Earlier on, I asked a question, looking for a source for a quote. Someone helpfully asked Chat GPT. The answer it gave was 100% wrong. (The answer that humans gave was right.) Please stop asking it. It doesn’t work. And it doesn’t help. Bad arseholes want you to use it. So stop. And everyone wins. - ThreadSky

gralefrit.bsky.social • 61 days ago

Earlier on, I asked a question, looking for a source for a quote. Someone helpfully asked Chat GPT. The answer it gave was 100% wrong. (The answer that humans gave was right.)

Please stop asking it.

It doesn’t work. And it doesn’t help. Bad arseholes want you to use it. So stop. And everyone wins.

Comments

mxp.bsky.social•61 days ago

This is the fundamental issue. Society is being nudged to rely on ChatGPT as a primary source and I don't see it being widely chastised in the same way Wikipedia was 10-15 years ago for the same thing.

Down the line, we'll be in a media illiterate world where the LLM models have eaten themselves.

paulbrock.co.uk•60 days ago

It's wild that chatgpt responses are shared as "proof" of an answer. Utterly unreliable

nookrat.bsky.social•60 days ago

Surely the only use of LLMs is to mock their answers.

charlie-or-similar.bsky.social•60 days ago

Have seen this doing the rounds on a.n.other platform. Would be nice to think there's a resistance brewing.

samanthabwriter.bsky.social•61 days ago

There’s a group I’m in where people post about games and books and movies that they can’t remember the full details about and someone always gives completely inaccurate answers they got from ChatGPT. And they’re always defensive when you tell them that their little AI friend just made that crap up.

vickythelabrat.bsky.social•61 days ago

I'm training a guy who insists on using it to find the answers to technical questions. He has a masters in AI, so he's probably biased, but I just wouldn't trust it!

mzdt.bsky.social•60 days ago

I have no interest in AI at all, but it should at least be more accurate in a fact-based discipline like engineering - but I still don't know why anyone would bother asking it.

seamas.bsky.social•60 days ago

Oh, AI, how I hate thee.

https://bsky.app/profile/did:plc:lss244cnzeoicm3p3nrg5ovh/post/3lecbuahrks2c

harecomedy.bsky.social•61 days ago

It's a bit like asking a furby.

fplshaman.bsky.social•61 days ago

What's your point?

gralefrit.bsky.social•61 days ago

Whenever someone helpfully throws AI answers at me, I shiver with the realisation that using LLMs has become routine for people. Maybe they use AI at work, or are told to, to do basic searches, simple summaries etc. I think if it’s become a habit, rather than a dumb toy, that’s dead sad.

mckelvie.bsky.social•61 days ago

I saw someone recently say if you use it, when it gives you an answer you know is actually right, tell it it's wrong, and watch it agree and then give you an incorrect answer instead. It's luck when it gets something right.

gralefrit.bsky.social•61 days ago

(I’m not blaming anyone for using it to help with my enquiry. I’m just gloomy that it was in any way a first port of call. That it’s normalised itself into casual exchanges. It’s bobbins. And it lies. Checking it was in fact wrong took longer than getting the right answer.)

rossmc11.bsky.social•61 days ago

It's somehow useless both for the specific and unhelpful in general. I asked it for a history of a local festival and it told me all about the first recorded one in 1815, which was a surprise after last year's 50th anniversary event.

dewiioan.bsky.social•60 days ago

This is the issue, it's sometimes right, sometimes wrong, but it's always equally convincing, there is no difference between a right answer, and a wrong made up one, with citations, references...

gralefrit.bsky.social•61 days ago

To check that, I had to go onto Internet Archive and take out the alleged source text, and search it. It came up blank. Now, the Internet Archive is cool, but shouldn’t exist (copyright, yeah). If the Archive closed, I’d have no way to check. The LLM would be a closed black box. That’s the worry.

gralefrit.bsky.social•61 days ago

It’s stolen everything, hidden its sources, and soon using it will replace our ability to quickly interrogate its findings, or to understand how it got to whichever mad conclusion it reached.

richardawkward.bsky.social•61 days ago

I had to research the oil industry in Burundi recently. ChatGPT invented a fictitious national oil company, Petrobi. I smelt a rat and when I asked to to share its sources it gave an electronic shrug and said (I paraphrase) "well, it's the kind of thing that usually happens there."

richardawkward.bsky.social•60 days ago

If I didn't already know a fair bit about the subject I'd have just taken it at its word and Petrobi would have become a fact.

geospacedman.bsky.social•60 days ago

That last quoted remark sounds like something somebody's racist uncle would say about Burundi.

ceoln.bsky.social•60 days ago

That is literally all that the LLM itself does! "Here are some words that tend to occur in this sort of context". All the big techs want to add "and are true", but so far they aren't having much luck.

I miss when LLMs were just like a crazy stoned friend saying obviously random BS.

ghosttapes.bsky.social•60 days ago

Genuine question... Why did you use ChatGPT at all? What did you expect to get from it bearing in mind it's not a research tool, it just makes up plausible sounding text?

pieist.xenoplasm.com•60 days ago

Amusingly, Nigel Balchin, the author of The Small Back Room, was a pioneer in the use of computers for research.

My guess is that when ChatGPT decides to hallucinate, its daddy issues come into play.

gralefrit.bsky.social•60 days ago

I did wonder about that…

angusmain.bsky.social•61 days ago

I tested it with that quote myself, because I often get students giving weird attributions for quotes and wondered if this was the source.

angusmain.bsky.social•61 days ago

ChatGPT told me it was from Dickens. When I asked it for more information it said it was from Chapter 2 of The Pickwick Papers, gave me a fuller version of the quote, and cited several books which had used the quote as an example of Dickens’s wit. All of it made up.

dottyeyes.bsky.social•60 days ago

Holy heck! Lying line a psychopath!

angusmain.bsky.social•61 days ago

It’s not just that ChatGPT is wrong, it’s that it actively tries to convince you that the information it made up is true. It’s a confidence trick.

gralefrit.bsky.social•61 days ago

The helpful reply I got told me it was from The Small Back Room by Ernest Balchin. That piqued my interest, because I’d just watched the Powell and Pressburger doc, so it sent me into a Balchin rabbit hole that was very exciting and totally wrong. That’s the danger. That it inspires human interest.

alex.breaks.news•60 days ago

I used Perplexity earlier today which gave me an extremely biased answer (although to a rather mundane and unimportant question). However, I knew it was biased because unlike ChatGPT, it tells me the sources.

AI is a tool, and I think if used correctly and at the right time, it can be useful.

cockbeard.bsky.social•61 days ago

It's the difference between 'phone a friend' and 'ask the audience'

Anything that you think you should know, it's fine, ask the audience, it's called common knowledge. Anything even remotely specialist, or contentious, then phone a friend

I'd not let a phone vote on a personal diagnosis

gralefrit.bsky.social•61 days ago

Maybe the prompt bar should include the word “reckon”. And instead of Alexa, all AI assistants should be called “Deano.”

“Deano. Who do you reckon wrote the Mill On The Floss?”

amazingmikeyc.bsky.social•60 days ago

Why read a summary of the topic on Wikipedia when you can get a rewritten version of the article with an extra bonus lie in it.

nookrat.bsky.social•60 days ago

The game is spotting the lie.

It's like that icebreaker they make you play at team building days.

amazingmikeyc.bsky.social•60 days ago

I asked MS copilot to list all the living former prime ministers and it missed rishi sunak. So I said "you missed one!" and it apologized, and listed them again plus Robert Walpole 🤷‍♂️

nookrat.bsky.social•60 days ago

I try not to use it too much, but I do love it's erong but confident replies after admitting it has been caught out.

pravg.bsky.social•61 days ago

Had a patient who told me they’d done extensive research on their condition then showed me their Chat GPT results on their phone which was depressingly inaccurate.

libroloverjen.bsky.social•60 days ago

The original ChatGPT was in NO way designed to function as a search engine. It spits out answers using complex mathematical relationships between words to create appropriate sounding responses with no regard to whether its answers are correct…that’s why it hallucinates. /1

libroloverjen.bsky.social•60 days ago

The fact that journalists and lawyers, among other professionals, have been caught using it to source information is scary. They do not understand what it was designed to do and have been embarrassed more than once by sharing incorrect information. /end

toucanmacaw.bsky.social•61 days ago

It's the blithe authority with which AI speaks that is so seductively convincing. It sounds so confident, so sure. No wonder people are tricked.

gralefrit.bsky.social•61 days ago

It’s like a robot Boris Johnson.

toucanmacaw.bsky.social•61 days ago

Now I have a picture of him at a crossroads, selling his soul to Tufton Street...

toucanmacaw.bsky.social•61 days ago

Egad. What an image.

mackemfox.bsky.social•61 days ago

And uses 10x the electricity as using a regular search engine.

gralefrit.bsky.social•61 days ago

And if you choose not to use it, still runs in the background, doing the environmental damage. It’s horrific.

nasazh.bsky.social•60 days ago

That's why I like https://kagi.com approach. It only generates AI summary if you ask for it.

acidrains.bsky.social•61 days ago

All this, plus every query wastes approx 500ml of water.

puntloos.bsky.social•61 days ago

So far the only thing I found AI to be helpful for is unblocking writers block.

nookrat.bsky.social•60 days ago

It's good for practicing foreign languages too. For example, I've asked it questions about outer space in French and then asked follow up questions about the info it has given me. As long as you aren't relying on the info to be factually correct...

radiohead319.bsky.social•61 days ago

For all the hype of AI, I find it astonishing how useless the AI results are at the top of every Google search. It's just another few lines of gumf to ignore along with the sponsored results (that usually take an age to load, if at all).

They've killed their key product!

palkopalko.bsky.social•61 days ago

There are some interesting long reads on the irresolvable conflict between being both a trusred search provider and an ad service, and the ad guys have essentially won the battle internally at Google and it's best product is now dying

radiohead319.bsky.social•60 days ago

You'd think the obvious thing to do would be to have the equivalent of the "I feel lucky" button that they used to have, but something like "Hit me with some AI". That way if you feel AI might be helpful you can use it, but if you want more precise result, use the standard search.

samanthabwriter.bsky.social•61 days ago

Agreed! In fact this is why I’ve stopped using Google altogether for searching. The combination of AI and ads that I have to scroll by to get an actual result annoyed me into paying for a search engine elsewhere.

rhodri.biz•61 days ago

Yep. I wrote a thing for the Telegraph recently (behind a paywall) about Google’s current woes. Came across https://kagi.com which I now pay for and feels as fresh and useful as Google did 20 years ago. (Intriguingly, Kagi just uses Google’s index but does a better job of returning results.)

samanthabwriter.bsky.social•61 days ago

That’s the one we’re using! So far I like it.

rhodri.biz•61 days ago

It’s working on AI assistant stuff too, but crucially it doesn’t foist it on you.

radiohead319.bsky.social•61 days ago

I used Ecosia for a while. At least it does some good. I think I'll give it another go!
https://www.ecosia.org/

samanthabwriter.bsky.social•61 days ago

That one looks good. We’re currently using Kagi. I’m pretty favorably impressed so far but we haven’t had it long enough for me to know for sure yet.

gralefrit.bsky.social•61 days ago

That’s my standard search engine.

Comments

Posting Rules

Reply