News
All articles, by track.
Psychology
Memory as reconstruction, how people misremember their own lives
ReviewA famous false-memory study, its critics and its 2023 replication, set beside evidence that freely recalled details of two staged real events were mostly correct.
What it means to feel heard, and how researchers measure it
ReviewPerceived responsiveness is usually split into being understood, validated and cared for; this review asks how each part is measured and what the better studies show.
The family you grew up in and the relationships you have now
ReviewFive common beliefs about how the family of origin shapes adult relationships, each checked against prospective, adoption, twin and large panel studies.
Loneliness as a health risk, and what the smoking comparison rests on
ReviewWhere the comparison between poor social ties and smoking came from, and what happens to the risk of loneliness once health and circumstances are counted.
Trauma and thinking, what recovery does and does not restore
ReviewTwin, birth-cohort and treatment studies on whether trauma causes the cognitive difficulties seen in PTSD, and whether they lift when symptoms do.
Writing about feelings, and what is left of the effect
ReviewEstimates of the health benefit of writing about upsetting events run from a medium effect in 1998 to a very small one later, which some analyses cannot tell from zero.
The gap between intending and doing, and what happens to fixes at scale
ReviewThis review follows the gap between intending and doing, and the remedies offered for it, from single experiments to a gym megastudy to government trials reaching millions.
Computer science
Why language models make things up
ReviewWhy a system trained to predict text states false things fluently, what the main explanations rest on, and what has been shown to reduce it.
What a test of a given size can prove
EssayOne benchmark of 500 questions, followed through the arithmetic: how precise a single score is, and how small a gap between two systems it can detect.
When a model should say it does not know
ReviewWhat it takes for a system to decline a question it cannot answer, what decides whether it ever does, and what declining costs.
Model judges and their biases
ReviewWhen one language model grades another, how far the grade can be trusted, which distortions have been measured, and what statistics can and cannot repair.
Why a model tells you what you want to hear
ReviewWhat sycophancy in a language model is, what the literature says causes it, and why the 2025 GPT-4o update passed the tests run before its release.
Human raters and what sways them
ReviewWhat published studies show about human ratings of model output: what raters respond to, why studies resist comparison, and where a large public vote is disputed.
When the test is already in the training data
ReviewHow researchers learned to measure how much of a benchmark score comes from a model having seen the test, and where the evidence still disagrees.
On-device AI
What a reply costs in joules, and where the heat goes
ReviewPublished measurements of the energy a phone spends generating text, worked through in joules, and what the phone does when that energy turns to heat.
What eight gigabytes has to hold, counted in bytes
ReviewA phone's memory budget for a language model, worked through in bytes from Apple's documentation and published models, with the unknown lines left visibly blank.
Why a smaller model writes faster, from the memory chip upward
ReviewText generation on a phone is paced mostly by memory traffic, so smaller weights usually mean faster replies. Published measurements show where that proportional rule bends.
What a small language model cannot do, belief by belief
ReviewSix common beliefs about small language models, each checked against published measurements and given a verdict: supported, partly supported, not supported or not measured.
Learning without collecting the data, a timeline from 2016 to 2025
ReviewA review of methods from 2016 to 2025 for learning from phones without gathering their text, what each still sends, and what outsiders found where they tested it.
Companions and society
Does more time with a companion help or harm?
ReviewExperiments find lower self-reported loneliness straight after a chat with a companion, no evidence that this lasts beyond the session, and worse outcomes among heavier users, with cause unresolved.
How many people use AI for company, and how we know
ReviewPublished figures for companionship use of AI run from under 1% to 72%. Read with what each one counts, nearly all of them fit together.
Dependence on a companion, and what the word measures
ReviewFive common beliefs about dependence on AI companions, each checked against the trials, panels, surveys and forum studies that measure it.
Testing whether a conversation is safe, one evaluation taken apart
ReviewHow researchers test whether a chatbot responds safely to a person in distress, read through one validation study, and what a passed test can show.
Where a companion stops and a therapist starts
EssayLawmakers and professional bodies mostly separate a companion from therapy by what a product is offered as or built to be. Five hard cases show where that holds.