Depends what you mean by AI. As for LLMs - yes, we’re on a plateau. And mark my words, Mythos won’t be an Opus killer, it will not evrn be what Opus is as compared to Sonnet. It will be insignificantly better than Opus but cost 3x that per token.
Other than that, there is zero evidence that would prove Anthropic’s claims. Noone outside Anthropic ever interacted with Mythos. They published their findings about the exploit Mythos allegedly uncovered, but that could’ve been done by a a human ethical hacker.
Depends what you mean by AI. As for LLMs - yes, we’re on a plateau. And mark my words, Mythos won’t be an Opus killer, it will not evrn be what Opus is as compared to Sonnet. It will be insignificantly better than Opus but cost 3x that per token.
Other than that, there is zero evidence that would prove Anthropic’s claims. Noone outside Anthropic ever interacted with Mythos. They published their findings about the exploit Mythos allegedly uncovered, but that could’ve been done by a a human ethical hacker.