CJR study shows AI search services misinform users and ignore publisher exclusion requests.
See full article...
See full article...
AI search engines give incorrect answers at an alarming 60% rate, study says
- Thread starter JournalBot
- Start date
AI companies want to bypass copyright laws , but aren't willing to wait for laws to be updated... I am shocked.
Upvote
181
(182
/
-1)
That's about the standard American rate of correctness. If most of us actually cared about factual correctness, we wouldn't be in the situation we're in.
Upvote
188
(190
/
-2)
Upvote
182
(182
/
0)
I'm shocked, I tell you! Shocked!
All kidding aside, it doesn't take more than five minutes to figure out that these AI engines are often wrong....in the worst possible way. Subtly incorrect with an air of authority. Rarely entirely incorrect....so people give the benefit of the doubt. Sigh.
All kidding aside, it doesn't take more than five minutes to figure out that these AI engines are often wrong....in the worst possible way. Subtly incorrect with an air of authority. Rarely entirely incorrect....so people give the benefit of the doubt. Sigh.
Upvote
119
(119
/
0)
If I was evil scum I'd totally be marketing MAGAI® Search to the deplorables. Guaranteed to return doubleplusgood truthiness.
Upvote
69
(73
/
-4)
But...but...but... President Musk said X is a much better source of news than anything else!
Upvote
41
(43
/
-2)
I know there's a lot in here to point out and discuss and whatnot, but this little bit here:
I don't know that I would pay even $5/month for any of these services, so I want to know who in the world would actually pay $40/month for that. I don't know if the number would be incredibly amusing or incredibly depressing.
I don't know that I would pay even $5/month for any of these services, so I want to know who in the world would actually pay $40/month for that. I don't know if the number would be incredibly amusing or incredibly depressing.
Upvote
90
(90
/
0)
C’mon you’re posting this in an article on accuracy? Please post any proof that Columbia expelled anyone for viewpoint. They have only expelled a handful of people, all of whom were involved in the Hamilton Hall takeover.
You can say that was unjustified but there is NO evidence there was any sort of Trump related purity test.
Upvote
40
(61
/
-21)
What does a released study about AI error rates have anything to do with university administration being cowed into submission by a fascist president trying to recreate 1932 Germany.
Upvote
82
(89
/
-7)
Yeah, shocking... The other day I asked o3-mini a question, it had no idea, but still answered some bullshit again and again. I saved the "reasoning" text when I called its bullshit out, because it surprised me how revealing it was:
Upvote
93
(93
/
0)
I think of dumb ai like malicious compliance from dumb people. The cheese will stick to a pizza if you add a 1/8 cup of glue; the request was fulfilled and the solution will work. I expect the same uselessness from a chatbot or from a stoned undergrad.
Upvote
17
(17
/
0)
Not really surprising. If you use Google more than once a day, you would know that. 30 minutes prior to this being posted I searched for something on my desktop (which is the only device I haven’t moved to DDG) and the “A.I.” changed 3 simple words that could have killed me if I hadn’t know. better.
I was looking up the max safe dose of a OTC sleeping pill in 24 hours. I needed to take another, the bottle mentioned nothing about a max dose or anything, I didn’t want to take too much, and definitely didn’t want to OD. I just wanted to go to sleep.
Had I followed Google’s advice, I would be hospitalized or worse right now. Thankfully I know enough to have caught it…this time.
Just 3 words in the AI summary could kill someone. Let that sink in, and they were small words. Unimportant words to many.
Click “web” or don’t use Google folks. Google launched my career decades ago when it became public. I was hugely successful because I knew how to use it. Now I am telling you: walk away.
EDIT: before someone asks, I may provide details later, but I'm baffled and honestly considering reaching out to my lawyer to see if maybe something can be done (probably not, but he likes challenges). Due to this, I won't give details (yet), but the tl;dr is: the LLM behind the Google AI stuff changed basically suggested that the max dose is the minimum if I had a huge issue falling asleep, and suggested another random value as the max that was 5X as much. None of the "sources" suggested anything like this, so it is unclear where Google got this information.
When I finally found a reputable page on the subject, the page noted that such a high dose can cause "respiratory depression, cardiac arrest, and death".
I don't rely on AI results in general because I know how they work, but had my spouse Googled that...or my kids, or anyone else...
I was looking up the max safe dose of a OTC sleeping pill in 24 hours. I needed to take another, the bottle mentioned nothing about a max dose or anything, I didn’t want to take too much, and definitely didn’t want to OD. I just wanted to go to sleep.
Had I followed Google’s advice, I would be hospitalized or worse right now. Thankfully I know enough to have caught it…this time.
Just 3 words in the AI summary could kill someone. Let that sink in, and they were small words. Unimportant words to many.
Click “web” or don’t use Google folks. Google launched my career decades ago when it became public. I was hugely successful because I knew how to use it. Now I am telling you: walk away.
EDIT: before someone asks, I may provide details later, but I'm baffled and honestly considering reaching out to my lawyer to see if maybe something can be done (probably not, but he likes challenges). Due to this, I won't give details (yet), but the tl;dr is: the LLM behind the Google AI stuff changed basically suggested that the max dose is the minimum if I had a huge issue falling asleep, and suggested another random value as the max that was 5X as much. None of the "sources" suggested anything like this, so it is unclear where Google got this information.
When I finally found a reputable page on the subject, the page noted that such a high dose can cause "respiratory depression, cardiac arrest, and death".
I don't rely on AI results in general because I know how they work, but had my spouse Googled that...or my kids, or anyone else...
Last edited:
Upvote
125
(130
/
-5)
Upvote
36
(38
/
-2)
AI is overhyped beyond belief and I have to spend so much of my time at work swatting down STUPID AI integration ideas. The hallucination problems aren't magically getting solved anytime soon.
Upvote
63
(63
/
0)
Is it just me or is the Google Search AI summary much worse than asking an AI chatbot directly?
Upvote
18
(18
/
0)
You're overlooking the fact that he could be wrong and it could get worst.
Upvote
25
(25
/
0)
I run several highly specialized content sites and routinely see AI reference our sites despite having the setting that we don’t want our sites crawled.
AI (and tech in general really) is yet another way to leech off the work of others and monetize it for the benefit of a few.
Yay. Murica.
AI (and tech in general really) is yet another way to leech off the work of others and monetize it for the benefit of a few.
Yay. Murica.
Upvote
61
(62
/
-1)
"Room for improvement" will be its epitaph.
And that whole "today is the worst that the product will ever be" ignores precedent and reality. No matter how bad something is today, it very much can be worse tomorrow.
Citation: the world today.
Upvote
55
(55
/
0)
...but whatever the use case... it WAS wrong 60% of the time on a case.
Reason enough to worry, no?
Upvote
35
(35
/
0)
Considering the enshittification epidemic going on I find it more likely that products are at their best at launch and will just get worse over time.
Upvote
45
(46
/
-1)
It’s understandable that no one here appears to be very surprised, but it needs to be stressed that this is a horrific result. Wrong ninety percent of the time is shocking and unacceptable under any circumstances. There’s a huge problem right in front of us.
Upvote
38
(38
/
0)
The Sheep Look Up
Ars Centurion
I scraped Elsevier's entire journal collection today. But I did it using a LLM, so that's perfectly okay, right?
Upvote
40
(40
/
0)
Still more accurate than the President of the United States. I don't see why people are being so harsh on AI that is orders of magnitude superior to the most powerful man in the world. I mean its probably 100 times more accurate...
Upvote
9
(19
/
-10)
sigmasirrus
Ars Scholae Palatinae
I’d say you could sue the manufacturer for not including the safe maximum does on the bottle! Usually I see stuff like “do not exceed 3 doses in 24 hours” on OTC stuff.
Upvote
25
(25
/
0)
Did you miss the news that Microsoft, Google, Perplexity, etc. are offering LLM search and average users are in fact using it?
How is this study of live products being used by millions of users not relevant or valuable?
Also, the problem of the search ignoring website rules against scraping is in the study and is mentioned in the this article.
Did you just read an AI summary of this article?
Upvote
87
(87
/
0)
Google AI results are worst than Google search… which are really bad. Double enshittifaction?
Upvote
28
(29
/
-1)
And AGI is right around the corner and these people love to claim that hallucination isn’t as big of a problem anymore >.>
Waiting for this article to be on hacker news for everyone to come out and defend and downplay this.
There is a huge amount of money and advertising effort in convincing the average user that these tools are reliable.
Waiting for this article to be on hacker news for everyone to come out and defend and downplay this.
There is a huge amount of money and advertising effort in convincing the average user that these tools are reliable.
Upvote
27
(27
/
0)
I wonder whether they'll ever circle around to 'so bad it's good' like D-grade movies?
Upvote
7
(7
/
0)
sigmasirrus
Ars Scholae Palatinae
It’s true that it’s a little to the left. But not by much. An ordinary user might ask “what was that BBC article about the polar bears recently?” And expect to find an answer. In fact, I’ve had ChatGPT answer vague questions like that successfully sometimes. You would think if the tool is any good, if you’re more specific it would do a really good job of finding the article. In the study, though, rather than say “I don’t know” it makes URLs up sometimes.
Maybe if the hallucination rate were 1 in a million requests it might be more worthwhile since now you’re approaching the reliability of the underlying dataset (I.e. all the top 10 results on Google, when Google was still decent, could still be wrong). But this is a much higher error rate. So you never know when to trust it, making it kinda useless.
Actually less that useless because it’ll lull you into a false sense of security.
Upvote
28
(28
/
0)
sigmasirrus
Ars Scholae Palatinae
Meanwhile in the search for quality information I came across a search engine for scientific papers that seems really good, better than Google Scholar.
https://openalex.org/
https://openalex.org/
Upvote
13
(13
/
0)
I didn't see it in the article, but the study had a narrower focus than just "AI is inaccurate"
That's setting aside things like citing a real article that is unrelated to the one being searched for.
It argued the AI as a search engine is worse than modern plain Googling. Given more people and companies are using AI instead of vanilla algos, I can see that as a cromulent warning.
That's setting aside things like citing a real article that is unrelated to the one being searched for.
Upvote
40
(40
/
0)
It's not always true though, look at Google Search. It was significantly better both at finding information and as a user experience some four years ago, before Prabhakar Raghavan sank his bean-counter talons into it. In our current era of late-stage capitalism, it's entirely possible - nay, likely - for technologies to in fact get objectively worse in the name of The Almighty Shareholder Value (pbui).
Upvote
54
(54
/
0)