Claude just BROKE the ENTIRE INDUSTRY...

Terminator: Die huma…bbzzzzw Claude: You have reached your token limit

How soon are they going to be releasing the source code ?

Remember when we used to joke about being in the permanent underclass? Good times.

if I can't use it, it's useless

So, what, we've now entered an age where the public won't be seeing these increases in AI intelligence anymore? We're being cutoff for our own "own good" while a few mega corps get the real AI? I guess the skeptics were right, if superintelligence were ever to be released, it's not the public who would be the ones benefitting.

The real worry is a company like Anthropic having this sort of power. Or any company, really.

... And this is why super intelligence will never be contained. Better make sure it's really keen on humanity. 🤞

This won't be released until a competitor releases an equally powerful variation... I give it until May at the latest!

DETAILS & LINKS: https://natural20.beehiiv.com/p/anthropic-says-its-new-ai-model-is-too-dangerous-to-release

Improving from 80 to 90% on a benchmark isn't a 10% improvement, it's an over 100% improvement. Test scores reward accuracy logarithmically. 80% on a benchmark represents 4 correct answers per 1 failure. Which is a 4:1 ratio of correctness. 90% on a benchmark represents 9 correct answers per 1 failure. Which is a 9:1 ratio of correctness. That 10% score increase represents a lot more than a 10% improvement.