jackdeansmith

I think it would be entirely reasonable to ban death stars in private hands, the same way I think it's entirely reasonable to ban uranium enrichment in private hands. My point is that it's a question of fact about how risky an AI model can be. I think it's clear that current models are not enriched uranium or death star level, but I don't think there's anything ruling it out in the near future!

>If the biggest danger of LLMs is that they can hack traditional systems

This is doing a lot of lifting. If the biggest danger of LLMs is they could uplift bioweapon development, the situation is different. If the biggest danger of LLMs is they reach capabilities allowing for recursive self improvement, the situation is very different still.

I see a very clear link, very capable AI models are clearly able to find and exploit cyber vulnerabilities at a speed and scale not possible before. In the right hands, that's clearly a weapon

And what are those ingredients for biology, which can be constrained as effectively as uranium enrichment? I'd argue there isn't anything which can easily be restricted or monitored.

What about uranium enrichment?

They can't. You have to explicitly give apps permission to access your photo library.

I think it's a lot harder to have this particular type of communications failure if you're writing on a chalkboard. Imagine trying to write out that whole slide, it would take forever. If you really did have to present that information on a chalkboard, you'd be significantly more likely to write something along the lines of:

"We checked the test data: possible to damage tiles significantly" "Foam that hit wing was way bigger than the tests"

Obviously you can miscommunicate via any medium, but I think the author's point here (which I agree with) is some mediums lend themselves to specific types of miscommunication.

I think this is clearly a joke

>The real tragedy is not that Hims exists, but that it works so perfectly. Every day, thousands of people choose their compounded weight-loss drugs over FDA-approved alternatives, their combination ED pills over established single-ingredient treatments, their algorithmic consultations over actual medical care. They make these choices not because the products are better, but because the entire experience has been optimized to feel more like shopping and less like confronting the mortality and vulnerability that define the human condition.

Strongly disagree with almost everything in this article, but specifically this. The reason people make these choices is not because of slick marketing working against them, it's because the existing process to get medical treatment is paternalistic, hard to navigate and often expensive.

If you want safe and really high quality medical care you should absolutely have a personal physician you have a personal relationship with, who understands your lifestyle, your risk factors for side effects, and your medical needs deeply. How many Americans have that? Maybe a few dozen? The market has responded to just how terrible the existing system is.

I misunderstood your comment then, my assertion is that models which have the same capabilities as current models will be cheaper in the future. I have no doubts that models with more capabilities will be more expensive.

Want to bet? I'll give you 5:1 odds that tokens from a model with some specific benchmark performance (we can sort out the specific benchmarks or basket of benchmarks if you want to bet) will be cheaper two years from now.

In my opinion most Anthropic models are the opposite, scoring well on benchmarks but not always way on top, but quietly excellent when you actually try to use them for stuff.

What would convince you that a system is too dangerous to release?

There was certainly some client side censorship, but also the model weights clearly had some of this baked in. You can try yourself by asking the same queries to r1 hosted by a US host like together.

As someone born in 1997 and working in the software industry, one thing I’ve learned is that the discipline of software engineering is sometimes mostly about understanding the history of a system. Love reading stuff like this because sometimes it really makes some concept click!