Anthropic News & Discussions

[firestar464]

The way I see it, Anthropic never denied the risk of their own models. That's why they gave early access to trusted entities so that they could patch their systems. That way, even if bad actors got Mythos-level capabilities, key infrastructure would already be appropriately upgraded. This is why they are now saying that the export controls are unnecessary.

It's also disconcerting to see people like Jimmy Apples yap constantly about "anthropic and their evil safety agenda" until Anthropic gets restricted, and suddenly safety is important

Apparently they really do think that Sam's boot is that tasty

[firestar464]

[weatheriscool]

[Cyber_Rebel]

Can only imagine the drama once actual AGI is here. Frictions between the government, private companies, citizens, and how it might be deployed internationally are going to be a more challenging ride than building it.

[weatheriscool]

[firestar464]

Instead, she wrote, the researchers simply asked Fable to fix open source code with public and known vulnerabilities along with “deliberately planted vulnerabilities,” after the model initially refused to “review the code for security issues.

Bro, they called this an issue? b r u h

[weatheriscool]

If he really feels this way he should disband his company and never release another model. If he releases more advance models in the future he is just trying to f*ck over other companies. The guys a snake.

[weatheriscool]

Hopefully he'll let Fable come back!!!

[firestar464]

ah yes, the epitome of sane policy

Really, any ideological/geopolitical motives are completely second with this guy. His whims and mood come first.

Whatever in this case, as long as I get Fable back.

[firestar464]

(bottom guy was reposted by Jimmy)

[weatheriscool]

I believe that cyber security will have more and more to do with a.i guards in the future that actively protect secrets and important online infrastructure . This is one reason why the US government will probably take control of the most powerful models as they'll be used for this purpose. Passwords and old methods won't work in this future.

Super mythos models will be made as guard dogs to literally defeat your hacking attempts.