Lots of folks on here complaining about Fable's response times, but I must admit that what should feel like a bug feels like a feature to me. I like delegating tasks and have some time to think some big-picture thoughts before my AI underling comes back with the results

I have mostly spotted AI slop by common vocabulary (overuse of "epistemic") and syntax ("it's not X; it's Y"). Interesting to see someone who has read a large sample identify indicators in topics/methods/conclusions of papers. Let's one speculate about the future of scientific *ahem* epistemology

Neighbor after hearing why I'm leaving instead of doing the green card thing: So, it's absolutely possible you'd have a uneventful and harmless path to permanent residency Me: Yes, but it's also absolutely possible to play Russian roulette 3 times and be fine and I'm not into that either

Anthropic to user: "AI can make mistakes. Please double-check responses." User to Claude: "Please always double-check your responses."

When working in govt policy, it comes with the job that most days you don't get what you want. (Your ideas the crushed, your proposals frankensteined, ... .) Rarely does it feel as intense as today though

Go science!

Add me on LinkedIn! That way my AI agent can like and comment the posts of your AI agent and vice versa. You and I will never have to interact again

peak Pygmalion displacement especially notice 1 to 5 in the table below (pdf here: doi.org/10.31235/osf...)

Dr. Velislava Hillman and Dr. Peter G. Kirchschlaeger argue that as AI erodes job markets, it exposes the reality that the crisis facing young people "was never primarily technological, but political." Time to rethink what education is for, they say.

What's a good tune for singing the words "late-stage capitalism" over and over again?

I still find it mildly amusing to start an argument with Claude in one tab, copy entire chatlog into a new tab, and then gossip with Claude instance #2 about how Claude instance #1 is full of sh*t.

Don't we all love a little "rm -rf *"? Do you feel your fingertips tingle when you think about typing it out?

I set out to understand why AIs love negative parallelism. The answer wasn't just surprising—it was illuminating. www.theatlantic.com/technology/2...

A convo that lives rent-free in my head and comes up whenever I fry bacon was with a UK woman who said "I've never slept with an American but I don't think I am missing much. I like them thick, but I've had their thick-cut bacon and I am pretty certain no one over there knows what thick even means."

Turns out my cat is getting fat bc she is depressed, which is probably connected to me working late hours instead of cuddling with her. So y'all have to stop having stupid ideas about AI in healthcare and planting them in the heads if Utah decision makers or my cats osteoporosis to answer for

Today marks a first in that I have found a mildly entertaining post on LinkedIn

Seeing these two posts back to back in my feed capturing two domains of automation resilience: 1. Where the value of an activity comprises not only the output but also a sense of human connection, 2. Where part of the value lies in the output showcasing human artistry/ craftsmanship

I think this same thing applies to math and (T)CS - as with much of the rest of the essay. (Literally change the one word "philosophy" to "mathematics")

I seem to have missed this—federal govt gatekeeping who gets to use latest #AI tools? Many reasons to regulate AI, but this way seems bad. @whitehouse-47.bsky.social pushes back against states + Congress reg but then goes ahead and regulates themselves? 🤔 www.nytimes.com/2026/07/10/o...

Excellent paper from @catsaintcroix.bsky.social

Ring of integers? You mean a clock?

Work in the AI economy is deliberately organized through an "outsourcing maze," writes the Weizenbaum Institute’s Sana Ahmad, enabling global capital to extract value from workers via layered intermediaries.

One thing that I am going to miss about living in an English speaking country is how severely you can insult people using very few words. A good German insult either includes profanities or a rant the length of an essay

AGI: An LLM gets my toddler to complete her bedtime routine ASI: An LLM gets my toddler to complete her bedtime routine and fall asleep by 9pm

Prompt engineering in 2026:

I would like to see a benchmark that is just 1000 people across all industries leaving going on vacation for a week and leaving their jobs to LLMs. Benchmark performance is a measured as the inverse of how much their bosses and coworkers hate the vacationers by the end of the week

My toddler is now teaching me the subtleties of speaking American English that I haven't learned in a decade. Like that the word "pants" has actually two syllables: "pay" and "ahnts"

Humor’s secret sauce: Coherent incongruency. 🧵1/4 New research with @drew-gorenz.bsky.social #SocialPsych #CogSci #Humor J Exp Soc Psych; no paywall until Aug 27 authors.elsevier.com/a/1nOzR51f8%...

More clarity for aspiring #entrepreneurs: Is the new German Spin-off Framework (GSOF) the long-awaited #startup booster for deep-tech innovations? Five questions for Bram Wijlands, Managing Director of Max Planck Innovation GmbH. #TechnologyTransfer #Innovation

Mayo Clinic’s former director of research operations alleges she was sidelined, then terminated, for raising objections over lax standards of studies involving AI.

I haven't even looked closely at the study design and directed the methods yet but I am already convinced that this paper's title was badly chosen Large language models can predict the results of social science experiments | Nature share.google/xaQJLA1JqD9s...

Curious. Do you trust LLMs more on their intuition, their facts, or their ability to do the work?

worthwhile set of anecdotes about the tech job market

We have reached the infallible no-one-understands-the-model-it-therefore-must-be-smart point in the discussion

On a related note, what is the normal amount of old chicken nuggets I should expect to find stuck with a pillow in its pillow case?

"You fake it til you make it" but said really depressingly

Schindler's list would be such a terrible movie to watch if Schindler had been an effective altruist

For mere mortals who are not trying to hack nation states, what are Fable use cases that actually warrant the extra investment? I have been comparing Opus and Fable output on a few research tasks. Both are doing just fine

Plausible, but I would take the headline more seriously if there was more research in the article. Here, the only piece of evidence is that anonymous financier's quote

How do neural circuits allow for #sensory processing to change with context? This study by @nuttidanuttida.bsky.social &co reveals how inhibitory microcircuits support flexible, context-specific integration, supporting adaptive decision-making 🧪 #AcademicSky #neuroscience plos.io/4vL2Qf9

If you have sent me an email today, please know that it's a holiday over here, daycare is closed, it is almost midnight, and my daughter is having her 8th half-hour tantrum of the day

Anybody who remembers how Blair Witch Project was advertised - we are all now waiting for some movie with a lot of shaky handcamera shots, right?

5% is for cowards. Make it 51

Lennart Finke and I will release a paper on Monday about how some AI developers tend to make models that differentially downplay company controversies. Below (🧵) is a link to a 1-question Google form for you to guess the results before they're out. (They might surprise you.)

Impressive blog post. I'd only add that in may cases, "Statistical control requires causal justification" (doi.org/10.1177/2515...).

Describe the economy in a single sentence

I knew the day would come where my daughter's bedtime would be later than mine. I just didn't know we get there before her 3rd birthday. It is 10pm. I am going to bed. One room over, my daughter is playing with Legos and singing to herself. This is probably going to continue for another two hours