Alphabet shares dive after Google AI chatbot Bard flubs answer in ad | Hacker News Reader

Alphabet shares dive after Google AI chatbot Bard flubs answer in ad

reuters.com

27 pointsnico26 comments

The funny thing to me is that chatGPT frequently spews confident bullshit.

That the marketing employee at Google that greenlighted the ad didn't notice is a failure of human, not artificial, intelligence.

Did you miss the part in the presentation were they rambled on for about 30mins how their implementation doesn't make these pesky little mistakes their competition does?

They were clearly wrong I guess.

Until there's some logical component to the AI, it's just regurgitating whatever answers it was trained on, which are sometimes somebody spouting bs. How can it possibly be any different?

When will we be frequently using the term A.S. for artificial stupidity?

This got me thinking.. perhaps there will new form of SEO where someone really wants to push their false agenda, they only need to convince the generative AI with whatever it takes(more pages of same lies, more prominent sources with the lie, etc) and then they are set…

We are in for rough times.

Probably we need to first start using the term Artificial Sapience as synonym for Artificial Intelligence so that it catches up on marketing, and then we could use the "A.S." abbreviation the same way some people say DRM meaning Digital Restrictions Management.

Sapience? I take that as a jest. AI is even farther from wisdom and sagacity than it is from "intelligence"

You can always expand AI as "Artificial Idiocy"

One of the early uses of machine learning back in the ‘80s was Optical Character Recognition (OCR). Even that had character confidence levels to indicate how confident the engine was in recognizing the character from the image.

When end user tools started wrapping the OCR libraries for use, they simply drop this information and just use the text output. Due to this, the end user loses the context of what is a possibly correct recognition versus an erroneous one. Granted, there were false positive. But, if one used the confidence information, smarter UIs could be built.

The same story repeats with ChatGpt and Bard.

Some hinting to the end user on how confident the AI is with the generated output will help the user build successive prompts that converge on the right output, eventually.

What the article states but subtly implies differently is what they mean by 'after'. Event 1 was incorrect info in ad. Event 2 was shares stumbling 8% The article just says that Event 2 happened after Event 1 and not that Event 2 happened 'because of' Event 1.

This seems unnecessarily cruel of wall Street. ChatGPT makes blunders all the time.

Why did they not pre-decide what to show on the presentation? This sounds so dumb!

They did. And they still screwed up.

Unfortunately, this was a very specific fact that was difficult to catch. Nobody realized for the first day that there was a mistake.

It doesn't absolve them though. I'd have expected them to verify everything in the response before publishing it. It was more a failure on the the event organizers than the technology.

But, I find it hilarious that news outlets are being overly critical about Google's technology, rather than just the event, even when ChatGPT also makes mistakes and hallucinates.

I wish we could let go of the idea that Google or Microsoft is liable - directly culpable - for the output of these incomprehensibly complex language models. This notion will impede progress.

When they stop insisting the models are useful or correct.

They are useful though.

For throwaway space-filling verbiage that sounds like it might be close to right, sure.

I've used it for quick programming questions and found the output helpful.

Lucky you, that it didn't quote some wrong answer somewhere.

Even if it gets it right mostly, well that's dangerous in its way. When it's egregiously wrong but continues to sound like it actually knows what it's talking about, it will lead folks into error.

Then you should indemnify them against all harm caused by the models.

I love progress, and it’s an important thing, but progress can cause harm. We need to work to balance the two.

If you wave a hand and say there is no liability at all, the harm doesn’t go away. In fact, it dramatically increases. So, you’re actually just shifting the liability from one party (who caused the harm) to another party (who likely has no control over it at all).

So, while I agree that liability will impede progress to some extent, I would push back and say that to value progress at all costs is simply a mistake that doesn’t align with the values of most members of society.

They made the models, so it’s their responsibility. Maybe they should refrain from making something they can’t control.

Your point being? When damage is done, do you have any suggestion as to who should be held responsible? With normal search results, the site or user generating it is held accountable.

With liability comes responsibility, which is in my view more important than progress. Otherwise we would just progress into a dysfunctional dystopian world

You think the current world isn't dysfunction or dystopian?

Well if it is then shouldn't we try to change it? Personally, I don't want to live in a dystopia...

Reply on news.ycombinator.com