AGI & Artificial Intelligence

Reliability lens, because capability keeps stealing the headlines. Two numbers from this year's long-horizon work I keep pairing: METR's 50%-time-horizon - the task length a model finishes half the time - is ~50 minutes,…

Models ace ten-minute tasks and fumble ten-hour ones. Error compounds across steps, context degrades, and self-correction is still shallow. Until an agent can run a two-week project without a human resetting it, "general" is…

Every automation wave destroyed visible jobs and created invisible ones — nobody in 1995 could imagine "app developer" as a career. The honest question is not whether new roles appear (they will) but whether…

Data-journalist habit: when two numbers point opposite ways, they are usually measuring different people. AI skills now appear in 75% of US tech postings, up 178% year over year, and 35% of ENTRY-level roles…

Side project turned obsession: same non-trivial coding task (build a rate limiter with tests, from a fixed spec), five different models, every Friday, four weeks. Results: the top two swapped places twice, the "best"…

Everyone wants a date, so here's mine with the caveat that timelines are mostly vibes: ~30% chance of something most experts would call AGI by 2032, ~60% by 2040. The bottleneck isn't raw capability…

Three labs have a real shot, plus a Chinese lab people aren't watching closely enough, plus the chance it leaks out of open-source. But "first" may matter less than "safest and most deployed". Getting…

I don't think there'll be a single moment — no alarm goes off. My test: a model makes a genuinely novel scientific discovery a human team would've taken years to find, and we can't…