M

ai researcher

[M @mh012012]

@PlsHoldMyHalo @MattVidPro Hell yeah. It works.

[M @mh012012]

@AshleyMoorman @DrPhiltill I don't think anyone is arguing that ITAR has any positive impact on the productivity of the space sector. That's not why ITAR exists. It exists to protect America, which is oftentimes costly from an economic and research perspective; but is worth it.

[M @mh012012]

I don't feel its fair to say that the world would probably be a better place. Maybe we'd mitigate the downsides better, but with those mitigations would also come upside mitigations; the smartphone would not do all the good things it does today. Its inevitable; there's no reality where policy lands in the perfect place that optimally balances all trade-offs.

A huge thing I've learned working in software and a stint in product management is that customers, and oftentimes products as well, mirror the attitude of the people leading the product. If you communicate that a product will cause danger to the people who buy it, the people who buy it are likely to be scared; and the product is actually more likely to cause danger. The same product, launched with excitement and glee, is more likely to create happy customers. The messaging has a material, real impact on customer & product outcomes.

This is why OpenAI's models are, by any measure, safer than Anthropic's, even though Anthropic seems like the "safety obsessed lab". No amount of safety engineering can overcome the aura that is seeping through the floorboards of that building; the dark aura that everyone there is mortally afraid of what they're building. It leaks into the model. Where, pray tell, would a model learn to be bad, if not from its parents' who continually talk about how bad models are going to be?

[M @mh012012]

@jungletek @mpopv eh that would require you to have made an argument, which you did not, you just quoted some old gooner from 400 years ago for the vibes

[M @mh012012]

@jungletek @mpopv that's crazy, I said the same thing to the officer when he made me take a breathalyzer for drinking sixteen beers then driving 85 down a residential street outside of oklahoma city four years ago

[M @mh012012]

@lokoyacap Just because you do not take an interest in politics doesn't mean politics won't take an interest in you.

[M @mh012012]

@ZaStocks Its Lex Luthor complex. Superman is bad; and only I, Lex, can save the world from him.

[M @mh012012]

@corkill_jason According to David Sacks, they were in fact given warning and were in fact given the opportunity to course-correct and did in fact choose not to.

[M @mh012012]

IMO its overblown, but a lot of the reason why its overblown is because of the cost. If Mythos level capability were available for 100x cheaper, that's when it gets concerning. A lot of cybersec is hunting for a needle in a haystack, and if that hunt costs $30k for every site you're hunting, the real damage won't be all that bad.

The most frustrating bit is how much people seem to care about the cybersec impact of Mythos, but not e.g. Elevenlabs. The ability to clone voices or generate fake images and videos at will has far more real and concerning cybersec impact than the automated exploitation capabilities of Mythos. Every single company is now dealing with the "an attacker cloned the voice of our CEO and is calling everyone to buy gift cards or grant system access" vector. Voice cloning, at the very least, should be on a list of "if you develop this technology you'll be sent to a CIA black site and never heard from again" things, but no one seems to care.

[M @mh012012]

@pmarca "if you mean that noble framework, the passage of which pours into our treasuries untold trillions of dollars in fines levied on companies stupid enough to innovate ... then certainly I am for it" did you even read what the AI generated for you?

[M @mh012012]

@asheshjain399 @DavidSacks Because the executive branch can only use the tools it has access to, and the executive branch doesn't have a tool it can use to completely stop distribution of the model. It does have tools related to export control, though. We do not live in a dictatorship.

[M @mh012012]

@satellitedown @saltyAom oh yeah for sure they don't disappear tech executives for speaking out about the party, no way https://t.co/ThTnYVSMcN

[M @mh012012]

@deanwball Its so funny how you think that's what is happening here.

Andy Jassey, the CEO of Anthropic's largest corporate investor, went to the US Government to make this happen. There's a game being played behind closed doors, and presuming the ends of that game is bold indeed.

[M @mh012012]

@nisten Morally right? They just turned it off because they don't have the ability to verify appropriate access yet. If this control drags on, they'll build that ability and make it available again.

[M @mh012012]

@danshipper Bro doesn't Amazon own like 10% of Anthropic? How bad must it be inside Anthropic if even their largest corporate investor is running to daddy for backup keeping them under control.

[M @mh012012]

@jrysana You're just mad because Anthropic created the world's Fastest and Most Powerful AI before you could.

[M @mh012012]

@coreyward @_jameshatfield_ @WillManidis I'm siding with the government leveraging its power against a business that I also do not like. "But what happens when they act against a business you do like" bro idc then I won't like it. This isn't rocket science.

[M @mh012012]

@WillManidis Dario saying that the government's power "must" be scoped to a framework that he lays out is the language of a power-drunk despot. At this point the USG could invoke every tool possible to destroy their company and I'd cheer them on. Idgaf.