Google stock tumbles 8% after its Bard AI ad shows inaccurate answer | Hacker News Reader

Google stock tumbles 8% after its Bard AI ad shows inaccurate answer

investors.com

154 pointsdjkivi83 comments

This article lacks any sort of reasoning or rationale to dispel the possibility that this is merely a timing fluke. Tech companies are sensitive to capital availability, even those who are publicly traded. As the fed tightens interest rates, we will see more stock dips in the short term. Additionally more layoffs across the industry were announced this week and Google, Apple, et. al. are a proxy “tech” to a huge number of people.

The biggest issue here is just the perpetuation of reductionist thinking on the part of many stock market blogs and thought leaders.

I suppose it doesn’t really matter what any human thinks about this article. I suspect it’s written to influence high-volume, automated trading systems given the other types of absolute hot garbage there is on that site.

I mean, you said it yourself - there's an HFT moment happening with negative articles being robotically written that is causing the stock to drop. If anything it's a bit of an over-correction from the positive bump GOOG had recently from the same robot articles of the initial Bard announcement.

Price action of Big Tech companies today:

AMZN: -2.2%

AAPL: -1.9%

META: -3.42%

MSFT: +0.45%

GOOGL: -8.21%

This is called "post hoc ergo propter hoc" and is near 100% of the content in mainstream financial publications. It mostly noise that reflects the interests of the authors and nothing more. Just take it for granted and move on.

There's a reason we have TVs running financial news through the work day in finance offices worldwide - but in mute.

This is why all of these media outlets use words like "as" and "after", and never "because of". They do all the "post hoc" lifting and assume that the reader will automatically add the "ergo propter hoc".

"Google stock tumbles 8% after its"

which every consumer of media reads as

"Google stock tumbles 8% because its"

This shady use of temporal sequence to imply a causal link is one of my most hated writing tactics.

"HN Reader barfs aggressively after reading yet another intentionally misleading headline!"

There's probably an algorithm out there that just writes financial news all day using the MadLib:

"[COMPANY] stock [RISES|FALLS] amid [NEWS]."

All you have to do is foreach company, look up whether the stock is up or down, and then copy the first news search result that comes in related to the company. Bam, you have a headline for any financial news site.

Yes some financial news outlets have been doing that since at least 2015.

https://www.theverge.com/2015/1/29/7939067/ap-journalism-aut...

Nonononono when two related things happen in sequence, one must have caused the other! But seriously, I can't think of any principle ignored more often than correlation != causation. It seems to be the easiest way to trick people into believing something is true without having to go through the trouble of finding real evidence.

"There was a man who sat each day looking out through a narrow vertical opening where a single board had been removed from a tall wooden fence. Each day a wild donkey of the desert passed outside the fence and across the narrow opening—first the nose, then the head, the forelegs, the long brown back, the hind legs and lastly the tail. One day, the man leaped to his feet with the light of discovery in his eyes and he shouted for all who could hear him: 'It is obvious! The nose causes the tail!'"

> correlation != causation

Thanks to that, I doubt everything else you wrote and will have to corroborate it with sources I trust more.

> is merely a timing fluke

What? Look at GOOGL and MSFT stock for the past 30 days. They follow eachother very well (as do most mega-cap tech stocks where there's nothing specific happening). Except for the last 2 days when MSFT announced a deep partnership and search integration with a leading AI LLM, and at the same time Google announced a clearly half-baked attempt to stay relevant.

> This article lacks any sort of reasoning or rationale to dispel the possibility that this is merely a timing fluke.

Why is it the author’s job to entertain hypotheticals like that?

Microsoft announced a generally available feature based on popular technology.

Google posted a gif of their chatbot malfunctioning as part of a blog post explaining that it’s not available.

Some things are kind of obvious.

This article lacks any sort of reasoning or rationale to dispel the possibility that this is merely a timing fluke.

True, and this is why I don't consider IBD a reliable source; it's basically a news outlet for people who think the WSJ leans too far to the left and has reeked of confirmation bias as long as I've known it (since the 1990s).

That said, this was a massive self-own on Google's part, like a Dad proudly unveiling a DIY project that's way better than that factory-made crap - only to have it catch fire seconds after being switched on.

Google has closed 5% down only six times in the last ten years.

Three of those where in March 2020 with the S&P 500 going down just as much if not more.

Two were following earnings releases.

The remaning one was because of concerns about big tech regulations: https://www.cnbc.com/2019/06/03/apple-google-facebook-amazon...

Or maybe that was just a coincidence and it was just a timing fluke - like going down 8% today while the rest of the market is almost flat.

> Bard wrote that the telescope took the very first pictures of a planet outside our solar system. In fact, the first pictures of these "exoplanets" were taken by the European Southern Observatory's Very Large Telescope, according to NASA records.

Sure, it's inaccurate, but it isn't worse than Google's current "featured snippets" that are so often bullshit and get meme'd.

Searching right now for "who took the first photograph of an exoplanet" gives back "NASA's James Webb Space Telescope", with a reference article about the first exoplanet photograph _by_ JWST instead of the first one ever.

They’re re-writing history to cover their tracks!

The difference in outrage is the same difference in outrage you get between a car crashing in cruise control and a car crashing in autopilot.

I just did that search on startpage.com and the second result was "NASA's Webb Takes Its First-Ever Direct Image of Distant World" at https://blogs.nasa.gov/webb/2022/09/01/nasas-webb-takes-its-...

To be fair, that's a delicate syntactic distinction to be made between "X did Y for the first time" and "X did Y for its first time". The first result was also from NASA and should be a correct result (judging by its headline).

it isn't worse than Google's current "featured snippets"

Another way to put this is that it's just as bad as the pain point it claims to solve. And though your basic point makes sense, Google's apparent indifference to quality undermining their own product launch is a huge PR disaster, much like Zuckerberg's performative excitement over Legs! in the Metaverse just highlighted how threadbare the concept was.

It shows a company that became huge by being ahead of the curve is now way behind it. Younger HN users may not appreciate the degree to which Google was the massively disruptive New Thing of its day.

Prior to its appearance in the search market, incumbent competitors all looked sort of the same, giving you a listing of semi-curated search results in a reasonable time, with different vendors having different advantages and disadvantages. Then one day, there was talk of this new thing called 'Google', what a crazy name. And it was just a search box with two buttons - 'Search' and 'I'm feeling lucky' - is this some sort of a joke website? Where are the categories? Where are the sidebars? You'd better not be wasting my time...then you'd try searching for a topic you were familiar with, and the results were an order of magnitude better than what you had become used to. It was like magic. I still remember the room and the desk where I first tried it and was amazed. And for over 10 years, they could just do no wrong - every product they rolled out was just massively superior and often so new as to have been previously unimaginable. Google Maps was astounding, and Google Street view even more so - in fact I'm still astounded by it every time I stop to think about it.

That's why having egg on their face with the Bard launch is so embarrassing. It's not just flawed, it shows a complete lack of imagination and suffers from exactly the same faults that increasingly plague its search product. Imagine if Exxon announced a new clean fuel technology and it turned out to be oil in barrels with a happy face logo.

In the later 90ies I usually relied on HotBot, which actually had better results than Google provided initially. Biggest issue, not just with HotBot, was the completely overloaded start page, which especially on modems made it slow to load. This very simple start page by Google was very intentional; for example right now it is almost 300kB, cookies and other such junk not included, i.e. based on completely different assumptions of the commonly available bandwidth (300kB start page would have been instant death spell for a search page back then). They worked hard to make the crawling + results better, and that was pretty much the final nail in the coffin for all the other options.

Headline doesn't really line up with everything else going on that affects a trillion dollar company...

It is probably some combination of the SOTU address, continued macroeconomic headwinds, and the poor/slow reaction to ChatGPT/Microsoft partnership. The wrong answer probably has SO little to do with this.

so what did their competitors move in a similar time, if its generalized market anxiety? Hint: nowhere near as much, and microsoft is actually up.

Google is a trillion dollar company but a lot of their revenue depends on their dominance in search. Google is a search/ads company, and now feeling pressure from microsoft/openai for the first time in a long time, their proposed answer was not well received because it doesn't satisfy the fundamental requirements for search.

There was also an AI event today: https://www.cnbc.com/2023/02/08/alphabet-shares-slip-followi...

The official video no longer available, but there's a copy at https://www.youtube.com/watch?v=npV4Kix7Td0 [1]

[1] https://news.ycombinator.com/item?id=34708255

The current state of these "AIs" seems like the equivalent of an automated overconfident charlatan, made more convincing with a ginormous corpus of bullshit^Wtraining-data to draw from.

There's no need to worry about being inaccurate. The answers provided by ChatGPT are usually inaccurate.

It's just funny that Google's answer to ChatGPT seems to have the same problems despite all the praise it's received from folks that got early access. It sort of reminds me of the hubbub around the Segway.

I'm imagining droves of tourists now in major cities, chatting with bot tour guides on tablets.

Tour guides often tell local legends instead of facts, just like AI.

Yeah, but ChatGPT is the new cool kid on the block everyone's been talking about, and with it being banned from school and academic work, it will increase its cool factor, and the kids will want it even harder.

Meanwhile Google is now your grandpa's search engine and Google knows it, which is why they're in full panic mode to come up with an equally cool competitor.

Ironically Google's search engine was fine up until 2015 or so, when they started to prioritize AI in their search algos. Now, of course, it's a sea of useless SEO spam. It's hard to say if their priorities are wrong, or if they're just losing the arms race against spam. Still, I would pay good money for a search engine as functional as Google in say, 2010.

It’s an arms race between content creators trying to win an audience and the algorithms trying to surface relevant content. Other trade off is that many people actually prefer the layman blogs than the technically precise information that I personally prefer.

Cool kids wanting it doesn't mean it has a profitable business model. Essaysforsale.com isn't a $billion business.

It's only a problem if potential consumers of your product expect it to be accurate and they do.

A huge advantage of search is its deniability, "just returning what is out there" or "not our fault nobody has that." These AIs are pretty exciting, but they are stepping out of that protection.

Not in reality, although perhaps in the public imagination. These models are totally dependent on the patterns they find in their input data.

They reassemble patterns into new text. If they weren't generating new text, people wouldn't be mocking all the nonsense they spew.

They plagiarize, so they can't pass the buck. Bing's version partially mitigates this, but it's still synthesizing its own text, not directly copying.

It has nothing to do with that it's just that they had nothing to show at their event

I imagine it's more likely due to the fact that the US government---both parties---has now basically declared open war on big tech, and the State of the Union was last night.

There are basically only two topics that can unite the parties: China and Big Tech.

Compare with Meta's "Galactica", which was a complete disaster: https://news.ycombinator.com/item?id=33611265

Insert obligatory "correlation is not causation" admonishment to clickbait headline writers.

That aside, it's pretty shocking that Google didn't check that the answers were correct before putting them in an __ad__.

Maybe they did using Google.

with BARD

You know that feeling you get when you send an important email and you suspect something is wrong even though you checked it thrice? And you probably had that feeling amplified when that email goes to a large email group.

So regardless of whether there is a cause and effect here (though I've never seen such fluke with $GOOGL), the main problem here might not be the technology, but the overly confident top management who publish such an ad without double checking every detail themselves, or have someone do it. Hell, have three someones, you can afford it.

It's not an elusive bug that can be hard to catch, it's not even a live demo, it's the equivalent of a false statement made in an advertisement.

It's not the trust in the AI that this ad brakes (it was never established to begin with) but investors trust in top management that act clumsily and with enough overconfidence to click the send button.

Management are so used to lying that they don't understand the concept of truth.

Bard's answer is the exact kind of sounds correct but could be wrong type of response that ChatGTP often gives.

I mean at least no one used wrong chatgpt answers in a showcase were they rumble around how much better they are than their competitors...

The bigger they are, the barder they fall.

I fail to see the value of an AI that basically just regurgitates something you could already find in a search engine, on top of just spewing out completely nonsensical garbage if it can't find something. All this AI chatbot crap just seems like a step-backward as a whole, not too different from cryptocurrencies and blockchain where it offers no tangible benefit over the status-quo, costs way more for worse performance, and is still outperformed by traditional systems.

I'm with you, but look at all the people in these threads who say things like "I just want one definitive answer, not a bunch of conflicting gobbldeygook results!" People want to be told what is true and get on with their day. Sure, the machine can't DO that, but it looks like it does. Tragically, I think this has legs.

So chatbots are the new blockchain*. Adding bot to company name for a 20% pop of the stock price.

*some people have a point that it might be a fluke

I'm surprised it caused that much of a reaction, if that indeed was the reason (which I am not sure it is). I mean, everybody knows by now these models are prone to fibbing and would invent factoids and references out of thin air. So, Google model is as good/bad as all others, and also somebody wasn't diligent enough to fact-check their own ad (which can, I guess, be taken as a sign of organizational decline but pretty much any organization of Google's side has the same problems).

Can someone explain how this is rational? We have a tiny sample of how Bard performs and people draw conclusions from it?

Because Google’s stock price was based on a lot of hype about their incredible capabilities and future growth. The contrast between the 2 presentations pop the bubble a bit. Plenty of people believed that Google could easily keep up with any thing other companies could throw its way and remain dominant in search indefinitely. This is perhaps the first time that is really being called into question.

If you read carefully, you'll notice that the article never actually claims there is a causal relationship between the ad and the share price dropping ("price dropped because of X"), just a temporal one ("X happened, and some time after that the price dropped"). This is a trick the business media pulls all the time. The reality is that they don't know why the share price changed, but reporting just on the change isn't very interesting, so they'll pick some plausible reason and run with it.

The irony is this type of business writing is going to be among the first replaced by AI (it already is in local sports), because this sort of vacuous "mention some things that relate to each other" approximation to correctness is a standard that chat bots have already reached.

I think a lot of those investment news articles already are automated.

Which is then picked up by other machines to derive what trades to do on the stock market. At what point do we decide to just shut all this energy wasting garbage down and use the machinery for more useful things?

This has already happened.

The causal implication in the title is basically made up (8% drop entirely because of incorrect example).

The way I see it, in this AI war, considering google already has an almost monopoly, they have a lot to lose and not much to win, and the opposite for the remaining players.

i imagine the stock price drop is more due to microsoft incorporating GPT than the wrong answer

We think that stock prices are based on earnings or profits. But they aren't (except for stocks that pay dividends.)

A more sophisticated way of looking at it is that stock prices are based on how people view earnings or profits. More specifically, what they expect future earnings or profits to be. From this perspective, it makes a bit of sense - if everybody thought that Bard was going to be a huge new pile of money for Google, and now it looks like it won't be, then people think Google stock is worth less.

An even more sophisticated way of looking at it is that stock prices are based on how people view how other people view future profits. So if you think everyone else is expecting Bard to make Google a boatload of money, then if the Bard demo goes badly, it may be rational to sell, even if you personally didn't expect Bard to be any good.

>A more sophisticated way of looking at it is that stock prices are based on how people view earnings or profits. More specifically, what they expect future earnings or profits to be.

Stock prices are based on how people expect future stock prices to be.

If your ad shows your brand new AI product making an _obvious_ error... that's a problem. It signals that you cannot trust anything else without actually checking...

rational? stock prices are speculative.

i'm glad that layoffs pump got wiped out, at least for now.

As people have said this isn't necessarily cause and effect, but if you can't get your new product working correctly in the controlled environment of an ad it certainly shakes my confidence that it will be working reliably in practice.

Has to be a timing fluke, the story was reported on before the close yesterday and the stock finished up fairly big.

That's why they were not releasing models - hallucination disease.

they can pat themselves on the back that they insisted on coding interviews for years, thereby losing 80% of experienced engineers who could have prevented this

Good! Hope it falls further below $97 or $95 then I will just buy some Google stock.

Perhaps Christopher Hohn (TCI hedge fund owner) is right and Google needs one more layoff in other areas, especially in DEI managers, etc at the company so that they can replace them with AI experts to compete against Microsoft and OpenAI. Seems like now is the time for them to put DeepMind to better use.

Like all the nonsense about Meta going to zero before and everyone scared to buy at $88, This is the same thing with Google with the ChatGPT hype. I would not bet against Google now.

You should have shorted Google last year. The move to short Google right now is already too late.

https://news.ycombinator.com/item?id=4101733

As an aside, I use uBlock Origin, and still I get an autoplaying video on that site. Extremely distracting...

ROFLMAO

Just yesterday I was being told on this website that Google didn’t clearly fumble this opportunity.

Yeah, Google's quick announcement of throwing $300M into an OpenAI competitor kind of reeked of desperation

This is what I thought right away upon hearing about Bard - plus stupid name. How will Google display 2 pages of ads using a perfect search engine? Literally, this is contrary to their interests.

How glorious it would be if it turned out the Google AI division is incompetent. Microsoft/OpenAI catches them with their pants down, like TikTok caught facebook. These arrogant incumbents puttering around, releasing no code, deep inefficient management structures, diversity and inclusion. How glorious for the forces of capitalism to punish them for their mistakes

They invented transformers which underpins the latest craze.

Leadership and product management failed here but the researchers did their job.

There’s a difference between pure research and releasing an innovative product

And also, with regards to "they [Google] invented transformers", from what I can tell only two of the paper's eight authors are still at Google.

(I work at Google, but this is based on public sources.)

> but the researchers did their job.

And they are still doing their job, but not at Google. Almost all of the people who invented transformers are at their own startups right now. Even OpenAI has a splinter group, AnthropicAI, the one Google invested in.

those people who created original transformer, large language models, and even this bard AI(lamda, meena) left the company

I mean, ChatGPT also lies very confidently; it's not like this is a Google-specific problem. Generative AI, perhaps, just isn't as useful as people seem to think.

Reply on news.ycombinator.com