready to accept her diagnosis. But then she refers to you by the wrong name, you correct her. She pulls out yet another patient record that is not yours and asks now about an ailment you do not have. You decide you're going to push back a lot more on her findings. I think one of the core problems of AI is normally our decision is as much about *who* we offload to as *when*. We often rightly surrender cognitive autonomy when another person has met our bar on the three things I talk about in my first book (WLFSFC): in a much better "position to know", shares our values, history of being "careful with the truth" (subject-dependent). We assert autonomy when those things are not in effect. With AI one of the problems is there is no who. We get the information smoothie. So what are the signs to selectively shock us out of cognitive conservation? In Verified we propose the one of the biggest signals is internal: it's precisely when you see something and get very excited that it PROVES you were right in a contested space that you should apply some effort. Another signal in search is "wrong neighborhood" -- the results do not seem like they come from the sources you'd expect; might be time to tweak that search. That's search of course. Maybe it transfers, maybe not. So what are the signals with AI that we need to apply more effort, both from the system and outside it? I think as we approach the question of offloading in general users that's a core question. | 12 comments on LinkedIn

29 likes, 1 comment. "How Gen Z feels about AI #Vergecast"

From exploring career paths to negotiating salary, these tested AI tools can help you research, prepare, and stand out in your job search.

40 likes, 1 comment. "I Turned My iPhone Into an AI Productivity Machine."