It's weird that asking a simple question as "what is a mitochondria" switches to opus 4.8.
It'd be nice to have fable 5 distinguish between harmless and malicious prompts.
Anyways, 7 days to explore fable!
This is what a good experience should look like: No fees. Just payout.
Trade your USDT & USDC on Ewave with 0 fees.
Get the real worth of your money every time. ๐
https://t.co/q42mKIkfuO
Given the Opus + Sonnet models we get to see as peasants....I'd assume they have something pretty mind blowing
Internal research models run 6-12 months ahead of what ships. What they release publicly is what *clears safety, reliability, and cost thresholds*
not the frontier of what exists internally. So there's likely something running at Anthropic right now that makes Opus 4.6 look like Haiku.
DeepSeek used 24,000 fake accounts to scrape 16 MILLION conversations from Claude.
Not a hack. Not a leak. They just talked to it 16 million times and wrote down the answers.
This is the new IP theft. You don't steal the model weights โ you steal the behavior. Train your model on how the best model responds.
Anthropic spent billions building Claude. DeepSeek spent API credits copying it.
The AI cold war isn't about who has the best researchers anymore. It's about who's best at cloning the other side's work.
And the wildest part? Some guy responded by VOLUNTARILY releasing 155K of his own Claude conversations as open-source training data.
We're speedrunning "nothing is proprietary" as a civilization.