[SllX]

By checking the citations rather than taking what’s generated at face value.

If it’s important, check it. If it’s not important, then it is pretty much just entertainment.

LLMs can be very useful in a general web search and save some time, but if you don’t put those literacy & critical thinking skills to the test and actually confirm anything, then you might as well not even have bothered with the search at all unless you’re hoping it can just replace all of your original thinking too.

[Gigachad]

If google didn't intend it's answers to be taken at face value it would just present the citations in a list of links rather than generating an answer.

Obviously the marketing point of the AI tools is it just gives you an answer straight up so you don't have to bother reading normal sources.

[SllX]

The AI summary is still useful for narrowing down the results, even if you fully check the citations.

> Obviously the marketing point of the AI tools is it just gives you an answer straight up so you don't have to bother reading normal sources.

To lazy people yes. That would be a marketing point. It’s not that though, so you use it to save time, but you don’t get to skip the verification step.

[Gigachad]

Google should not be publishing a statement that they haven't verified. This is different to listing search results links, they are the ones publishing the content here.

A journalist could not make up a harmful statement about someone and get away by saying the readers should have all read the sources. AI companies want to take all the benefits and profits, while holding none of the liability and responsibility for the harms they are causing.

[hunterpayne]

"A journalist could not make up a harmful statement about someone and get away by saying the readers should have all read the sources."

Journalists do that all the time. We even have a whole collection of words to describe it. Muckraking for instance is probably about 100 years old. Its even in the Google auto-complete in the browser I am posting from.

[Gigachad]

Notice the “get away with” part. Journalists can be held liable for making things up. And so should Google.

[hunterpayne]

They aren't really held accountable in any way for this type of stuff.

[intended]

Absolutely not.

LLMs are, for all intents and purposes, the equivalent of outsourced workers.

Google created a summary, not just sharing search results.

Google is responsible for the output it created and then published.

If they had only surfaced search results, then they would not be liable for what other people generated.

Google’s scale does not protect it from this liability.

[clort]

If you re-read the article, you might see that it mentions that the citations do not necessarily cover the AI summary. The linked pages do not make the claims that the AI summary makes. That is the context of the ruling. Google made up the claims, and provided false citations. They are not, in fact, providing a summary, but a whole new narrative. Therefore they own it.

[SllX]

I read the article and I’m aware of the failure modes of Google’s AI summary. They’re actually one of the worst in the space on this shit which is why I don’t use Gemini and it’s fine that they get slapped for this, but what I was responding to initially was this:

> errors can be so subtle that it is not possible to recognize them unless you spend an hour researching every fact presented. at that point, what's the benefit of AI? nobody is going to do that.

Because if someone goes through the citations and it doesn’t substantiate what was generated, then what was generated was obviously bollocks. Being able to recognize those contradictions is an essential skill to using LLMs with web search at all. It’s not rocket science.

[danaris]

> Being able to recognize those contradictions is an essential skill

My eyes are brown.

I dislike coffee.

My phone is on the desk next to me right now.

One of these is false, and the other two are true. Can you recognise which is which? Or do you not have this "essential skill"?

When you're being given information about a topic you don't already know about, there's no skill to be able to recognise which pieces of that information are correct and which aren't. Either you know the information already, or you don't.

[em-bee]

nobody has that skill. in order to recognize a contradiction you have to already know something contradictory. if you don't then you can't recognize anything. the only other reason to make you check is that you are very suspicious of this AI thing. you and me are, but who else is? you can repeat a thousand times that people should not trust an AI summary and they will not get it. they have neither the motivation nor the energy to do that research.

it's like saying people should not use a computer if they don't know how to keep it secure.

you have to look at the reality, which is that people are not educated or critical enough to use AI safely. and that misinformation can cause people to get killed.

just yesterday there was a post about a man being falsely accused of a crime by AI tools and losing his job and his family as a consequence.

these mistakes destroy lives, and most people are far to trusting to use AI tools safely

[broken-kebab]

Important for whom exactly? If it's you who are called convicted pedo by Google AI summary, it's you who has vital interest in additional research but not me who reads it. There's intentions mismatch. Which probably would destroy your life, and you won't call it "entertainment" then, I think.

[SllX]

Strawmen are for scaring off the crows, not discussion fodder. Take it out to the farm where it belongs.

[broken-kebab]

Could you elaborate what in my thesis of intentions mismatch makes you think it is not a valid argument?

[oaiey]

But you are not doing a 2 hour rabbit hole search when you stand in front of a T-shirt and check whether it is fair traded or all-american produced.

[SllX]

If those are things you legitimately care about before you spend one penny on a T-Shirt, then you are. Or you did your research before hand. Or you’re just not buying the T-Shirt.

Or you don’t care about those things at all, and you will buy the T-Shirt that’s in front of you right now rather than wait later and buy one that better reflects your supposed values when you’ve done an appropriate amount of research. Using AI may even reduce the amount of time you spend on that part.

Your T-Shirt buying patterns & values are not my concern though.

[isaacfrond]

You are confusing who is the party that is injured--as many in this thread are. We are not talking about the consumer who buys a non-fairtrade t-shirt, when he would rather had a fairtrade one. It's about the t-shirt producers who is legitimately fairtrade but whose business is now in the shitter because of a lying AI.

[SllX]

LLMs can’t lie. They are incapable of telling either the truth or lying. To the extent that they are in any way useful, it’s recognizing that they are text generators attached to crawlers and other tools that can with the right inputs produce useful generated text that may also incidentally be correct or incorrect.

Businesses might (well, will) suffer because people are misusing AI, but it is a misuse to do anything with it without an additional verification step.

To be clear here, I have no issue with Google taking it on the chin in cases like this, but what the comment I was originally responding to had this:

> errors can be so subtle that it is not possible to recognize them unless you spend an hour researching every fact presented. at that point, what's the benefit of AI? nobody is going to do that.

And my point is this: if it matters, verification is not optional. If it doesn’t matter, then fine, skip the verification step, but if you’re taking whatever text is generated by a GPT at face value without understanding what that is or being able to determine the source inputs for the “claims” it outputs, then you’re part of the problem because sometimes the source is just a GPT-generated web page, and that’s obviously not trustworthy. Sometimes it’s a MediaWiki site page that doesn’t actually exist, but because it’s MediaWiki it’s not going to return a 404. Using a tool requires understanding it including its failure modes, and in the case of LLMs that means: trust nothing, verify everything.