The PM-to-engineer ratio at Mercor is rising not because of more PMs but because AI has removed the headcount bottleneck on engineering output, shifting the constraint to product judgment.

The fastest-growing data type at Mercor is environments, meaning interactive simulations used for training AI agents.

Foundation models trained on descriptive single-cell data do not yet outperform simple linear models on causal or perturbational prediction tasks, a finding observed across multiple groups in the field.

Baron Capital's Mississippi data center investment is generating approximately $900 million per month from Google.

It takes approximately three supervised deals of hand-holding before an AI agent can be trusted to handle finance tasks autonomously.

Mercor already spends more on AI tokens than on developer salaries.

Pape draws on 25 years of studying suicide attacks to identify threatened or actual ground control of homeland territory as the single biggest risk factor for suicide terrorism.

Some of the biggest companies today were originally wrappers around SQL or S3, undermining the dismissive framing of AI wrapper startups.

Grok (xAI) is worth watching as a fast-closing competitor to Anthropic on both price (65% cheaper) and capability, with Baron predicting parity within six months.

“We were the fastest robot from zero to launch in three years and we've only spent about 15 million total.”

“We're spending half a trillion dollars more than we're taking in.”

“We've not been able to discover a single psychiatric condition in which sleep is normal.”

“From Bitcoin's code perspective, it didn't bend the knee. The institutions bent the knee to bring Bitcoin into it.”

“If you live in a state where there is an arbitrary power who can have you put to death, you are a slave.”

“I would say something like 70% chance that this goes horribly wrong like human extinction.”