Hacker News Reader

Comment by ppalata | Hacker News Reader

> Then they'll be indistinguishable from the text of the AI but still separate. Then they'll start folding it in to the AI prompts. Then they'll start folding it in at the training phase so the weights themselves drive you towards Thinking Ford First, this step being inevitable because it almost certainly will be by far the most effective of all of these techniques. Opus 7.2 Brought To You By Mountain Dew. Just a coincidence that every time you ask it to build an authorization system it writes code to ensure that the user has drunk a verification can, and is just incompetent at removing that code for some reason.

I know close to nothing about the US regulation in this space but at least in Europe, when you pay for advertisement, it has to be always clearly communicated to the user. Are you assuming that they will just ignore the law in such a blatant way?

View on HN

I would guess they’ll spend ages arguing about what an “ad” is.

Eg if I ask Llama 3 to find me the best headphones, I think we can agree its response isn’t an ad.

If Claude responds with a banner that Sennheiser designed and paid Claude to show, that’s an ad.

They’ll live in the probabilistic grey area between those. If Sennheiser pays Anthropic to add “Sennheiser makes really good headphones” to the system prompt, is it an ad if Claude recommends Sennheiser headphones? What if they pay to ensure their site is included in the web search tool results for any queries about headphones?

It’s sort of akin to paying Google to rank higher in searches if that were a thing they did.

I do think they should have to declare those, but they’re not traditional advertising to me.

> If Sennheiser pays Anthropic to add “Sennheiser makes really good headphones” to the system prompt,

I think you mean "OpenAI is partnering with Sennheiser to improve its AI."

So, while it's tempting to be cynical and sarcastic, my honest answer is that I expect them to try it, and it would actually be a value judgment as to whether that constitutes ignoring the law.

Common Law-based systems prioritize the consistency of precedents. As a result, the real law is often built on the margins, to borrow an economics terms. Which means that in a sense, we find out what the law "really says" by someone first breaking it and someone else challenging them. What I expect to happen is that we will eventually get to the point that someone is going to judge that the value of trying this is more than the expected value of the resulting lawsuit, and then this will eventually get to trial, and then, a couple of years later, we'll "know" whether or not this is breaking the law.

But there is a sense, a real, non-sarcastic sense, in which until that point, we don't really know if it's illegal or not.

Most of Europe does not operate under Common Law, so this may seem strange.

Load up on verification cans because the companies who got away with flaunting EU advertising/tracking laws will get away with obnoxious ad-driven AI too.

(The cookie consent banners that you see everywhere were anticipated and forbidden in the laws themselves, companies just gambled that the EU wouldn't be able to enforce the law, and they were correct. I expect something similar to happen with for-profit AI steering.)

Reply on news.ycombinator.com