Archived link: https://archive.ph/Vjl1M
Here’s a nice little distraction from your workday: Head to Google, type in any made-up phrase, add the word “meaning,” and search. Behold! Google’s AI Overviews will not only confirm that your gibberish is a real saying, it will also tell you what it means and how it was derived.
This is genuinely fun, and you can find lots of examples on social media. In the world of AI Overviews, “a loose dog won’t surf” is “a playful way of saying that something is not likely to happen or that something is not going to work out.” The invented phrase “wired is as wired does” is an idiom that means “someone’s behavior or characteristics are a direct result of their inherent nature or ‘wiring,’ much like a computer’s function is determined by its physical connections.”
It all sounds perfectly plausible, delivered with unwavering confidence. Google even provides reference links in some cases, giving the response an added sheen of authority. It’s also wrong, at least in the sense that the overview creates the impression that these are common phrases and not a bunch of random words thrown together. And while it’s silly that AI Overviews thinks “never throw a poodle at a pig” is a proverb with a biblical derivation, it’s also a tidy encapsulation of where generative AI still falls short.
My friends would probably say something like “I’ve never heard that one, but I guess it means something like …”
The problem is, these LLMs don’t give any indication when they’re making stuff up versus when repeating an incontrovertible truth. Lots of people don’t understand the limitations of things like Google’s AI summary* so they will trust these false answers. Harmless here, but often not.
* I’m not counting the little disclaimer because we’ve been taught to ignore smallprint from being faced with so much of it
Ok, but the point is that lots of people would just say something and then figure out if it’s right later.
Quite frankly, you sound like middle school teachers being hysterical about Wikipedia being wrong sometimes.
LLMs are already being used for policy making, business decisions, software creation and the like. The issue is bigger than summarisers, and “hallucinations” are a real problem when they lead to real decisions and real consequences.
If you can’t imagine why this is bad, maybe read some Kafka or watch some Black Mirror.
Lmfao. Yeah, ok, let’s get my predictions from the depressing show dedicated to being relentlessly pessimistic at every single decision point.
And yeah, like I said, you sound like my hysterical middle school teacher claiming that Wikipedia will be society’s downfall.
Guess what? It wasn’t. People learn that tools are error prone and came up with strategies to use them while correcting for potential errors.
Like at a fundamental, technical level, components of a system can be error prone, but still be useful overall. Quantum calculations have inherent probabilities and errors in them, but they can still solve some types of calculations so much faster than normal computers that you can run the same calculation 100x on a Quantum Computer, average out the results to remove the outlying errors, and get to the right answer far faster than a classical computer.
Computer chips in satellites and the space station are constantly have random bits of memory flipped by cosmic rays but they still work fine because their RAM is special, error correcting ram, that can use similar methods to verify and check for errors.
Designing for error correction is a thing, and people are perfectly capable of doing so in their personal lives.
and this is why humans are bad, a tool is neither good or bad, sure a tool can use a large amount of resources to develop only to be completely obsolete in a year but only humans (so far) have the ability (and stupidity) to be both in charge of millions of lives and trust a bunch of lithographed rocks to create tarrif rates for uninhabited islands (and the rest of the world).