Technology

Even the number is a bit misleading. First of all, anyone who has ever done LLM benchmarking knows that this isn’t an exact science, at all. You can totally get a 99% on a benchmark and fail every single task on another.

But even this particular claim is nuanced. From the original article:

But with Gemini 3, Google’s A.I.-generated answers were more likely to be ungrounded than when the system was based on Gemini 2, meaning the websites they linked to did not completely support the information they provided. In October, correct answers were ungrounded 37 percent of the time. In February, with Gemini 3, that figure rose to 56 percent.

See https://www.nytimes.com/2026/04/07/technology/google-ai-overviews-accuracy.html

Meaning that 56% of the time, users cannot even verify the information given by the LLM with the sources the LLM claims it’s using.

https://xkcd.com/2236/

And that’s a problem too. If everything ai says should be fact checked, the burden is shifted back to the consumer making the convenience or ‘productivity’ aspect virtually nonexistent. Such a cop out. Here’s your answer! *be sure to fact check. Okay so google it basically? Why the ai then? Just stupid

Everybody with some sense knows AI's can excrete total hogwash and it's answers need to be fact checked down to the most minute detail. Some people take what they get from AI's as gospel anyway, but that is a them problem.

The vast majority of people have less than zero idea of what an AI chatbot is. They're purposefully marketed to ensure it makes as little sense as possible to anyone who doesn't do a good bit of research.

People hear or read 'artificial intelligence', and picture the batcomputer or C3-P0, which were, for the most part, always 100% correct.

But Google a: calls these summaries, and b: presents them as the top search result.

Also c: gives it a colorful, more pleasant looking window to manipulate users into subconsciously prioritizing it over 'plain, ugly' results below.

Here is an example... Apparently Zelensky was a US president at one point and we all missed it.

It still seems like letting people off the hook for media literacy is a knee jerk reaction. Since the dawn of Google, the VAST majority of people who use it have just treated the first result as gospel. I don't know if scraping that same content and putting it in the same place with the word "Summary" above it is materially that different.

The core problem here is still individuals not taking accountability for their own education. I would actually argue that holding Google to the standard of somehow being "arbiters of truth" is even more dangerous. No one should trust any information presented to them by an entity that has vested financial interests in influencing consumer behavior.

That's actually a really good summary of the issue. It's the tacitly implied authenticity and "goodness of match" that being the top result implies that shifts the balance.

If they'd put a "generate AI summary of search" button to display the AI result, I the they'd be on firmer ground.

What needs to happen is:

1. an enforceable certification process, like part of FMVSS, to state "this vehicle is certified as L[0-5] self driving per https://www.nhtsa.gov/sites/nhtsa.gov/files/2022-05/Level-of-Automation-052522-tag.pdf". Put the certification on the window sticker. Have it reported to insurance.

2. Levels 0-2 cannot be advertised as self driving, even though there may be hands-free driving capability in some limited cases. The driver remains fully responsible and liable (no change to current liability rules essentially)

3. Levels 3 and 4 will be required to have shared liability with the driver and manufacturer in all conditions where the vehicle is in control of itself. This includes roads that the vehicle should be able to navigate autonomously and the driver has requested it to, but it is not for any reason.

4. Level 5 would place liability on the manufacturer solely, as there is no indicated driver in this case. This is the only one that can be advertised as "self-driving".

Yes, but at the same time can we stop marketing as "self driving cars" normal cars with a somewhat sophisticated cruise control, like Teslas, and stop pretending their "super full self driving unsupervised for realsies plus plus" is a "self drive mode"?

VPN getting set to Germany if that happens.

I was thinking the same thing.

An AI output is EITHER an original work (either as a wholly original work or as a derivative of another work), or it's not (and is thus a republication of an existing work).

If it's a republication, then Google owes a ton of copyright fees and the original publisher of whatever bit of training data got regurgitated is liable. If it's an original / derivative work, then Google owes nobody anything, but is responsible for whatever the AI outputs.

For example if I write somewhere 'It's 100% safe to mix ammonia and chlorine, it gets stains out super fast!' (note- DON'T do this, it's toxic), I'm the author of that statement so if someone does that and dies I've got partial responsibility for that death.

Same thing with Google.