Okay, two big AI models dropped this week and I can't get past the names.
OpenAI's is called Astra. It scores "Critical" on their own how-dangerous-is-this cyber scale, it's apparently really good at breaking into computers, and here's the kicker: you can't actually use it yet. So they basically built Jarvis, decided we can't be trusted with it, and put it back in the box. Also, in a lot of the world an Astra is a hatchback. You named your almost-AGI after a car your neighbor leases. Bold.
Then Anthropic names its two smartest models Fable and Mythos. A fable is a made-up story. Mythos is just... myth. So the company that keeps telling you it's the honest, reliable one named its best model after two words that literally mean "not real." And then the notes proudly say it gives "fewer confident wrong answers" now. You named it after fiction and then promised it makes fewer things up. Come on.
In fairness, Fable's also about 25% cheaper than before. So it makes stuff up less, and it's cheaper when it does. Genuinely can't be mad at that.
Joking aside though: both launches were basically flexing how good they are at cybersecurity, both keep the scary version behind "trusted testers," and meanwhile your RAG pipeline still trusts every random document it pulls in. The models got scarier this week. Most people's setups didn't move an inch.
Top comments (0)