551–560 of 587 posts
Earlier quoted context omitted.
What if they just make a mistake, unintentionally?
Publishing lies that were generated by the plausible-lie-generating-machine is a very intentional action, it's not a "mistake". The legal profession uses the term "reckless disregard", see, e.g., [1]. [1] https://en.wikipedia.org/wiki/New_York_Times_Co._v._Sullivan
Earlier quoted context omitted.
Do LLMs knowingly make mistakes? Do we sue scientists for unintentional procedural errors?
This isn't about what the LLM knows. No one's trying to hold a computer liable. This is about what Google knows. And Google knows perfectly well that LLMs hallucinate all the time. They will provide incorrect information, confidently and often. Which is why this whole article is explicitly about holding Google accountable.
Earlier quoted context omitted.
Publishing lies that were generated by the plausible-lie-generating-machine is a very intentional action, it's not a "mistake". The legal profession uses the term "reckless disregard", see, e.g., [1]. [1] https://en.wikipedia.org/wiki/New_York_Times_Co._v._Sullivan
It’s illegal to make factual errors. Wow, cool, I guess we solved science very easily!
Earlier quoted context omitted.
I don't even know that I would call it slowing down so much as constraining/focusing innovation. There's three basic paths for a company hit by this ruling to comply: 1. Stop showing users generated content. 2. Figure out how to generate the content with more quotes and attribution to source websites, to regain the protection offered to search engines. 3. Figure out the hallucination problem, so that every statement…
> If this ruling forces companies to put more money into #3, whereas now they're coasting on good enough, I'd say it was speeding up innovation. The thing is, no one has the slightest idea how to stop hallucinations. The models are fundamentally "hallucinatory" at core - they generate what is _probable to follow the string thus far in its training corpus_, modulo RLHF and friends. Notice that nothing there has any ri…
Earlier quoted context omitted.
We're talking about a country where someone had their house raided and their devices stolen by the police for tweeting "Du bist so 1 Pimmel" (you are such a dick) to a politician.
This was later deemed inappropriate by courts. Shit happens -- as long as the corrective mechanisms work it is just exceptions.
Earlier quoted context omitted.
it's not clearly labeled. it's a search engine. i expect deterministic responses from that.
It literally says “AI can make mistakes.” And yeah, I’m sure everything else on the Internet was true before.
that trust level obviously varies, but for example if the page found is from wikipedia i know that the content has been reviewed by multiple people, so i can trust that more than say a blog post on the same topic.
on the other hand, AI results are not certain to be an accurate reflection of that wikipedia content, and therefore it is less useful than the page itself.
practically speaking i don't even look at the results in detail. i just check if my search keywords match and then i open the pages and read them myself.
the AI summary does not tell me which keywords on the pages matched. it is therefore completely useless.
Earlier quoted context omitted.
This isn't about what the LLM knows. No one's trying to hold a computer liable. This is about what Google knows. And Google knows perfectly well that LLMs hallucinate all the time. They will provide incorrect information, confidently and often. Which is why this whole article is explicitly about holding Google accountable.
You honestly think Google is intentionally giving wrong answers??? Google would love to have a perfect oracle that never makes mistakes! Maybe German courts can conjure one up.
Google 100% has the choice not to include AI overviews. Which they know are often false.
It would mean egg on their face, to be sure, but would it even meaningfully impact revenue?
Earlier quoted context omitted.
It doesn’t matter who you’re angry with in that case, because you’re going to suffer the consequences regardless. No lawsuit is going to reimburse for that.
No one is going to suffer consequences because of a Google AI summary. If they do, that's not Google's problem. It's someone else's fault, principally those who blindly acted on information they were told was potentially incorrect. In the world according to you, AI couldn't exist. Or more likely, it would be accessible only to academic and corporate/financial/government/military elites. That's not OK, and I'm unwilli…
No one is, until someone is. You could say the same about newspapers, TV, or the internet.
> If they do, that's not Google's problem.
Well yes, it is, at least in Germany for now, where at least one judge seems to see it that way. And that's correct, IMHO, because Google would (in that case) make something appear as factual information when it is not.
> It's someone else's fault, principally those who blindly acted on information they were told was potentially incorrect.
Those people are definitely at fault; history is full of examples of mobs acting horribly based on shallow, incomplete, misleading, flawed, or flat out wrong information. But you can't just put all blame on those mobs, when someone gave them this information - or do you think, for example, Donald Trump played no part in the January 6 riots and should not be held accountable for it?
> In the world according to you, AI couldn't exist. Or more likely, it would be accessible only to academic and corporate/financial/government/military elites. That's not OK, and I'm unwilling to join you in pretending that it is.
That isn't my stance at all. AI should exist, and it should be widely available in my opinion. I don't even think this is primarily about AI, but entities in a position of power acting responsibly, such as Google (which has worked hard to position itself as a gateway to information and steward of facts.)
I think we should not put the burden of verifying information entirely on consumers, we shouldn't allow big corporations to run the largest social experiment in history on their own terms, and we need to talk about responsibility and safety.
> So your position is that the general public should be given access to AI only when it is either capable of flawless accuracy, or when the AI provider is prepared to assume unbounded liability despite warning the user that perfect accuracy is not possible.
That is neither my, nor the sibling's, nor this ruling's position. The issue the court took was that Google presented AI-generated information in a shape that was indistinguishable to common users from the previously statistically sourced factoids in Google Search - a product that historically allowed to search the internet for things and get back search results, not AI-provided guesses and hopefully-correct information.
The entire point is that companies should not be allowed to use AI recklessly. This point is also one that the Pope made in his encyclical, and the EU posits in the AI act, by the way.
Earlier quoted context omitted.
This was later deemed inappropriate by courts. Shit happens -- as long as the corrective mechanisms work it is just exceptions.
Why is it even allowed to raid someone for victimless crimes? It's a fundamental issue, not just a mistake.
Idk what exactly happend. But somehow they got approval from court to collect evidence of the crime (electronic devices) via house search. As said, the harm caused by the house search was later deemed unproportional to the crime.
Earlier quoted context omitted.
No one is going to suffer consequences because of a Google AI summary. If they do, that's not Google's problem. It's someone else's fault, principally those who blindly acted on information they were told was potentially incorrect. In the world according to you, AI couldn't exist. Or more likely, it would be accessible only to academic and corporate/financial/government/military elites. That's not OK, and I'm unwilli…
> No one is going to suffer consequences because of a Google AI summary. No one is, until someone is. You could say the same about newspapers, TV, or the internet. > If they do, that's not Google's problem. Well yes, it is, at least in Germany for now, where at least one judge seems to see it that way. And that's correct, IMHO, because Google would (in that case) make something appear as factual information when it i…
I'm missing the comparison here. Trump (in)famously did not post any disclaimers. He did not leave any room for doubt when he accused the Democrats of stealing the election. He spread feces and called it fact, so yes, he should certainly have been held accountable for the consequences. His voters, however, decided not to do so, and that was the end of it.
The issue the court took was that Google presented AI-generated information in a shape that was indistinguishable to common users from the previously statistically sourced factoids in Google Search - a product that historically allowed to search the internet for things and get back search results, not AI-provided guesses and hopefully-correct information.
Did they, or did they not, include a highly-visible disclaimer that the results might be incorrect?
I'm hammering on this, not just because I think it's what should have determined the outcome of the case, but also because the last few times I've seen Google AI summaries, they did not include such a disclaimer. Gemini itself still does, but the instant results on the search page don't appear to. Which is obviously not OK.
If they have stopped warning users, or if they didn't warn the users in the situation leading up to the lawsuit, then that more-or-less instantly flips me over to your side. We wouldn't actually have anything to argue about in that case.