Reflection on @dwarkesh_sp
's reflections on his interview with Rich Sutton, and why LLMs are exciting because they are Skinnerian, Popperian and Gregorian creatures. Minds with finite capacity cannot adapt forever without having to forget previous knowledge. This is true of
RESEARCH
-
LLMs as Skinnerian, Popperian, and Gregorian Creatures: Cognitive Limitations
By
–
-
World Model turns your house into a Holodeck
By
–
This is how a World Model turns your house into a Holodeck. https://t.co/veLd456wIH
— Robert Scoble (@Scobleizer) 12 décembre 2025This is how a World Model turns your house into a Holodeck.
-
SoTA Model Benchmarks Reveal Evaluation Methodology Flaws
By
–
The evals that folks care about and publish at any time are those that SoTA models are just shy of being good at. So this is what such pics *always* look like.
-
Lightweight Discriminative Models: Overlooked ML Approach
By
–
Somehow I entirely missed the discriminative models that are lightweight, easy to understand, train, and infer.
-
Odyssey-2 Pro: Intelligent Interactive Real-Time World Model Available Soon
By
–
https://t.co/IrMUUAtjfg https://t.co/8fazB7BxN6
— Robert Scoble (@Scobleizer) 12 décembre 2025Say hi to Odyssey-2 Pro—our most intelligent, interactive, and general-purpose world model yet. Every pixel is generated in real-time, and frame-by-frame the model simulates the world forward. Today for minutes, soon for hours. Available to developers and our API very soon.
-
Gradual Tech Improvement Expected Until Field Maturity
By
–
Ongoing gradual minor improvement in product capabilities is what you'd expect in pretty much any tech field at pretty much any time, until that field is extremely mature and stable.
-
GPT-5.2 Thinking vs Pro Model Review Comparison
By
–
My review for GPT-5.2 Thinking (not Pro) was not as positive. Pro, however, is an amazing model, which is why I say so.
-

Anthropic Opens Fellows Program Applications for Safety Research
By
–
We’re opening applications for the next two rounds of the Anthropic Fellows Program, beginning in May and July 2026. We provide funding, compute, and direct mentorship to researchers and engineers to work on real safety and security projects for four months.
-
Anthropic Expands Fellows Program with Strong Research Outcomes
By
–
40% of fellows in our first cohort have since joined Anthropic full-time, and 80% published their work as a paper. Next year, we’re expanding the program to more fellows and more research areas. To learn more about what our fellows work on: https://
alignment.anthropic.com/2025/anthropic
-fellows-program-2026/
… -

GPT-5.2 aces AIME math without tools
By
–
GPT-5.2 Thinking scored 100% on AIME (math without tools) Pretty good progress