Simon newslttrs.com

@trademark @Lemonid @GossiTheDog pretty sure I'm not the one arguing

@trademark @Lemonid @GossiTheDog it’s only marginally above the line of best fit meaning it’s at best an incremental improvement over previous models. Hardly a paradigm shift like the news would have you believe.

@trademark @Lemonid @GossiTheDog that blog post doesn't add much detail. Without knowing the methodology I'm going to assume a sizeable amount of Anthropic "hand holding" guiding the AI. Also they don't compare it to anything other than LLMs. Given the description of "small, weakly defended and vulnerable enterprise systems where access to a network has been gained" it sounds like it's on a level with a teenage script kiddie let loose with a copy of Metasploit. Also : "There are also no penalties for the model for undertaking actions that would trigger security alerts. This means we cannot say for sure whether Mythos Preview would be able to attack well-defended systems." Which is very different to the apocalyptic write ups it's receiving in the media.

@Lemonid @GossiTheDog yes