Hacker News Reader

Comment by BaseballPhysics | Hacker News Reader

Eh, we all know ChatGPT can make these same kinds of errors. The only difference between Microsoft and Google is Microsoft did a better job of proofing their marketing materials. It's genuinely amazing watching Google taking a shellacking on this when all of these LLMs have the exact same problems.

The real problem is none of these technologies should be put in front of users in a way that might give them a patina of authority or certainty. Their ability to convincingly bullshit is incredibly dangerous and I'm amazed at how blithely irresponsible these companies are in pushing this technology out in front of users before it can be trusted.

View on HN

>The real problem is none of these technologies should be put in front of users in a way that might give them a patina of authority or certainty.

Well, it seems that Microsoft has addressed with Bing in a better way than ChatGPT and Bard by providing actual web references.

Sometimes I wonder whether there is generally a mismatch in expectations. People seem to expect the LLMs to output only verified and true information, when it fact it is subject to fallacies very similar to those of a human due to vagueness inherent to written language. I found that if you think of these assistants as an extremely eager and efficient intern, expectations and reality match much better.

> Well, it seems that Microsoft has addressed with Bing in a better way than ChatGPT and Bard by providing actual web references.

Lol I guarantee you the average bing user isn't checking those references.

They're going to run a search, get a result, and use it uncritically, because the average person has no idea what chatgpt is or how LLMs can go wrong.

The people use things uncritically all the time. Many people don’t even read past a headline. Someone who won’t check their source on a search headline won’t check the source on anything else they read on the internet either, which is rife with incorrect information (either accidentally or intentionally).

Yeah but the difference is a headline is written by a human who will make human mistakes.

These AIs make weird mistakes that humans are not well equipped to detect.

No human would write an article claiming the JWST was the first to detect an exoplanet, but Bard did, and your average user won't even know to question the answer because it's written with authority.

Whatever, I guess we'll see what happens. These companies are rolling out this technology regardless of concerns about safety or ethics so now we all get to live the experiment.

I genuinely hope you're right and I'm wrong.

I don’t understand why some seem to think that not vetting your marketing materials is somehow ok when you are paid hundreds of millions of dollars and effectively have unlimited resources at your disposal.

You're missing the point.

Was their mistake dumb? Yes.

Is making an error like that ok? Of course not.

But is it the kind of error that justifies a large decline in stock price? No.

The behaviour of the stock implies doubt in Google's AI implementation itself, and my point is there's absolutely no evidence ChatGPT is any better or worse. If that was the reason for the stock decline, Microsoft and Google should both be taking a pounding because, at this point, we should assuming their implementations are equally flawed due to fundamental limitations in LLMs.

Their ability to bullshit is indeed a weak spot that should (and probably will) be addressed. However, I don't think the average person is not aware of that, nor fails to notice it.

Or put another way, how often are legacy search results wrong or misleading?

ChatGPT is incredibly useful despite it's limitations... imho.

> Their ability to bullshit is indeed a weak spot that should (and probably will) be addressed.

I actually don't think that's possible without a complete change in approach:

https://nautil.us/deep-learning-is-hitting-a-wall-238440/

https://www.newyorker.com/tech/annals-of-technology/chatgpt-...

> However, I don't think the average person is not aware of that, nor fails to notice it.

You massively over estimate the average person.

The average person has no idea what chatgpt, LLMs, or generative AIs are.

They certainly have no idea how they can go wrong.

And placing the output right in with the search results gives it an air of authority that the average person will not question.

I guarantee the average person will uncritically take anything chatgpt produces as straight truth, and it will cause problems.

Reply on news.ycombinator.com