AI at Work · Randomized trial
A university gave 2,379 students a free GPT-4o tutor inside their course platform, and randomized who got it. Final grades fell by about 4 points. The students who never opened the tool lost ground too. Here is what the paper actually shows, and what it doesn't.
Body & Longevity · Phase 3 result
Novartis lowered Lp(a) in 8,323 patients. Heart attacks and strokes did not fall. And for now, the biggest trial in the field exists only as a press release.
Evidence Guides
Six checks that separate what a study measured from what the headline claims. The same checklist I run before I write a word.
2 articles
1 article
2 articles
2 articles
1 article
2 articles
1 article
1 article
One email a week: the 5 facts that mattered in AI, health and longevity, each checked against the original source. No spam. Unsubscribe in one click.
By subscribing you agree to receive the newsletter. See the privacy policy.
A randomized trial of 2,379 students. The tutor was free, patient, and built on GPT-4o. Grades fell anyway.
Hinton, Bengio and about 400 colleagues are asking for an international body to oversee the most capable AI systems. A public call, not yet a policy.
Meta's new flagship scores 56 on an independent index, two points behind Claude Opus 5.5. And the stronger model is not the one you can call today.
Pelacarsen lowered the biomarker in 8,323 patients. Events did not follow. What that gap teaches about every longevity claim.
A Yale reanalysis of 12 epigenetic clocks says single clock drops don't reproduce. Then come the claims: GLP-1s, a skin cream, 2 to 3 years younger.
ACHIEVE-3, 1,698 adults, 52 weeks: HbA1c down 2.2 points against 1.4, weight down 9.2 percent against 5.3. A real head to head, run by one of the duelists.
Twelve weeks of supervised lifting: the vegetarian group gained 0.38 kg of arm and leg muscle, with no detected strength gap against omnivores.
In patients with obesity, office systolic fell 16.0 mmHg in four weeks on the endothelin blocker. A subgroup signal, on a short clock.
Sironax's FORGE-LM trial is half enrolled. The endpoint is body composition, which is the right question. Enrollment, though, is not a result.
809 adults, seven days of wrist sensors: weaker, less stable light and temperature rhythms track with obesity. Association, not a verdict.
Six checks: design, endpoint, absolute effect, population, status and funding, limits. The house method, in the open.
Every story here starts from the paper, the filing or the data, never from the launch post. Claims are labeled by evidence level, and the labels are on the page.
Start with the guide, then audit any article on the site with its own checklist. The sources sit at the bottom of every page, where they belong.
