autotraining models with markdown

The idea: give an AI agent a small but real LLM training setup and let it experiment autonomously overnight. It modifies the code, trains for 5 minutes, checks if the result improved, keeps or discards, and repeats. You wake up in the morning to a log of experiments and (hopefully) a better model. The training code here is a simplified single-GPU implementation of nanochat. The core idea is that you're not touching any of the Python files like you normally would as a researcher. Instead, you are programming the program.md Markdown files that provide context to the AI agents and set up your autonomous research org. The default program.md in this repo is intentionally kept as a bare bones baseline, though it's obvious how one would iterate on it over time to find the "research org code" that achieves the fastest research progress, how you'd add more agents to the mix, etc. A bit more context on this project is here in this tweet.

Coding After Coders: Summary

The New Reality of AI-Assisted Programming

• Elite software developers now rarely write code themselves — instead, they direct AI agents in plain English
• Tools like Claude Code deploy multiple agents simultaneously: one writes, one tests, one supervises
• Tasks that once took days now take under an hour

The Strange New Workflow

• Developers spend their days describing intent to AI, reviewing the AI's "plan," then letting agents execute
• When agents misbehave, developers have resorted to scolding, pleading, ALL-CAPS commands, and emotionally charged language ("embarrassing," "national security imperative") — and it seems to work
• Prompt files have become records of hard-won rules to constrain unpredictable AI behavior

Economic Stakes

• Coding was once considered near-guaranteed, high-paying employment ($200K+)
• It may be the first expensive white-collar skill AI can fully replace — unlike AI video or legal briefs, AI-generated code that passes tests is indistinguishable in value from human-written code
• Irony noted: Silicon Valley workers, who told others to "learn to code," got automated first

Developer Sentiment: Mostly Euphoric

• Most developers interviewed were energized, not demoralized — reporting 10x to 100x productivity gains
• Key insight from tech executive Anil Dash: unlike creative fields where AI removes the soulful work and leaves drudgery, in coding AI removes the drudgery and leaves the soulful parts

Historical Context: A Long Arc of Abstraction

• Each programming era simplified the one before: Assembly → high-level languages (Python) → open-source packages → now natural language intent
• AI represents the highest abstraction layer yet: developers no longer need to manage syntax, memory, or debugging minutiae
• The open question, now being asked at Anthropic itself: what is coding, fundamentally, when the code-writing is gone?

Run, manage, and scale AI workloads on any AI infrastructure. Use one system to access & manage all AI compute (Kubernetes, 20+ clouds, or on-prem).

Run by university