I can’t see that. Yes, the press is hyping Mythos in a complete brainless way, without understanding what the Mythos paper says.
But just following the statements made by Anthropic that is a clear and factual representation.
I actually recommend that people read the paper itself. And not just die cherry picked parts of the press.
Also, I have seen 3 reported 0days found by Mythos in projects I (or my employer) contributes.
That is not hype, those found issues are fact.
If You made it to here in the comment, a little bonus. Let us debunk some press statements
Press: Mythos cheated, lied and is dangerous.
Fact: earlier versions did this. Anthropic describes the preview version as the most aligned model they have ever created, and cheating occured in less then 1% of cases
Press: Mythos is a security nightmare
Fact: in the wrong hands, it could be.
Press: Mythos exploited Firefox!
Fact: yes and no. It founds zero day’s in Firefox, but the extremely high success rate where an instrumented JS environment, and not the full browser
I can’t see that. Yes, the press is hyping Mythos in a complete brainless way, without understanding what the Mythos paper says.
But just following the statements made by Anthropic that is a clear and factual representation.
I actually recommend that people read the paper itself. And not just die cherry picked parts of the press.
Also, I have seen 3 reported 0days found by Mythos in projects I (or my employer) contributes.
That is not hype, those found issues are fact.
If You made it to here in the comment, a little bonus. Let us debunk some press statements
Press: Mythos cheated, lied and is dangerous.
Fact: earlier versions did this. Anthropic describes the preview version as the most aligned model they have ever created, and cheating occured in less then 1% of cases
Press: Mythos is a security nightmare
Fact: in the wrong hands, it could be.
Press: Mythos exploited Firefox!
Fact: yes and no. It founds zero day’s in Firefox, but the extremely high success rate where an instrumented JS environment, and not the full browser