Google's AI search bot Bard makes $120b error on day one About 10 percent of Alphabet's market value – some $120 billion – was wiped out this week after Google proudly presented Bard, its answer to Microsoft's next-gen AI offerings, and the system bungled a simple question. In a promotional video to show off Bard, a web search assistant to compete against Microsoft's ChatGPT- …
The so called AI was already exposing its weakness by not being actually intelligent. But now it screams it loudly into the web: "I am dumb, and I am proud of it!".
Pattern recognition is, up to now, the only thing that somehow works, though actually useful in niche where the training data is very good. Known best example is cancer (and similar) detection, or LSD simulation where "detect eyes" or "detect ducks" training data is applied to normal videos.
Reproducing something out of an input is the current limit as those picture creators show - but inherently they are stealing from source pictures. It is still a pattern search-match-and-mix, but calling it AI is quite a stretch.
As for text: Nice for playing, but the results show no actual intelligence, exposing the weakness of AI. Showing that there is no intelligence, only pretending to be so. Something we see a lot in humans as well. The leftpondians suffer from the fact that "free speech" is abused to spread bullshit on a level that many gave up on expecting truth from anyone - something where Germany has to suffer less from thanks to some enforced laws after World War 2.
Maybe it is time for a US lawsuit that companies do not have the right of "free speech and opinion" as humans do. Reverting a famous "Fox News" lawsuit where they, practically, won the right to spread lies of the worst kind under the cover of "free speech and opinion". There are several Southpark episodes where, as usual, Eric Cartman exposes that logic.
Pretty good summary -- these things are pattern matchers, and surprisingly good ones when fed enough data. But they're also quite fragile when fed less than immense amounts of data, as you can never tell exactly what features their recognizers have glommed onto. In all likelihood, you will be surprised sometime in the future when they fail on something obvious.
ChatGPT is worse. It looks smart, but it doesn't really have a good model for the world, so it often ends up just making stuff up. And it doesn't even know that's what it's doing. Bummer that Google's demo screwed up visibly, but really, this stuff isn't ready for prime time with either Google's or Microsoft's technology. Not even close.
Are the developers of these models aware of the old saying 'garbage in, garbage out'?
It takes ten or more years to train a human to be even vaguely critical of what it is being taught - and even then, some of them never make it... why would one expect statistics to do any better?
I had a go with chatgpt:
Jamie Landeg Jones (Welsh Git)
What's the speed of gravity?
The speed of gravity is the speed at which a gravitational field or gravitational force travels through space. It is considered to be infinite, meaning that it travels instantly and simultaneously to every point in space. This is because gravity is a long-range force that acts between massive objects, such as stars and planets, regardless of the distance between them. The speed of gravity does not depend on the distance between objects or on the time it takes for the field or force to travel from one object to another. Instead, it is an inherent property of the force of gravity itself, which always acts instantly and uniformly throughout the universe.
Jamie Landeg Jones (Welsh Git)
That can't be correct. It most likely travels at the speed of light.
Yes, you are correct. The speed of gravity is generally considered to be equal to the speed of light. This is because gravitational forces and gravitational fields travel at the speed of light. This means that if two massive objects were suddenly to change their position relative to each other, the change in the gravitational field would propagate through space at the speed of light, and the change in the gravitational force between the objects would be immediately reflected in their motion.
It is worth noting that the exact nature of gravity and how it is related to other physical phenomena, such as space-time and the fabric of the universe, is still a subject of ongoing research and investigation by scientists and physicists. Despite the progress that has been made in our understanding of gravity, many aspects of it remain mysterious and the subject of ongoing debate and study.
Jamie Landeg Jones (Welsh Git)
Didn't you just contradict yourself?
Yes, I apologize for the mistake in my previous answer. The speed of gravity is indeed equal to the speed of light. Thank you for bringing this to my attention.
It seems pretty clear that we will realize exactly when they turn it on based on likely hilarious and embarrassing impact on the output.
Hilariously these results were entirely foreseeable, but the people on these projects can never seem to find the brakes even when they can see the cliff they are about to go over.
Sometimes the bubble just has to pop, and going into a recession seems as good a time as any. Fake AI has been the king of the hill for the valley's self delusions for a couple years now. Blockchain is no longer sexy, time to shuffle on to the Next Big Thing.
Funny though that this is Google's response to a challenge to their ad dominance. They have been on a decades long project to return the crappiest search results possible, destroying their own technical lead and making their search results as bad as the competition. All Bing needed was to re-implement Google's search results circa the mid aughts, and they could start clawing back market share.
This post has been deleted by its author
I've asked ChatGPT for fancy poetry and programming tasks. Very impressed. Especially when I'm rusty, I'd be struggling with quotes and comas in the right places and so on. But what it gives isn't a finished thing. If it's given you say a cool bash script in one minute then most likely you'll go 'Oh. Maybe I should have asked a slightly different question.' Then with the slog done, I'm free to manually upgrade that work to a finished thing. Just because 'It's that magic AI' doesn't mean there's no need to check and test.
The poetry is at the 90% of humans would struggle to do so badly. ie. Tedious verse but definitely a good first attempt. AI is also good at suggesting ideas in some sort of structure. "Draw me a fox on a horse in the style of a woodcut" gives half a dozen suggestions in a minute or two. That's planted a few creative seeds I can explore as I hit the ground running. AI isn't like a coffee machine where you press some buttons and you get the finished cup, it's a recipe book where you start from. Use your own parameters to adjust to your needs and add your own special sauce.
This post has been deleted by its author
And all apologies to fans of Queen and the best comic book movie ever made*.
(* No, seriously- It's over the top, there's COLOR instead of 50 shades of brown**, and enough wonderful scenery that even Brian Blessed (Giant Ham (WARNING: TV TROPES LINK)) couldn't fully chew, and lots of cheese to go with the ham. A fantastic movie for just turning one's brain off and watching the spectacle. )
(** I would have liked the 2013 movie Man of Steel better if it had been in color.)
Judging from the places I've worked, it's more likely to have been:
Manglement: "Microsoft have just announced AI-based search. We need to demo the AI search prototype you're working on"
Devs: "It's not ready to be shown publicly yet. There's still a lot of issues with it"
Manglement: "Demo it anyway"
"If Microsoft and Google are to use AI chatbots as the new user interface for web search, they better make sure their technology generates factual up-to-date information if they want people to use it."
Was that sentence written by an AI chatbot? Because it seems to be chock full of misinformation. (A) They can't make sure. (B) They will each roll out their chatbot anyway, because competition. (C) People will use it even knowing it is not 100% reliable, because it looks reliable _enough_. People put up with crap search results today, they will put up with crap AI tomorrow.
It'll probably go the same way voice input in day to day use, touchscreens on big laptops, personalised adverts , sponsored Google search results and 3d TV sets went. Only a handful of people actually find them useful.
Actually, for personalised adverts and Sponsored Google search results make that no-one.
I guess a lot of people who want easy answers (and don't necessarily care about accuracy or even correctness) may be impressed. When I search for information on the web I don't look for an answer, I look for references. I will figure out correctness and accuracy myself then.
I guess from my PoV substituting an answer for references may well be worth $120B less in market value. It's not often that the markets agree with my assessments of value, but when it happens it gives me a warm and fuzzy feeling inside.
Oh please, don't you yet realise how parasitic ambulance-chasing market makers and traders turn a profit after manufacturing and factoring in a contrived inconsequential disaster.
Pick a valuable stock and wait for a novel opportunity/new beta product
Highlight and criticise an anomaly during a session of valuable stock sale.
Stock price drops.
Anomaly then realised to be human error related so deprecated stock bought back at lower price, which delivers a market maker's/trader's profit, and stock price rises again because of returning interest and stock market maker and trader support.
That's the wash and churn. Rinse and repeat and there you have it, the ponzi stock and dodgy financial markets in a nutshell.
Only 2 weeks ago Brady Haran's "Sixty Symbols" showed ChatGPT being challenged with Physics questions with less than perfect results:
https://m.youtube.com/watch?v=GBtfwa-Fexc "ChatGPT does Physics - Sixty Symbols"
I think that chap was a bit generous with his marking as well!
BTW should one assume that Google had done a rehearsal of this question before capturing this sample output? So either they didn't bother checking the answer first time (!) or they hadn't noticed that asking the same question twice gave different answers! Either way, a remarkably stupid thing to release in a demo!
Did they just ask Bard if it felt it was ready? "I have great enthusiasm for this mission".
I fairly well remember writing in these august pages, about a slightly different matter: "When someone tells me something on the Internet, I ask myself, 'Who is telling me this?' and 'Why are they telling me?'" Now, a Google™ search, while laden with false drops and adverts, does answer both those questions - it shows me exactly which sources it thinks match my query, and demonstrates very easily that I have to dig around a bit to get past the sponsored stuff and SEO to find things that (based on the words in it) are about the topic I want. As far as I can see, Bing and Bard, in serving up Large Language Model output, are going to be unable to say what the sources are, or why the Allegedly Intelligent back-end thinks that its output is relevant. That they can be simply inaccurate is deeply unsurprising.
"The above error somehow made it past Google's various engineering, legal, PR, and marketing depts, and found its way into a demo of Bard, right when issues of accuracy and trust are at the top of everyone's minds."
The level of incompetence is mind-bending..I thought at first this was a live demo, but in fact it seems to have been a pre-made video presentation, thus implying that all of the above had a chace to review it before it was published. And not a single one of them felt the need to sanity-check the information??? Youn can't make this stuff up.....
"The above error somehow made it past Google's various engineering, legal, PR, and marketing depts, and found its way into a demo of Bard, right when issues of accuracy and trust are at the top of everyone's minds."
Would this not imply that someone in some department is suspected of tampering with BARD output to make it look legitimate?
"The level of incompetence is mind-bending..And not a single one of them felt the need to sanity-check the information??? Youn can't make this stuff up....."
Donald Trump meets Max Headroom.
This tech should be able to concoct absolute doozies of conspiracy theories. (As Hazel Burke might have put it.)
So you were there when support.microsoft.com, from a perfect way to find service packs and updates, got changed to something which found nothing unless you used google with site:support.microsoft.com ?
Before that change "office service pack" and so on always led to the newest expected update, and the second and third were the previous service packs. After that change: Fail. Still fail until today, unable to catch up. (Personal opinion: 'cause doing it half-assed does not work...)
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
