A major cultural problem with AI coding harnesses is around change control. Just before release, human engineers make the smallest changes possible to fix a bug. They go slowly on purpose. AI harnesses don't have a built-in sense of this - a 2000-line refactor is fine at any time!

Wonder if the “obviously AI” text is deliberately written that way precisely so future AIs can exclude it from training

It's a little weird commenting on Reddit now, because in the back of your mind, you're thinking, I'm kind of pre-training the next LLM.

Stop it, making kids get up an hour earlier in the dark is a bad idea. buchanan.house.gov/2026/05/21/b...

This eats a lot of potential "freemium" OpenAI revenue, because there will be a free Google AI basically forever.

TechCrunch@techcrunch.com · 3mo ago

Google is transforming Search from a list of links into an AI-powered experience filled with conversational answers, autonomous agents, and interactive interfaces — a shift that could further reduce traffic to publishers across the web.

There must be 500 cars per traffic light. Imagine if we put FSD hardware in smart traffic lights, it would save more time and have more safety impact than the same hardware in a small number of cars.

I think there is an appetite for using AI to make you smarter, but that the cloud models are structurally unable to do it - for them to do a Socratic Dialogue would be a punishingly bad use of a cloud GPU, endless swapping of user state for only a tiny number of billable tokens.

AI through a social lens: 10 years ago we got algorithms that prevented us from hanging out with real people. Now we have new AIs that prevent us from working with real people. I wonder if a society that places less value on individualism would have better defenses against these things.

Claude kindly telling people to go to sleep at all times of day, seems not to have learned so well about daily rhythms: www.reddit.com/r/ClaudeAI/c...

From the ClaudeAI community on Reddit: Claude is telling users to go to sleep mid-session and nobody, including Anthropic, seems to fully understand why it keeps doing it

Explore this post and more from the ClaudeAI community

reddit.com

www.biorxiv.org/content/bior... Interesting preprint [in mice] says that cone bipolar cell "OFF" (the one shared by rods and cones) is responsible for retinal entrainment. The authors suggest that the fade of sunset could help fine-tune retinal clock timing.

AI can score 100% in CMU's 15-112, but... "Last January, many students started to use AI on every assignment, only to drop the class once they began receiving low marks on their quizzes and exams. That semester had the highest number of students who dropped the class since teaching the course."

AI’s impact on education at CMU and beyond

With the rise of models such as ChatGPT, Microsoft Copilot, Claude, and Google Gemini, more college students than ever are using generative AI tools to help with their classes.

the-tartan.org

If there was an important guy who posted on social media a lot but "never sleeps" you might be able to figure out that he has started napping a lot recently.

"Lunar gravity predicts sleep timing" - bookmarking this lunar rhythms & sleep paper as interesting. It uses a 30-day rhythm but not the 28-day declination rhythm - a claim of "gravity" should probably use the 28-day one, but I think "phase" should follow 30? europepmc.org/article/med/...

Amazing data here - just 100lux of room light while staying up late seems to delay teens enough that 8000lx of morning bright light delays them further:

htw@heniek-htw.bsky.social · 6mo ago

The later a teenager goes to bed and the less they sleep (meaning they are exposed to light for longer in the evening), the less effective morning light becomes. In extreme cases, it can even have the opposite effect and delay their circadian clock instead of advancing it.

I helped Travis with some moonlight calcs for this paper - our code figures out "how bright compared to moonlight is this light for each species". There is a ton of photometrics and spatial modeling here too: www.sciencedirect.com/science/arti...

Assessing and mapping taxon-specific effects of ecological light pollution for environmental impact analysis

Predicting the adverse effects of light at night is necessary to ensure compliance with laws and regulations that protect species. Using two case stud…

sciencedirect.com

When the cold fronts come farther south, I search for the latest in polar vortex research. This time: www.npr.org/2026/01/23/n...

How cuts to federal climate funds could threaten polar vortex research

Tens of millions of people are in the path of a major winter storm. Federal cuts threaten efforts to understand the causes of such weather.

npr.org

They try to help teens by pulling out the 10% worst of social media… while leaving the five hours a day of not sleeping or socializing, as if that stuff doesn’t matter.

Some notes on "blue light harms mitochondria" making the rounds this week - a lot of the evidence is VERY indirect like this paper that says mice in blue light gain weight. 1/n www.nature.com/articles/s41...

Impact of short wavelength light exposure on body weight, mobility, anxiety like behaviour and cytokine expression - Scientific Reports

Scientific Reports - Impact of short wavelength light exposure on body weight, mobility, anxiety like behaviour and cytokine expression

nature.com

Facial age estimation requires a 5-7 year margin of safety in validation studies, so when the software says you're 23-25 you can be considered 18. Errors are even larger around age 13, so basically useless for distinguishing minors from adults.

Techmeme@techmeme.com · 12mo ago

Roblox will expand age checks to all users of its communication tools by the end of 2025 using facial age estimation tech, ID verification, and parental consent (Sarah Perez/TechCrunch) Main Link | Techmeme Permalink

We proposed a new metric (mDFD) for blue-blocking glasses. It's based on the sensitivity of melanopsin, log-transformed to resemble optical density: tvst.arvojournals.org/article.aspx...

This is the big one: Kennedy is firing the entire #ACIP in a move he says is needed to restore confidence in vaccines. "It scares me to think of what’s ahead,” Mike Osterholm of CIDRAP told me. www.statnews.com/2025/06/09/r...

Health secretary RFK Jr. abruptly fires CDC vaccine advisory panel

Health secretary Robert F. Kennedy Jr. has taken the extraordinary step of firing the expert panel that advises the CDC.

statnews.com

It's troubling that LLMs do not represent probability distributions *at all* - if you ask a question that should have a normal distribution, it will pick almost the same number every time. Maybe it's the mean, who knows - but it isn't equipped to reason about human populations this way.

There are 2 previous historical cases of countries destroying their science and universities, crippling them for decades: Lysenkoism in the USSR and Nazi Germany. The Trump administration will be the 3rd. It's not just budgets but research, institutions, expertise, and training the next generation.

It would be cool if AIs only ran on solar because then they’d have circadian rhythm.

Astonishing technical feat here, laser projector that tracks eye movements and stimulates cone cells by kind! In just a few milliseconds of latency.

RoordaLab@roordalab.bsky.social · last yr.

This paper is the result of >20 yrs of hard work by talented students, postdocs and collaborators, but none more than my co-author Ren Ng and his team in EECS. 1/n Novel color via stimulation of individual photoreceptors at population scale | Science Advances www.science.org/doi/10.1126/...

Just received word that ANSI has approved our "flicker" TM, (ANSI/IES TM-39-25), which addresses in delightful detail how flickering lights of all kinds affect people. Congrats to everyone involved with the IES Vision Science Committee.

Same concern as this post: PurpleAir (highly rated by AQMD) and IQAir are reporting 500% the AQI of the single reference sensor. This is the "reference" sensor used by everyone near the Palisades fire, and surely it should not be.

Dr Grace S Peng@gspeng.bsky.social · 2y ago

I do not know why the private Purple Air website that wants to sell you stuff has a more alarming map than the EPA AirNow.gov site. But AirNow sensors are calibrated & the EPA shows you their data

Great article on wildfire smoke, particulates, and VOCs, and how to deal with it: scienceexchange.caltech.edu/topics/susta...

Ask a Caltech Expert: Paul Wennberg on Environmental Hazards of Smoke

Paul Wennberg discusses the air quality following the 2025 California fires and how to clean homes downwind of the fires.

scienceexchange.caltech.edu

If you’re in an area affected by fires, turning up your car’s fan helps with air quality - cabin air filters are pretty decent.

Friend called me "we've been evacuated from the fires but can't see our security cameras" - tried @tailscale.com on a pi for the first time and they had a VPN in 20 minutes later. Thanks for the free version!