Thing is, chatgpt can easily answer this question correctly. So it's not an LLM issue, it's just that google has managed to combine google's horrible results with an LLM to give us the worst of both worlds.
I'd bet it's wholly dependent on their shitty results. They're basically passing it a prompt like "parse these 10 $cached_webpage_results[] to answer this $question" and since your prompts tend to heavily prime your answers it's gonna pull from the shitty included search results rather than its own training.
Was just reading the tech post about Nvidia being worth more than Amazon and Tesla combined due to hype over Nvidia AI chips. This is the value proposition that is manifesting that hype lmao.
I'm pretty skeptical of the value of these LLMs, but Google's isn't even good enough to call it half baked..
These models are ultimately crap in my opinion. They're not optimizing for intelligence or correct answers, they're optimizing for seeming correct and intelligent. All the feedback they get from users is uninformed, you only ask an LLM what you don't know.
I used to think these tech companies knew what they were doing, at least somewhat. It's just a bunch of business dipshits running from fad to fad burning tons of money until they fall backwards into a business. Then they monopolize it and enshittify it.
Google's isn't even good enough to call it half baked
Google has a problem where they let OpenAI get way ahead of them in terms of institutional knowledge and product capability, and with the number of people who really understand how these LLMs work being countable on one hand, no amount of money they throw at the problem can bridge that gap.
The smart thing to do would be to develop slowly and not release anything until they have something good - but they have to show something to their investors, so that's not an option. They have to release something in a short amount of time, and they simply don't have the capability to release something on that timetable that doesn't suck.
i was trying to look up strategies or solutions for trunk organizing, specifically my older-but-once-common model of vehicle. there are of course a gazillion hits with fake reviews/recommendations for crap products to sift through, but now that LLMs have been around for a minute, there are fake articles that state and restate the topic over 500 words saying nothing substantive.
eventually i realized that the quickest way to sort through answers is to do an image search of the question, because my eyes could quickly find a relevant picture which would take me to some old vbulletin forum where people discussed this topic back in the world before the internet became a huge pile of revenue-seeking shit wizardry.
Google isn't using their most expensive inference requests on serious questions from the wealthy decision makers of hexbear dot net, so you see, this means this technology can never produce good results. This also means my shitty job is and will always be completely safe from being automated. It's all a bunch of grifters and nothing to be concerned about!