Talking to a UX designer today I realized that AI app design is different. instead of focusing on discerning user intent and minimizing cognitive load in business apps, now we let users give vague signals and then count on “the AI” to do something intelligent that effectively becomes their intent.

This is wonderful. I especially like “Don’t be prissy.” There is no problem more sure to go away on its own than an ink smudge on your finger. open.substack.com/pub/nealstep...

Writing by Hand is Good for your Brain - Here's how to do it

Because I am known to write using a fountain pen on paper, a number of people have pointed me to this post and its underlying research. I won’t rehash what is said in those sources, but the gist of it...

open.substack.com

This kind of analysis applies to other fields as well. The move from proof scarcity” to “proof abundance” has analogues everywhere that AI is being used to do hard technical work. The backlog of reformulation for real insight is growing rapidly. teorth.github.io/tao-web/slid...

Troubling, but not at all surprising. I am reminded of Socrates’ warning to Hipppcrates in Protagoras 314b that you can’t “return” cognitive experience to “the vendor” like bad food because you have already been changed by it.

Prof Sam Illingworth@samillingworth.com · 2w ago

🧠 AI is reshaping the people who use it A peer-reviewed paper warns AI in healthcare and the military changes the humans using it: their empathy, responsibility and critical thinking. The tool changes the user. 🔗 doi.org/10.1007/s43681-026-01197-x #SlowAI #AIethics #Deskilling 🧪

New #AI #addon helps developers #automate everyday #programming #tasks Researchers developed Program-as-Weights (PAW), a new method that lets developers turn AI instructions into tiny, local tools, enabling powerful, private & offline AI processing scitechupdates.com/n...

Happy to be a small part of this colossal effort led by @zouharvi.bsky.social to reinvent the future of machine translation evaluation! 🔡 Join us in building the LTB and become a coauthor, help wanted!

Vilém Zouhar@zouharvi.bsky.social · 3w ago

There are many things machine translation still can't do. Help us steer the next direction by contributing hard-to-translate inputs (and be on a cool paper).

Enterprise AI has a data ingestion crisis: unscalable pipelines, and critical document security (ACL) leaks. Launching the Open Ingestion Standard (OIS) and OpenCrawling! www.open4dev.com/journal/2026... #Java25 #RAG #AISecurity #ZeroTrust #VectorDatabase #SpringBoot #MCP #ModelContextProtocol

Beyond the Silos: Announcing the Open Ingestion Standard (OIS) and OpenCrawling — Open4Dev

Introducing the Open Ingestion Standard (OIS) and the reference implementation OpenCrawling.

open4dev.com

Turns out if you have complete and total disregard for human creative effort you eventually steal it from someone with the resources to hit back.

regular meghan 나영지 🇵🇸@ruemcclammyhand.bsky.social · 4w ago

Apple is suing OpenAI techcrunch.com/2026/07/10/a...

I’m a little late with this message having been offgrid for the last week. Happy Birthday, USA! We will never give up on you.

eshumarneedi.com/2026/07/03/z... Nice rant.

Communities around the U.S. are putting on epic fireworks shows for America's 250th birthday — but that spells trouble for dogs and cats. Here's a guide to help the animals in your life.

Tips for taking care of your pets when fireworks are booming

Communities around the U.S. are putting on epic fireworks shows for America's 250th birthday — but that spells trouble for dogs and cats. Here's a guide to help the animals in your life.

n.pr

A coworker gave me a compliment today. She said I was a “great storyteller.” I cringed. If a child told me that I would be happy, but in the workplace…not so much. The whole “story-telling” push among the b-school gang today reeks to me of epistemic decay.

Our ICML mechinterp workshop paper demonstrates how feature geometry can lead to model failures, and analyzing that geometry can help us to efficiently build adversarial test sets based on concept combinations.

Naomi Saphra@nsaphra.bsky.social · 2mo ago

We don’t always know what problems are hard for LLMs. So devs evaluate on tasks HUMANS find hard or on broad benchmarks. What if we could instead anticipate which scenarios a model will fail on—all without evaluating specific input examples? 🧵NEW PAPER by @jenniferlumeng.bsky.social

Promising advance! 🚀🧠🇨🇭 EPFL researchers create an AI model that thinks like we do. More with this article: bit.ly/3QpoXs8

EPFL researchers create an AI model that thinks like we do

An EPFL team has created a new Large Language Model that is structured similarly to a human brain, allowing users more control and moving away from “black box” AI.

bit.ly

I'm so happy when other people write papers on nondeterministic factors in training. embrace the chaos

Emergent Capabilities Arise Randomly from Learning Sparse Attention Patterns

Neural scaling laws for transformer language models predict smooth improvements in pretraining loss with increasing parameters, but downstream capabilities such as in-context learning are known to eme...

arxiv.org

As scientists we should remember that when we talk science with an AI model we teach private companies how to do science and not a new generation of scientists that will contribute to society. Time spent doing faster research with AI agents is not spent mentoring students. Better CV but worth it?

This squares with my experience. The HN comments are delightful. Yes, this absolutely is cargo cult, but at this point the planes are still landing. news.ycombinator.com/item%3Fid%3D...

Not enough people know the expression "Bread and circuses."

We used to have this word, “presidential.”

50% of the time that comparativists come to my office for methods help, they are actually just looking for this paper. What do you do if your "treatment" affected all units, but you expect it to affect some units *more* based on covariates? This. You do this. www.tandfonline.com/doi/abs/10.1...

Factorial Difference-in-Differences*

We formulate factorial difference-in-differences (FDID), a research design that extends canonical difference-in-differences (DID) to settings in which an event affects all units. In many panel data...

tandfonline.com

🧵Feeling safe against data poisoning in post-training? Think again! Individual components of LLM post-training pipelines are surprisingly robust to data poisoning attacks. In work led by Jack Sanderson (co-advised w Yiwei Lu), we show they crumble when attacked together. 1/n

Interesting thread. I like the features vs futures concept. Also the pathologies that result from “single player mode.”

Hillel is on Social Media Break@hillelwayne.com · 2mo ago

Day two of #CraftConf! Gonna try skeeting from Edge in hopes it doesn't kill my battery as much as Firefox. I just got in a couple minutes late for @kentbeck.com 's "Forest & Desert & Genie". About how AI unlocks so much creativity but also has so many challenges

Wrote a bit about the data center backlash, and why it is Good, Actually. The Data Center Bankshot davekarpf.beehiiv.com/p/the-data-c...

The Data Center Bankshot

The

davekarpf.beehiiv.com

What are the real problems to be solved in continual learning? In my latest post, I tackle this question — reviewing where I think the field went astray in the past, how language models changed things, and where the real challenges remain. infinitefaculty.substack.com/p/what-are-t...

What are the real problems of continual learning?

Reflections on catastrophic interference, plasticity, and learning for the future in the era of large language models

infinitefaculty.substack.com

So honored to receive the Glushko prize, and just grateful that @cogscisociety.bsky.social is willing to recognize whatever I’m doing as the study of the “mind” :)

Santa Fe Institute @sfiscience.bsky.social · 2mo ago

SFI Complexity Postdoctoral Fellow Marina Dubova (@mdubova.bsky.social) has received a 2026 Glushko Dissertation Prize from the Cognitive Science Society and the Glushko-Samuelson Foundation. The prize recognizes recent Ph.D. dissertations for groundbreaking work in cognitive science.

“Never-skilling”

Eric Topol@erictopol.bsky.social · 2mo ago

We've written about AI-induced physician deskilling that has already surfaced thelancet.com/journals/lan... @tberzin.bsky.social AI-induced never-skilling among newly trained doctors, while not yet proven, is a serious concern that needs to be addressed nature.com/articles/s41...

Our new paper reformulates tokenisation as a linear program (LP), which we solve to get SOTA tokenisers 😁 As a bonus, this LP tells us how close to optimal any tokeniser is! Check it out 👇 w/ J. Tempus, @philipwitti.bsky.social, @craigschmidt.com, D. Komm Paper: arxiv.org/abs/2605.22821

I totally agree, science is not a simple optimization problem, or at least not in the form conceived by most existing systems. @battleday.bsky.social and I wrote about this recently from a somewhat different perspective (emphasizing the 'problem problem'): gershmanlab.com/pubs/Battled...

Melanie Mitchell@melaniemitchell.bsky.social · 3mo ago

"Science is not simply a sequence of tasks that can be optimized....The scientific system thrives on inefficiency: redundant efforts, failed attempts, and divergent paths. These are not costs to be eliminated but sources of discovery." Excellent essay! ⬇️ www.science.org/doi/10.1126/...

NEW — My first true foray into AI haterdom is about how those of us not convinced by AI’s inevitability are a real constituency worthy of a voice. Give it a read:

Hating AI is good, actually

LinkedIn may be awash with boosters, but shunning AI is the human choice.

thehandbasket.co

The model is so fast and easy to use that I vibe-coded a small game with it in 1h 😅 Runs flawlessly on a consumer GPU if you're looking for a small local model to tinker with.

Nicolas Dufour@nicolasdufour.bsky.social · 3mo ago

Everything is fully open-sourced, including the codebase, the model + all individual single reward model variants! 🌐 Site: nicolas-dufour.github.io/miro 📄 Paper: arxiv.org/abs/2510.25897 🛠️ Git: github.com/nicolas-dufo... 🤗 HF: huggingface.co/nicolas-dufo... 🎨 Demo: huggingface.co/spaces/nicol...