Wits and scribbles
Some thoughts, almost worth reading.Sententia
Conjectures and refutations
The limits of my prompt
February 2026Post status: incomplete thinking, workshopping timeboxed writing. In plain words: What technical interviews might look like in a few years. Writing style: Weird math formalism interspersed w...
Tokens for nothing and your (cognitive) chips for free
May 2025Post status: ad-hoc reaction to X post. In plain words: Thinking before doing looks like cheating. Writing style: GPT-curated blockquotes, then marginalia on Krakauer.
Archive
Marginalia
Demanded post-scriptum.
- Conformal Classification of Neural Networks
- Back to Square One: Artifact Detection, Training and Commonsense Disentanglement in the Winograd Schema
- Probing Emergent Semantics in Predictive Agents via Question Answering
- KG-based world models
- Simple Few-Shot Learning with Language Models
- Learning to communicate about shared procedural abstractions
- Reward is enough
- Making Sense of Sensory Input; Apperception code
- Are Convolutional NNs or Transformers more like Human Vision?
- Flexible modulation of sequence generation in the entorhinal–hippocampal system
- 26 more →
Symposia
Compelled a record.
- Learning salon: Kimberly Stachenfeld
- Learning salon: Jane Wang
- Learning Salon: Recap & Discussion
- Learning salon: Zenna Tavares
- Ideas about a new programming language for games
- Learning Salon: Simon Kornblith
- Learning Salon: Blake Richards
- Learning Salon: Katja Hofmann, Deep RL for Games
- Learning Salon: Recap and Panel Discussion