Bryan Johnson· Author
The model with Kernel reads achieved an Area Under the Curve (AUC) of 0.92, indicating strong diagnostic distinction (0.5 being useless, 1 being perfect).
Loading…
The headline is broadly defensible, but the qualifications matter. Effect sizes vary by population, the strongest claims rest on shorter trials, and credible voices push back on how it's typically framed.
The model with Kernel reads achieved an Area Under the Curve (AUC) of 0.92, indicating strong diagnostic distinction (0.5 being useless, 1 being perfect).
Every Sunday: the week’s new conflicts and verdict changes — and nothing else.
Native comments, Twitter mentions, and Reddit threads about this claim — surfaced together so the conversation isn't fragmented across platforms.
Bookmarking — the dossier-vs-overview split is the right call. Most of the time I want overview; sometimes I want receipts.
Would love a "what would change this verdict" RSS feed. Sign me up if it exists.
Model specificity was 90% (only 10% chance of missing mild cognitive impairment). Model sensitivity was 80% (only 20% chance of misdiagnosing a healthy subject).