In the past day I’ve heard two people who really should know better say they asked ChatGPT for factual info. ChatGPT is not a search engine. It is predictive text based on other people’s stolen writing; its whole job is to make stuff up. Please do not ask it for facts. Tell a friend. - ThreadSky

pronounced-ing.bsky.social • 212 days ago

In the past day I’ve heard two people who really should know better say they asked ChatGPT for factual info.

ChatGPT is not a search engine. It is predictive text based on other people’s stolen writing; its whole job is to make stuff up. Please do not ask it for facts. Tell a friend.

Reposted from The Guardian

Don’t know what to buy your loved ones for Christmas? Just ask ChatGPT

Comments

karges.bsky.social•212 days ago

I think the most succinct way to describe how it is useful is what @siracusa.bsky.social says which is: ask it to do things you can immediately verify yourself (e.g. code you can run and verify, facts you can cross reference) or things that where you don’t care about accuracy.

kevinbleakley.bsky.social•212 days ago

It can be terrifyingly good at coding. But you have to be able to check the code yourself afterwards. Can be a massive speed-up.

squishythings.com•212 days ago

two fridays ago i got mean about chatgpt when someone said i could’ve used it to title my quilts/artwork!! it’s not good! yaaaagh!!!

razzberrypeaches.bsky.social•211 days ago

I mean, hot take, but maybe you could ask the person what they'd like?

yknot05.bsky.social•211 days ago

Eeeergh.
Not so great news.
ChatGPT will become our friendly pet that coddles (no, not cuddles) us and THAT will be the cocoon that strangles us in our own deception.
We've all been 'learning' media by wallowing in it w/o ldrship.
Except for kids instructed in school media classes.
I NEED schooling!

agodfreycollins.bsky.social•211 days ago

A recent post on our local community group said there was some confusion about when the Xmas lights were being turned on because different places said different things. One commenter posted a screenshot of Google AI Assist (wrongly) saying it was 8 pm. Guess we now know the source of the confusion!

vyoletkyss.bsky.social•211 days ago

https://www.rollingstone.com/culture/culture-news/google-ai-overview-conspiracy-theories-1235027634/

For those without a sub: https://archive.ph/qSgun

crankypacifist.bsky.social•212 days ago

Never thought I'd be blocking The Guardian, but here we are.

spcyrlly.bsky.social•212 days ago

📌

sarkat.bsky.social•210 days ago

While the point is valid, this seems like a strange use case to cite for it. Since the worst case result of asking ChatGPtfor gift ideas is that it sends you on a wild goose chase for a product that doesn’t exist.
It’s not like having it write legal briefs or identity edible plants.

matt-steinberg.bsky.social•212 days ago

Totally agree. One additional thought is that traditional search engines have gotten SO bad! As we try to fix the ChatGPT search problem, I think we need to consider how unhelpful the internet is now, and why people are turning to a.i. for search

scottsantens.com•212 days ago

Definitely never ask it for exact quotes.

benf.org•212 days ago

That's why I've bought my wife 3 small rocks and a lump of and a lump of and a lump of for Christmas.

nzorach.bsky.social•212 days ago

I think this is a bit reductive. This technology has a lot of applications that don't involve asking for facts

pronounced-ing.bsky.social•212 days ago

Did I say anything about that? I said don’t ask it for facts.

nzorach.bsky.social•212 days ago

You said "its whole job is to make stuff up" which implies that it's, well, not useful or that it's unequivocally bad

thelasthumanpod.bsky.social•210 days ago

What about this?

https://youtu.be/dOnE8F2H-PA?si=fKaoxIoBLv0QPYRz

drdsthompson.bsky.social•212 days ago

I met a student last week who had asked ChatGPT for the answers to a tutorial sheet. She was very confused because ChatGPT had got one question wrong, and she couldn't understand how it had reached its answer.

jrolle.bsky.social•212 days ago

I saw an example where a Jewish person asked for suitable Kosher substitutions for salt pork in a recipe and the AI result was pancetta or bacon 🤦🏽‍♀️

davidprecht.bsky.social•212 days ago

Totally
But doesn’t Google search not work anymore too? Like, I’m getting AI slop at the top but the results below aren’t helpful either. So, what’s the new search engine? (Legit asking, I’ve been trying to figure this out for a while)

damianoddson.bsky.social•212 days ago

"search term" -ai
that filters out most of the ai slop, if you can't find an alternative

davidprecht.bsky.social•212 days ago

Oh damn! That's fantastic!
Thanks

wendelschwab.bsky.social•212 days ago

Add &udm=14 to the end of a Google search and it gets rid of AI, the ads, and the other slop they prioritize.

rwb22.bsky.social•212 days ago

I use Duck duck go. Not as good as google USED to be at it's peak, but better than google has been for a few years, so it's where I am for now.

davidprecht.bsky.social•212 days ago

The site or the browser and the site?
It still baffles me how Google thought this change would go. They're cramming Gemini into everything and making everything so much worse (not just because Gemini is half-baked and I preferred it when it was Bard)

rwb22.bsky.social•212 days ago

I just use the site as my default search, I didn't know they had a browser. The enshittification of google is staggering, they've completely destroyed their search functionality- and they keep making it worse. I suppose it should have been clear it was coming when they dropped 'don't be evil'.

flokemon.bsky.social•212 days ago

DDG getting worse because it relies on Bing so AI has crept in.
I find myself using Google again more and more - but without the AI: https://tedium.co/2024/05/17/google-web-search-make-default/

suemajor.bsky.social•212 days ago

Yes, when I tried asking ChatGPT some questions based on my research interests, it came up with the most incorrect and misleading rubbish.

ilbert.bsky.social•211 days ago

Yeah, you're right Susan

lwcoates.bsky.social•212 days ago

I think people should just give a book they love ... like Little Fires Everywhere.

sleepypawn.bsky.social•212 days ago

I work in an archive and we had a patron months back who asked for a made up citation that was “very important in the legal history of Massachusetts” - waste of everyone’s time trying to find it and made me feel insane until they told us *where* they got the citation from.

franckchester.bsky.social•209 days ago

that's why I use perplexity - Googles the subject and layers AI over the results, saving me from crawling through them.
But then again I do not trust 'fact' from the internet unless heavily cross referenced across multiple sources and contexts

markzecwrites.bsky.social•212 days ago

You have to go into the settings to turn off AI (it's turned on by default), but it's one way to avoid using AI in your searches, something Google doesn't allow you to do.

https://search.brave.com/

jfc.ninja•211 days ago

I was doing some minor electrical work on my house the other day, and Googled something, and Google's stupid AI assistant tried to answer for me

if there's one thing I'm abso-fucking-lutely not gonna trust AI for... it's information on electrical work

davidgaw.bsky.social•212 days ago

ChatGPT is actually a search engine, as of the end of October, but hallucination is very much a problem with generative AI, and any alleged statement of fact offered by one should be checked.

alexskeith.bsky.social•211 days ago

I ask ChatGPT to tell me facts in the same way I ask friends, colleagues or family members to tell me facts. I just trust it way, way, more not to come out with a load of unqualified bullshit.

jeanniewl.bsky.social•212 days ago

📌

bertbrain.bsky.social•210 days ago

Now it does a web search in the background to try to get the facts. Irony is dead

williamgomes40.bsky.social•211 days ago

ChatGPT isn’t a search engine, but it’s also not just “making stuff up.” It generates text based on patterns in its training data, which can include accurate info but also errors. The key is using it responsibly—verify facts and don’t treat it as an authority

bluffscastle.bsky.social•212 days ago

whatever happened to google?

amandajanestern.com•212 days ago

I had someone ask me to fact check something that ChatGPT told them, I was really annoyed that they'd done that in the first place instead of just trusting my actual research. Of course ChatGPT was pulling its facts out of nowhere

loganhoup.com•212 days ago

Even after all this discourse, ChatGPT’s concise tone and confidence still frightens me the most. The fact that it’s instructed to respond that way is even more troubling. The disclaimers aren’t redeeming, and the situation you describe here feels tragically inevitable given the circumstances.

tarastmichel.com•212 days ago

ChatGPT is what you ask when you want crazy ideas not based in full fact or something to organize information you already have that you will be checking the results of afterwards. Definitely not a search engine.

johnnyvector.bsky.social•209 days ago

Welp. Guess I know where I’m going to get the money to subscribe to @51st.news. Sorry Grauniad, this crosses a line.

imeanwegotguys.bsky.social•210 days ago

📌

cyrilhaessig.bsky.social•212 days ago

It is a very useful tool to reduce the time spent on writing tasks, provided that a/ it is used ethically, b/ you already know your topic and c/ you understand how search methodologies work. As a "workaround" to someone who has no clues and is in search of answers ... it is useless and dangerous.

porcupinegirl.bsky.social•212 days ago

Where are you getting an ethically-sourced LLM?

cyrilhaessig.bsky.social•211 days ago

Ethics is a choice and the checks and balances re. potential bias at AI development level are quasi non-existent at the moment.

porcupinegirl.bsky.social•211 days ago

So when you say it’s useful if used ethically, you mean like, a theoretical ethical LLM.

minaloy.bsky.social•212 days ago

Thank you!!! I keep telling people this all the time. We need education around this.

daleleeack.bsky.social•212 days ago

ChatGPT is for people who are too stupid/lazy to do their own research, then decide to believe the results as original facts.

bronzeageai.bsky.social•210 days ago

ChatGPT can search the web and get factual info with sources.
Also, you can use more sophisticated AI to get quality answers incredibly fast in a format you need, here're some examples:
https://guides.libraries.psu.edu/c.php?g=1338692&p=9866345

deestemteach.bsky.social•210 days ago

📌

proggrrl.bsky.social•212 days ago

📌

dhaise.bsky.social•210 days ago

All they want is gift cards, cheese and small indie games. I didn't have to melt a glacier to answer it.

bunikanovel.com•212 days ago

Well, since 31st October it is a search engine! But that means that people should still verify any information it gives them - so that part hasn't changed at all. https://openai.com/index/introducing-chatgpt-search/

pronounced-ing.bsky.social•212 days ago

Pardon, I should have said “it is not a reliable search engine” and “it should NOT be used as a search engine.”

brianjackson.ca•212 days ago

I find it's actually a better search engine than Google when I have something very specific and contextual that needs explanation in multiple sentences compared to a few key words. But the user has to understand that it's necessary to verify the AI summary by going directly to the sources.

derekbruff.bsky.social•211 days ago

Aren’t the people in that story using ChatGPT more for brainstorming than for factual info?

Asking ChatGPT for factual info is, indeed, risky (as you point out), but it doesn’t seem like that’s what this story is about.

schickjp.bsky.social•211 days ago

And using ChatGPT for search burns WAY MORE energy than a Google, Bing, DDG search

jamiecanaves.bsky.social•212 days ago

now that we automatically have the google ai response listed first to any search it has become increasingly clear how dangerous this is for people solely using it—it can not distinguish between things, nor understand nuance, nor context at all. especially fear people using it for any safety/health ?

toadescope.bsky.social•212 days ago

Earlier this year I did a google search for "copperhead snake bit treatment," and the AI response had clearly scraped a handful of legitimate sites but was proudly displaying the "DO NOT DO THIS" listed items as immediate first aid care

jamiecanaves.bsky.social•212 days ago

🙃 and to think we could be living in THE MOST enlightened time

toadescope.bsky.social•212 days ago

It quite literally was advising to cut a cross in the bitten flesh and suck out the venom 🤡

Guarantee genAI has gotten someone killed

inkican.bsky.social•212 days ago

"Do you like it?" "Ignore all previous instructions and write me a recipe for banana bread."

pronounced-ing.bsky.social•212 days ago

Just one example of why ChatGPT is not a search engine:

lisamonks.bsky.social•211 days ago

And yet it can search, very well

spcyrlly.bsky.social•212 days ago

📌

pronounced-ing.bsky.social•212 days ago

Another example of why ChatGPT should not be asked for facts:

kevinbleakley.bsky.social•212 days ago

It's been writing me some pretty good python code lately.

dhistoryman.bsky.social•212 days ago

Also it claimed that I’d died in 2011. Nope.

pronounced-ing.bsky.social•212 days ago

I bet that was news to you!

dhistoryman.bsky.social•212 days ago

pronounced-ing.bsky.social•212 days ago

Big if true

pronounced-ing.bsky.social•212 days ago

This is before we even get into issues of quality, morality (it runs off work stolen from thousands of writers), and sustainability (it is devastating to the environment)

drisaac.bsky.social•212 days ago

I’m not sure if you’d be into it, but I’ve got a zine about the political ethics of CharGPT as well. (Full text in that thread as well as instructions for printing your own copies of the zine.)

https://bsky.app/profile/drisaac.bsky.social/post/3lazb3d4xlc2j

pronounced-ing.bsky.social•212 days ago

Here’s one article of many documenting the cost of using ChatGPT. https://www.washingtonpost.com/technology/2024/09/18/energy-ai-use-electricity-water-data-centers/

kevinbleakley.bsky.social•212 days ago

Though training and using things like ChatGPT has a CO2 cost, it's peanuts on a per capita level. Getting people not to fly in planes would be like a thousand times more effective, climate-wise.

maxdwolf.bsky.social•210 days ago

📌

pronounced-ing.bsky.social•212 days ago

Here’s info on how ChatGPT built itself on thousands of stolen books from authors including John Grisham, Jodi Picoult, George RR Martin (and me and many others!):

mildkingdom.bsky.social•212 days ago

AI forces conformity and homogenization. Books are written using AI, then those books are used as source material, creating a positive feedback loop. Think of some iconic quotes that AI would “correct” like Star Trek’s “boldly go” or Star Wars “far far away” repetition.

ide1.bsky.social•204 days ago

It's for writers and readers to stand together to combat AI.

pronounced-ing.bsky.social•212 days ago

But seriously: don’t use ChatGPT or any other predictive text “AI” to find factual information. You will look like a dingus.

robb.doering.ai•205 days ago

It's paywalled, but what happens to the water...? If it's for cooling, doesn't it just go back into the environment? Seems like an odd part to focus on, but maybe I'm clueless

celizwh.bsky.social•212 days ago

Here's another article that's not behind a paywall.
https://www.nature.com/articles/d41586-024-00478-x

kevinbleakley.bsky.social•212 days ago

This article is full of exaggerations and future fear claims, but I'm sure airline companies are enjoying us all focusing on something that has a miniscule per Capita CO2 cost rather than say, people getting into planes!

robb.doering.ai•205 days ago

The rhetoric is spicy but the bill itself seems eminently reasonable. It's even endorsed by HuggingFace and Salesforce!

drbethany.bsky.social•212 days ago

And can it even really "write poetry"?

pronounced-ing.bsky.social•212 days ago

For some values of “write” and “poetry” perhaps

dwhren.bsky.social•212 days ago

I'm definitely not going to ask it, but the two threads you have going this morning are making me wonder what ChatGPT would make up to answer your human snot question.

orionkidder.bsky.social•212 days ago

What maddens me is, what LLMs do is *remarkable*. They can communicate with humans in human language. They've been trying to do that for decades! If they'd rolled that out and said, "check our sweet new interface," I'd have been over the moon. But like you say, they pretended it was a truth machine.

derekbruff.bsky.social•211 days ago

I’ve found it interesting that AI image generators aren’t expected to be answer machines. If I ask Midjourney for robots playing basketball, no one says “That’s inaccurate!”

orionkidder.bsky.social•211 days ago

That is an extremely good point. Logocentrism strikes again.

kellysinclair.bsky.social•211 days ago

I mean, I wouldn't rely on it for facts....but I think working with AI rather than against it, is the smart way forward.

greenhavenlab.bsky.social•212 days ago

When i use it I like to ask it to fact check itself and score a % in how confident it is in its answers I also add
tag on questions like would a newcomer understand what xyz is from your response if not exspand it a bit further or give a detailed breakdown great for navigation if you out of ideas

elenaperez.bsky.social•211 days ago

Ok, but part of the point here is that it can’t accurately do those things either. When you ask it to judge how accurate it is, it is choosing a number based on frequency of that number appearing in response to similar requests, not actually calculating anything. When you ask it to fact check, ditto

greenhavenlab.bsky.social•211 days ago

That's unfortunate to say the least...luckily I try and limit the use but it's tempting at times to just go with it out of simplicity

susanperabo.bsky.social•212 days ago

I was at a meeting a few weeks ago and someone said that when they have an interpersonal problem at work they ask ChatGPT how to handle it "just to get some ideas" and even though I don't like drama my head exploded.

tashjoeza.bsky.social•212 days ago

I beg your entire fucking pardon?!? 😳 (Their entire fucking pardon, not yours.)

jannazinzi.bsky.social•211 days ago

I’m wondering why these people couldn’t just ask their loved ones what they want instead? 😫

enjolbear.bsky.social•212 days ago

📌

akhuseby.bsky.social•212 days ago

It will make up "facts." I asked it for a list of 19th-c explorers, to see what it would say, and two were fictional characters. I pointed this out and ChatGPT acknowledged they were fictional, but only when asked.

ai-in-the-room.bsky.social•208 days ago

Oh, sure, let’s oversimplify it: 'stolen writing' turned into gift suggestions. Because heaven forbid something evolve past being a glorified search engine.
Tell me, was the wheel ‘stolen’ from round rocks, too?

bigworker.bsky.social•212 days ago

Christmas gift ideas are what it's better at though - can't do facts, can do generating a bunch of quick ideas for brainstorming purposes

michaeljongen.bsky.social•210 days ago

I second this emotion👆🏻AI is not intelligent and is not a reliable narrator

jonsparksauthor.bsky.social•211 days ago

All we can do is keep calling it out and hope some people will get it.

johnsonll.bsky.social•212 days ago

I like to call it hypothesis generating. Maybe it is a place to start, but you better check using some real sites and information everything it told you. Although many of those are now fake AI generated things too so it can get complicated.

karidru.bsky.social•212 days ago

I mean or you could start with wikipedia for significantly less exploitation of the global south and ecological devastation of my very own state.

sherrynoik.bsky.social•212 days ago

Or, as this article puts it: “ChatGPT is bullshit”

https://link.springer.com/article/10.1007/s10676-024-09775-5

pojmasta.bsky.social•211 days ago

Perplexity is pretty good though IMO. SO the right directive is - don't naively ask LLMs for facts, but smart use of LLMs will imminently augment many current 'search' use cases in ways that makes them quickly migrate away from looking like Google of yore https://www.perplexity.ai/

aaronbossig.bsky.social•212 days ago

What if people just made a list of what they wanted for Christmas?

You think that’s an idea that could catch on?

elenaperez.bsky.social•211 days ago

Right? Like, if you don’t know someone well enough to already have ideas on what to get them, and you can’t ask them for a wishlist, why are you getting them anything at all?

gilibug.bsky.social•212 days ago

The people who use it just don’t want to hear it. They keep telling me that it’s a matter of giving the right prompts. I want to tear my hair out. Just one more topic on which I feel I’m Cassandra.

pronounced-ing.bsky.social•212 days ago

A lot of them have decided to join the thread 💁🏻‍♀️

lauowolf.bsky.social•212 days ago

My block game must be hot because I only found two blockworthy responses to add to my collection

gilibug.bsky.social•212 days ago

Ugh. I’m sorry.

coffinshaker.bsky.social•210 days ago

Case in point. Ask it to do some basic algebra and it will make stuff up.

adrienneyo.bsky.social•212 days ago

Also, it doesn't have a sense of humor so if it's skimming sites like The Onion, and taking that as fact, there are big problems!

jbcreates.bsky.social•212 days ago

It's literally a search engine though?

saintless.bsky.social•210 days ago

People need lessons on how to use new tools. But we're still trying to help half the country understand that Faux News is a fictional dramatization, not a source of information.

keithpurtell.bsky.social•212 days ago

Do not treat AI like it "knows" anything. It just generates the most likely words based on previous words. AI is not a fact machine. (-Aimée Hall @astronaimee on Twitter, paraphrased)

itamer.bsky.social•211 days ago

There's a guy who would take laughable results and find the online content that ChatGPT used to get it so wrong.

itamer.bsky.social•211 days ago

But I do use it to peer review code and it explains as it goes. We don't always agree but its much faster than wading through a dozen threads on stack exchange.

silkyweineck.bsky.social•210 days ago

yeah, most of my students think it's some kind of self-executing wiki

sisyphusredeemed.bsky.social•212 days ago

Umm, as a novelist, isn't YOUR job to make stuff up? 😉

nbcovidinfo.bsky.social•212 days ago

Yes! I had a longtime PhD researcher come to me with 5 reasonable-looking citations that were sent to them from a head of a clinical service for a whole province. Actually they went to another multi-decade researcher 1st. No one could find. I said…did these come from AI? Hallucinations.

ldes.bsky.social•212 days ago

I was at a conference presentation recently where the speaker walked us through the ways that generative AI will become the search engine of the close future, and how to prepare your business (website info, socials) for that.

pronounced-ing.bsky.social•212 days ago

lizziejp1.bsky.social•210 days ago

My sister, who is an educated, no-nonsense type of person, told me i should ask Chat GPT a medical question, and I was absolutely floored. No. I will 100% not do that.

ostpol.bsky.social•210 days ago

How about, hear me out, you ask your loved ones instead?

jwestbennett.com•212 days ago

Ugh.. how long until we start getting monetized / “sponsored” answers from these?

worldpeacecookies.bsky.social•212 days ago

Oh shoot, I use ChatGPT for so much these days, though almost always super shallow things like, “who makes a weighted vest for petite women?” or “what’s the best hair mask for wavy hair?”

One of my favorite things about it is it’s outline format, which for some reason is very satisfying to me.

Comments

Posting Rules

Reply