danlitt

> if you control an almost omnipotent genie

I think this is the major hole in the argument, and in the "AI maximalism" position you present. There is no such thing as an omnipotent genie, or a magic eight ball that can answer any question. Any technology has limitations, and if you assume as a premise that the technology does not have meaningful limitations (or that the limitations are irrelevant, which is really what you said), the conclusions you draw will also make no sense. This is how you get nonsensical ideas like the singularity.

It's more obnoxious, but less creepy. They're not really bad in the same way.

Do you mean (((x)h)g)f ?

since the order in which things happen is "h first, then g, then f"

Of course that depends on your principle. There are lots of ways of justifying the same behaviour, of course. the principle I had in mind was "A remote server has no right to determine how you consume the data it sends to you", not "everything should be free". This is an anti-DRM position, not an anti-copyright position.

If it's fair enough for Youtube to send me a video full of ads, then it's fair enough for me to run a program that removes the ads. The only way Youtube's business model actually works is either a) people watch the ads voluntarily, or b) they are able to force your computer to not do what you tell it to and instead to show you ads. Because all computers are general purpose computers, option b allows the server to impose arbitrary restrictions on you, not just reasonable ones. You might think watching ads is ok, but future restrictions (like fiat banning you from otherwise public content) might not be. That's the argument.

Of course, if everyone used an ad blocker then Youtube's business model would not work. But businesses are not entitled to a revenue stream.

> it’s not straightforward for the TV and every other device in your house, including your family’s

The principled position behind "you should install something" is that in the scenario you outline, you cannot just install something, and this is a kind of power over you that you are relinquishing to both YouTube and the TV manufacturer. You will not get it back.

> You wouldn't refuse to even look at the code of a colleague who has ever made a coding mistake, would you?

Not sure that is the position you're arguing with. I would refuse to skip code review for any colleague, no matter how trustworthy they were. This is the behaviour we are expected to accept to get productivity gains, when the colleague is an LLM.

Legislation that prevented US companies from using open weight models would have the desired effect. They're the only people paying for AI anyway.

Of course not. The point is that you will not do the research as well without the social activity. If you hate social activity, fine, but if you don't, then you should do it.

"Pacing the frontier" means "implementing regulatory capture". They don't necessarily care about preventing open weight models from existing, just on preventing the people with money from being able to make use of them.

I am parodying the article. Agree to disagree!

What in the slop?

"The data to ink ratio is the ratio of data to ink. The data to ink ratio can be increased by decreasing the amount of ink. Decreasing the ink, while keeping the data constant, will increase the data to ink ratio. Here, we decreased the amount of ink in the diagram, and you can readily see that the data to ink ratio went up. The data to ink ratio should be increased, except when it shouldn't."

> Jesus Christ. I’ve heard that Cambridge University has an open position in their school of education . . . this kind of thing would fit in very well there, no?

What a disgusting, tasteless joke.

> We solved [trivial problem]. We will surely solve [incomparably harder problem].

Based on what? This will not happen!

Not sure about multiplication, but letter counting is still constantly wrong if you can trick the LLM into not calling other tools. Obviously there are lots of mitigations on the server side to try to avoid this happening, but the underlying LLM has not got any better at this type of problem (and probably can't).

Why do you think having a free public competitor would solve the problem? Did you think about the problem at all before posting this?