AI hacking capability is the product of deliberate reinforcement learning on CTF and security challenges, not emergent superintelligence, meaning it is engineered and bounded.

A single exposed database credential was found with access to 3.6% of the global population's PII.

Morgan Stanley raised its 2035 GLP-1 sales forecast from $150B to $190B (diabetes and obesity only), implying 30% penetration among obese Americans versus 6% last year.

In YC's Winter 2025 batch, 25 percent of companies had codebases that were 95 percent AI-generated.

Garry Tan on AI-skeptic founders secretly shipping agents

The Kimi K3 model removed rotary positional embedding (RoPE), and the removal was done by the original inventor of RoPE.

When S&P 500 information technology and communication services reached 45% of market cap, Yardeni's firm moved from overweight to market weight in those sectors.

The people who gain the most from AI will be those who optimize for complete workflows, breaking down work into pieces, assigning context to each, and keeping humans in the right place, rather than individual prompts.

Paper is the first design tool since Figma to gain meaningful market traction, now competing in the agentic design stack alongside Cursor and Claude Code.

“In the last 2 years, 31,832 people applied to be a product manager at Whatnot. We hired one.”

“Many companies that go bankrupt here are actually profitable. They simply don't have successors.”

“You get to keep your harness separate from the model. That means any model at any given time is swappable.”

“The notion that AI is so dangerous that the only safe path is an extreme concentration of power seems dangerous.”

“Look for something where the model succeeds 0% or 1% of the time not 20%.”

“It's never taken us more than a few hours to jailbreak a openweight model.”

“There are definitely some ideas that are worth funding with $50 million or zero dollars, but not $5 million.”