mamma mia is a masterpiece of a film for many reasons but turning “does your mother know” into a cougar anthem by christine baranski instead of a song from a man’s POV is actually so genius
This is the biggest story in AI this year.
Karpathy is one of the very few people on Earth who can bridge LLM theory and actual large-scale training. He was OpenAI founding team, then ran Tesla's self-driving AI, then back to OpenAI for the synthetic data and mid-training work.
He has now joined Anthropic's pre-training team under Nick Joseph.
Pre-training is the deepest, most expensive, most compute-heavy part of building a model. It's where Claude gets its raw intelligence. A person of Karpathy's talent will definitely make the core engine even better.
My take on why this could have happened;
One.
> The AI race has shifted from compute to research efficiency.
> For three years the assumption was "whoever has the most GPUs wins." That's breaking.
> His recent autoresearch project, an AI agent that runs ML experiments on a single GPU, got 66,000 GitHub stars in a month. Fortune called it the Karpathy Loop.
> Anthropic reportedly wants him building a team that uses Claude to speed up Claude's own research.
So AI improving AI is the frontier now, and he's the perfect hire for it.
Two.
> He chose Anthropic over going back to OpenAI. He'd already returned to OpenAI once and left again. Picking Anthropic instead is a signal about where he thinks the most interesting work is happening.
> Talent is the truest market signal in AI. Money is everywhere. Karpathy's time is not.
Three.
> He paused Eureka Labs, his own education startup, to do this.
> A founder doesn't shelve his own company unless he believes the next two years of LLM research are a genuinely rare window.
> In his own words, people walk away from their startups only when they think they'll miss something that won't come again.
Now, OpenAI vs Anthropic vs Google DeepMind was already tight. This tilts the talent narrative toward Anthropic at exactly the moment the game is moving from raw scale to research cleverness.