Email or username:

Password:

Forgot your password?
Top-level
Jeen

@paninid I'm all for bashing Google, but in this case I'm not sure what you're complaining about to be honest. The result seems spot on given what you asked for.

40 comments
INTENTIONALLY blank

@paninid @abrokenjester

I've been running a "catch the AI" contest at the library. The goal being to gameify learning about LLMs/generative AI. It's been hit and miss in terms of participants except for one boy who enthusiastically answers every new challenge and always gets it right.

I've never actually seen the kid, but those who have say he's really happy the library is "doing something so important". It's for adults and teens. Anyhow, I like doing it, if only for the one die hard fan that fills out the slip every week without fail.

And it's also gotten people to check out some books about critical thinking, machine learning, data science, and AI on the adjacent shelf.

I keep hoping one day I'll walk in and see the box filled with slips, but it's still nice to see my one fan hanging in there 😂

@paninid @abrokenjester

I've been running a "catch the AI" contest at the library. The goal being to gameify learning about LLMs/generative AI. It's been hit and miss in terms of participants except for one boy who enthusiastically answers every new challenge and always gets it right.

I've never actually seen the kid, but those who have say he's really happy the library is "doing something so important". It's for adults and teens. Anyhow, I like doing it, if only for the one die hard fan that fills...

Jordan Biserkov

@migmit @IntentionallyBLANK @paninid @abrokenjester
It takes one to spot one. And it helps them improve their detection evasion techniques. While reassuring humans there's nothing to worry about.

INTENTIONALLY blank

@migmit @paninid @abrokenjester
We'd have bigger problems, then, because that would mean the AI has taken human form, walking undetected among us, and has decent penmanship.

INTENTIONALLY blank

@migmit @paninid @abrokenjester
True, could be three monkeys in a trenchcoat for all I know. 😉

Mostly, it's that I just don't work on that side. I'm on the grown up side. Other people have seen and interacted with the kid, so I'm reasonably certain he's neither a bot nor three monkeys in a trenchcoat. Not that I have anything against monkeys or trenchcoats.

Ricardo Carvalho

@IntentionallyBLANK woah nice! Care to share more details about it? How is the dynamics, sources, prompts etc

INTENTIONALLY blank

@rabc

That might be a lot, even for my long-winded self, but basically this gist of it is to make an article or image (or image paired with an article) and have people try to sort out which one has mistakes or "tells" that it might be AI generated.

I put a disclaimer on it that I'm actively trying to get the bots to give bad information re: the articles. Although, sometimes I don't need to, the LLMs are either working with out of date material or tend to hallucinate on certain topics.

The real and not real items set out for a certain amount of time. There are slips for each one asking which image for each round was AI or real with an open question asking people "how they knew it was AI or knew it was real".

After that I post which AI was used, the original prompt, and information about how AI works (in simple terms) and remind people to always verify sources, ask questions about what they're reading based on logic. Rinse repeat. 👍🏼

@rabc

That might be a lot, even for my long-winded self, but basically this gist of it is to make an article or image (or image paired with an article) and have people try to sort out which one has mistakes or "tells" that it might be AI generated.

I put a disclaimer on it that I'm actively trying to get the bots to give bad information re: the articles. Although, sometimes I don't need to, the LLMs are either working with out of date material or tend to hallucinate on certain topics.

INTENTIONALLY blank

@rabc
I try to use a mix of generative AIs--not just ChatGPT--Llama/Meta, Claude/Anthropic, Gemini/Google, Adobe AI "generative shape fill", DALL-E, Mid journey, etc. I stay away from political (or hot button issues) and try to stick with science/technology topics, biographies, and the like. I also set out books on critical thinking, AI, and data science. Plus I try to provide links to reputable sources discussing AI and give people ideas for things they can do to test/evaluate AIs performance. For making pictures, I suggest they have AI make signs with lettering, images involving hands, the inner workings of machines or simple machines they may be familiar enough with to notice problems. For written content, I recommend they take a topic, person, or fandom they know a lot about and ask highly detailed questions that a moderately knowledgeable person would know if they were comfortable in the topic, etc.

@rabc
I try to use a mix of generative AIs--not just ChatGPT--Llama/Meta, Claude/Anthropic, Gemini/Google, Adobe AI "generative shape fill", DALL-E, Mid journey, etc. I stay away from political (or hot button issues) and try to stick with science/technology topics, biographies, and the like. I also set out books on critical thinking, AI, and data science. Plus I try to provide links to reputable sources discussing AI and give people ideas for things they can do to test/evaluate AIs performance....

Jeen

@tomjennings @paninid if you search for "austria hungary space" a result showing a fictional description of Austria-Hungary in space is arguably just as accurate as independent factual results about Austria in space and Hungary in spcace. Ymmv.

Andreas K

@abrokenjester @tomjennings @paninid

The problem is the AI summary on top.

The results below make it more or less clear that they are about a fictional game.

The AI summary is misleading by leaving out any hint that it summarizes fiction.

And that is basically the issue with "AI" summaries. They don't get the context. They are convincing sounding text manipulation without understanding.

(Irrelevant fun fact: we did have a "real" colony in Asia for less than a decade.)

Ash_Crow

@yacc143 @abrokenjester @tomjennings @paninid it's a "featured snippet", an excerpt from a relevant page. Google has had that for years, it's not the same as the new "AI Overview" bullshit.

Niko Trimmel :veriqueer:

@Ash_Crow @yacc143 @abrokenjester @tomjennings @paninid exactly. This has absolutely nothing to do with AI and all with literacy. Click on the link and check the source. It being hosted on Fandom should give enough clues on itself that it's fictional.

Andreas K

@abrokenjester

Plesse note that humans get these clues often from subtle things like in which website we find a certain text, potentially even the URL. Or what other links are offered on the page.

So in a way the summary might be a perfect example of LLM AI art its best, being used totally wrong.

It's a bit like my personal position on computer translations: they are generally not good enough to translate something and hand it out to unsuspecting victims.

@tomjennings @paninid

Andreas K

@abrokenjester
OTOH computer translations can be extremely useful when an user is confronted with a text in a language he does not know: between knowing three context, the so so translation, searching for clarification on the web, asking clarification, you can usually get the meaning, without needing someone who knows the language.
@tomjennings @paninid

Andreas K

@abrokenjester
Now the point is, the "AI" has to work under human supervision so the human can interpret and verify its results. If you are not in the spam business, you generally care for a correct result, and stochastic algorithms are exactly incapable to provide that one guarantee.
@tomjennings @paninid

:blahaj: Why Not Zoidberg? 🦑

@abrokenjester @paninid The problem is that Google AI has repteatedly shown an inability to NOT post nonsense (or in this case from a game wiki) posts wholesale as search results.

Yes, the result on top is from a Steampunk wiki. The problem is that it is on top, and from a steampunk wiki.

jonoleth

@WhyNotZoidberg @abrokenjester @paninid if you search for a fictional concept, why wouldn't said fictional concept be the top result? Showing anything about modern Austria OR Hungary in space would be less accurate

:blahaj: Why Not Zoidberg? 🦑

@jonoleth @abrokenjester @paninid

Because

1. Not clearly marking it ON TOP as Fiction automatically spreads disinformation and anti-facts

2. I disagree. Even if the top post would just be a once sentence "The Empire of Austra-Hungary ceased to exist before space travel was made possible"

3. The search is without the "-" and without the word "empire" meaning it might as well be "Austria, Hungary In Space".

In short, facts should always be on top, ads should be obliterated, and AI sucks.

Old Man in the Shoe

@WhyNotZoidberg @jonoleth @abrokenjester @paninid

How do you wish, without AI, that Google classifies the internet into fiction and non-fiction?

The screenshot isn't even of AI. It's of the knowledge graph. Noticed it doesn't say AI anywhere on it? Google has done this for years.

:blahaj: Why Not Zoidberg? 🦑

@jenzi @jonoleth @abrokenjester @paninid

The way desktop Google does it is better at least, it shows the source on top, which is the most important thing.

As for classifying... Seeing the track record of public not exceptionally specialized AI (like in search for cancer cells) I wouldn't trust an AI classifying anything, to be honest. It would be worse than having it non-classified.

Old Man in the Shoe

@WhyNotZoidberg @jonoleth @abrokenjester @paninid Sorry you wanted the world's information to be classified into what's true and what's not true. It seemed simple when you suggested it.

:blahaj: Why Not Zoidberg? 🦑

@jenzi

Disinformation is the biggest threat to humanity as we speak.

Even if this particular search result wasn't a result of Googles abysmal AI, Google has admitted publicly that there is no way of making it stop making things up and present it as fact... and they are FINE with that.

Old Man in the Shoe

@WhyNotZoidberg Thanks for repeating that to me instead of backing your original position that Google should be the arbiter of truth.

María Arias de Reyna

@jenzi @WhyNotZoidberg No one said that in this thread except you .

What everyone here claims is that without a clear source, there is no way a user can distinguish between truth and fiction. Google now presents results as paragraphs that looks like they are valid answers but there's no way of knowing if they are true or not.

It is not Google who has to be the arbiter of truth. It is the user. Google is removing that possibility.

Old Man in the Shoe replied to María Arias de Reyna

@delawen but you can click the source… you’re being weird

EDIT: You're being emotional and rallying against AI when it's not even in use here and the source is right there in the screenshot. You're mad at me for not grabbing a pitchfork. The original post is misleading and wrong. And yes, originally someone asked for things to be deprioritized if it's fiction and that's not even trivial.

María Arias de Reyna

@jenzi @WhyNotZoidberg @jonoleth @abrokenjester @paninid
How to classify between fiction and non fiction? The same way all searching engines do: showing the source and allowing the user to research the source.

And no, Google hasn't done this for years. The fact that you are confusing the summary some search engines do taking information from reliable sources like Wikipedia with this kind of output only proves further the point that this kind of AI is not suitable/well trained for this use case

Jeen

@delawen @jenzi @WhyNotZoidberg @jonoleth @paninid I won't go into all of that, because it was all meant a bit tongue in cheek originally, but I will say that the source of the snippet is right there, in the screenshot, in big friendly letters, and with a nice link too.

Google does plenty of nefarious things, with or without generative AI. This just isn't a particularly good example of it.

jonoleth

@WhyNotZoidberg @abrokenjester @paninid this is less "Google's new AI initiatives suck" and more "Google should do way more than they ever have or even could do reliably"

Sampath Pāṇini ®

@jonoleth @WhyNotZoidberg @abrokenjester

I find it interesting that the original post didn’t mention or reference “AI” at all…that was just y’all being irritated by something that was implied 🤷🏻‍♂️

jonoleth

@paninid @WhyNotZoidberg @abrokenjester I was thinking of this post

mastodon.social/@WhyNotZoidber

but that's a fair point. Guess AI is just hot on everyone's mind when it comes to Google's many screwups right now.

:blahaj: Why Not Zoidberg? 🦑

@jonoleth @paninid @abrokenjester Heh. Definitely.

AI is NFTs on steroids, but it infects everything, not just techbro wallets.

René Hoffmann :fckafd:

@abrokenjester @paninid I don't see the problem either. Context is provided, that the information is from a fandom wiki of some steampunk world. People have been bad at "googling" all the time, but now they get even worse, as they always should check the source anyway but don't. It's a search engine, not an answer machine.

Go Up