High-quality human supervision

This week: Unsupervised elicitation, open tokens, the illusion of thinking, gravity

What do people want?

This week: what do people want, humanity’s last exam, thinking beyond silicon, a bottomless pit of plagiarism

A2A

This week: A2A, Who is AI replacing, prompted honesty, human-llm grounding

Denoise this noisy target

This week: NoProp, Sudoku-Bench , Virtual reality arena, coding is not software engineering

A deceptively persuasive title

This week: Counting, Claude, persuasion, deceptive persuasion

Two views of the causal chain

AI as normal tech, predictive optimization, continuous thought machines

Alignment with human preferences

This week: Evaluation, measurement, span annotation, openai evolves, hegemons

Algorithmic Glazing

This week: llm finetuning, algorithmic management, perceptual possibilities, ChatGPT won’t stop glazing

Not for private gain

This week: Reinforcement Learning with Verifiable Rewards, naming eras, model welfare, investors welfare