ANTHROPIC'S CEO JUST ADMITTED WHY CLAUDE HAS BEEN SO BAD LATELY
dario amodei at anthropic's developer conference in san francisco said the company saw 80x growth in Q1 on an annualized basis
they planned for 10x. they got 80x
that's why:
> the throttling happened
> the usage limits got nerfed (silently ofc)
> opus 4.7 felt slower
> max 20x customers were getting less than they paid for
> support was replaced with bots
> they had to rent GPUs from spacex
they literally couldn't keep up with how many people started using claude
this also explains the spacex GPU deal. they weren't shopping for compute because they wanted to, they were desperate for it because 80x growth broke their infrastructure
and it explains every single pricing change from the last 2 months
they weren't being greedy, they were drowning in demand they didn't build for
doesn't make the user experience any less frustrating but at least now we know the actual reason
80x growth sounds incredible on a slide deck
for the users it meant 2 months of degraded service while anthropic scrambled to catch up
and the models still feel weak compared to prime opus 4.6
claude PEAKED when opus 4.6 was released. everything since then has been anthropic trying to scale a product that got too popular too fast while the quality fell off
📌Anthropic Accuses Chinese Firms of Stealing Claude AI Data
Source: cybersecuritynews.com/anthropic-clau…
Anthropic today accused three prominent Chinese artificial intelligence companies, DeepSeek, Moonshot AI, and MiniMax, of running coordinated “distillation” campaigns to steal advanced capabilities from its Claude models.
The San Francisco-based lab said the operations involved roughly 24,000 fraudulent accounts and generated more than 16 million exchanges with Claude, in violation of its terms of service and regional access restrictions.
The company said the labs used proxy services and networks of fake accounts dubbed “hydra clusters” to mask their activity and evade detection.
#cybersecuritynews
2025: "AI will replace your job"
2026: "Please just let me write one function without 17 hallucinations and a 400-token apology"
we're not in the future, we're in beta