John Lago

It's amazing how much faster my packages install with the Jetify Prebuilt Cache:

@TheStalwart There's a lot of questions that LLMs can't solve directly unless they write code. Similarly, I suspect complex images won't be solved until they can create references using Unreal engine or something

Tried this on pro mode, and got one of the best attempts I've seen yet. But still no dice

A man with his feet planted firmly in the future. Why send emails when your agents can do it for you?

@ibuildthecloud Maybe there’s more but it’s hard for me to see how Ant handles this differently given the contract change and their red lines. I think they’ve navigated better than their gov’t counterparts. But I suppose agree to disagree. All the best Darren.

Yes, no, maybe. There's just too much nuance here at this point. Which ever side you think is "right" is going to be based on predisposed biases. I don't really care about that. The only thing I was ever commenting on his how terrible the CEO of Anthropic has been in navigating this. Completely ineffective in having any positive impact. But my original post in this thread I did say I suspected the DoW wanted to get rid of Anthropic. Which is of course complete speculation and can't be substantiated. Just a dumb opinion. Anyhow. If you want to say Anthropic did the right thing and DoW is evil. That's fine. I'm don't care much about that.

Watch this video. If you read between the lines, I don't think the govt wanted the deal to work. Just watch the CEO, it's clear this is not a company you want to work with. I wouldn't be surprised if he's replaced before they IPO. youtu.be/MPTNHrq_4LU?si…

@ibuildthecloud The point is that it wasn't a hypothetical or about future control, they actually had a planned use case that violated the provision

@jetpack_john Wasn't that always the stated reason? Anthropic has two provisions the DoW wouldn't agree to.

@ibuildthecloud Atlantic today had more details over where the negotiations broke down. It was the mass surveillance provision: theatlantic.com/technology/202…

Totally reasonable. Everyone is concerned about that. He has every reason to be concerned. The DoW stated they share the same concern and/or desires. But this has nothing to do with those concerns. There's just a total misunderstanding from Anthropics side on how these matters work. The points he's making are just not grounded in any rational logic. This is why I'd guess the govt doesn't want to work with them. And if they gave them a 3 day deadline, they were probably cutting ties.

@ibuildthecloud The whole thing is very confusing, because OpenAI is saying in their announcement that autonomous drones and domestic surveillance will not possible via their "safety stack". Is that different from Anthropic's safeguards?

Yes, but that doesn't track with the fact that OpenAI was not required to remove any safe guards. But it would make sense that if Anthropic has already added safeguards to would prevent the model from doing "any lawful action" they would have to remove them. So you are probably right that the goverment would need any such safeguards removed. But "mass surveillance" and "fully autonomous killing" are not legal and as such would not require safeguards for those things to be removed. But if a future law did make them legal (which obviously would be bad) then they would be required to remove those safeguards at that time. On that condition OpenAI has agreed and Anthropic clearly hasn't.

@ibuildthecloud I mean, a gun that fires on an rng would also be a liability

@jetpack_john Models don't make decisions either. It's a random number generator. The decision was made prior when the application was put into place. But this is purely speculative and neither of us know anything about law. So who knows.

@ibuildthecloud Gun's don't make decisions. Under the drone scenario, Anthropic's models would

They wouldn't be liable for that. I don't see how that would be possible. A gun company is not liable for people killed by their guns. But yes, Anthropic is totally free to say no and not sell to the govt. As they have and they absolutely should have that right. I'm commenting just that I think they are stupid. Going about this in a ridiculous way that doesn't really accomplish anything. But that's the beauty of America, you are free to be an idiot and fully applaud them for embracing their stupidity.

@ibuildthecloud Anthropic mentions removing safeguards in their statement. OpenAI was not working with military before this afaict, so they didn't have the same safety concerns.

as far as I understand it was not that the govt needed them to remove safeguards (openai didn't remove anything) it was that they couldn't agree to the conditions and terms. Anthropic made the point they didn't think the models were safe for those conditions, which are currently illegal and were never asked for.

@gurtis @ibuildthecloud True, by all accounts Ant was great at fulfilling its current contract. If they don’t want to expand the contract for the two scenarios, why not get another LLM and leave the existing contract as is?

@ibuildthecloud @jetpack_john It sounds like what Anthropic is concerned about is *not* already illegal. The gov't can legally gather large amounts of data and run it through an LLM to build a profile on whoever. Why try to force Ant to do something they don't want to do instead of just going somewhere else?

@ibuildthecloud Also, Ant would have brand/business/maybe-legal liability if drones on Claude kill mode violated the law. They’re free to refuse that risk

@ibuildthecloud If the govt needs a model that is unfettered for lawful uses, and they agree that the two scenarios are unlawful, why do they need Ant to lift the safeguards? Ant has a right to not build a products they think are dangerous or create liability. Govt can choose a different model

@devahaz Every new beverage trend is just "How can we disguise this milkshake?"

Tea dessert acquired

> What if a nuclear [ICBM] were hurtling towards the US with only 90 seconds to spare, and Anthropic’s AI were the only way to trigger a missile response to save the country, but the company’s safeguards wouldn’t allow it Is... that really how they plan to set this up?

@almonk is there a way to hit multiple modifiers (Ctrl + Alt) in Echo?

@danhockenmaier It’s a big assumption that the agents work seamlessly, and even then I don’t think you’d get vibecoded apps taking share. But maybe some adjacent companies find it more viable to compete.

@danhockenmaier Customer support? You’ll use something like Fin with Voice support. Routing? Each dasher has an agent on their device that bids aggressively for jobs that are on their route.

JUNE 2028. The S&P is down 38% from its highs. And you're hungry. You open your phone. Yesterday, you asked Claude to order you a burrito. It routed the order directly to the restaurant and saved you $1.25. But it took an hour and half to show up. It was cold. And they forgot the gauc. Who owes you a refund? Does Anthropic have a customer service number? So today, you open DoorDash. The burrito is the first recommendation. You tap on it. 19 minutes later you're holding your burrito, still warm. They remembered the gauc. "I guess AI was a sustaining innovation after all" you mutter as you take your first bite