·7 min read
Grade an AI agent by its decisions, not its wording
An October 2026 paper trains small agents to match a big model's decisions instead of its words. Decision models point the same way. What it means if you run agents.
- ·16 min read
An AI agent's notes can shape what it does
Two unrelated October 2026 papers, read side by side. In agents that keep notes, a wrong note may persist. What they show, what they don't, and our rule.
- ·7 min read
Personas are a landscape, not a straight line. The gain grows with the bend.
A preprint says AI persona activations sit on a curved surface, and following it beats a straight line where the surface bends most. Authors' numbers, three small models.
- ·6 min read
The harness is turning into a skill the model can learn. The numbers are early.
A September preprint lets a model edit its own context as a file. The gains are the authors' own, mostly zero-shot, and not yet replicated. Here is what holds.
- ·8 min read
Google's ScientistTwo was graded by two AI reviewers. Only one of them was held out, and that is the whole result.
Under the reviewer inside its own refinement loop, ScientistTwo clears the human bar. Under the held-out one, its lead depends on which papers you count.
- ·12 min read
A new robotics paper is titled around topology. Its own appendix shows the topology is not what finds the doorways.
A cited read of a three-day-old preprint on transferable robot subgoals: a real idea, an honest baseline, an abstract figure 3.3x the result, and a control that moves the credit.
- ·7 min read
A new AI paper claims a 2.29x efficiency gain. The real number is closer to 1.21x.
A cited look at a new AI reasoning-compression paper: a real method, verified numbers, and one headline figure that bundles two separate effects into one.