Stories by thejash

What's needed for AI agents to work

Decentralization or Dystopia: Pick 1

Show HN: Sculptor – A UI for Claude Code

Sculptor: Catch and fix issues as you code

Carbs, a hyperparameter optimizer that scales small experiments to LLMs

Open-source, sanitized evaluation datasets for models that reason and code

Infrastructure setup and open-source scripts to train 70B model from bare metal

Scaling Laws for Every Hyperparameter via Cost-Aware HPO

Jim Fan, Nvidia: On foundation models for embodied agents and scaling data

Noam Brown, FAIR: On achieving human-level performance in Diplomacy, and more

Nicklas Hansen, UCSD: On long-horizon planning and much more

Jack Parker-Holder, DeepMind: On open-endedness, offline learning, and much more

Chelsea Finn, Stanford: On the biggest bottlenecks in robotics and RL

Hattie Zhou, Mila, on supermasks, iterative learning, and fortuitous forgetting

Generative AI startups we want to fund

Minqi Jiang, UCL, on environment and curriculum design for general RL agents

Oleh Rybkin, UPenn, on exploration and planning with world models

VICReg: Tutorial and Lightweight PyTorch Implementation blog post