Reminder that the last 12 months are plausibly the least crazy months of AI for the rest of your life!
It's only getting crazier from here ...
Marius Hobbhahn, who leads Apollo Research in testing frontier models for deceptive or scheming tendencies, posted that the past year’s AI developments may represent a relative lull compared with the acceleration he expects ahead, though the claim arrives without attached benchmarks or timelines.
Reminder that the last 12 months are plausibly the least crazy months of AI for the rest of your life!
It's only getting crazier from here ...
Hobbhahn’s post supplies no datasets, capability thresholds, or dates, so the concrete changes he foresees remain an open matter even as his background in model evaluations lends the statement weight.
The group has reported roughly 30-fold drops in certain scheming behaviors during joint tests with OpenAI, yet the June 2026 statement does not connect those results to the broader progress forecast.
Reminder that the last 12 months are plausibly the least crazy months of AI for the rest of your life!
It's only getting crazier from here ...