I don’t need perfect. I need good enough
Comment on Google has sent internet into ‘spiral of decline’, claims DeepMind co-founder
Excrubulent@slrpnk.net 11 months agoOkay but the problem with that is that LLMs not only don’t have any fidelity requirement, they can’t. They are analogous to the language planning centre of your brain, which has to be filtered through your conscious mind to check if they’re talking complete crap.
People don’t realise this and think the bot is giving them real information, but it’s actually just giving them spookily realistic word-salad, which is a big problem.
Of course you can fix this if you add some kind of context engine for them to truly grasp the deeper and wider meaning of your query. The problem with that is that if you do that, you’ve basically created an AGI. That may first of all be extremely difficult and far in the future, and second of all it has ethical implications that go beyond how effective of a search engine it is.
Touching_Grass@lemmy.world 11 months ago
Excrubulent@slrpnk.net 11 months ago
Sure but if that becomes the norm then a huge segment of the population will believe the first thing the bot tells them. You might be okay, but we’re talking about an entire society filtering its knowledge through an incredibly effective misinformation engine that will lie rather than say “I don’t know”, because that simple phrase shows a level of self-awareness that eludes a lot of actual people, much less a chatbot.
Touching_Grass@lemmy.world 11 months ago
That’s already a problem. The thing j think about is what will serve me better. Google or chat AI. The risk of bad information exists with both. But an AI based search engine is something that will be much better at finding context, retiring results geared towards my goals and I suspect less prone to fuckery because AI must be trained as a whole
Excrubulent@slrpnk.net 11 months ago
Except we already know that AI lies and people in general are not aware of this. Children are using these. When you as a person have to sift through results you get a sense of what information is out there, how sparse it is, etc. When an AI word-vomits the first thing it can think of to satisfy your answer, you get none of that, and perhaps you should be aware of that yourself. You don’t really seem to be, it’s like you think the saved time is more important than context, which apparently I have to remind you - the bot doesn’t know context.
When you say:
an AI based search engine is something that will be much better at finding context
It makes me think that you really don’t understand how these bots work, and that’s the real danger.
We’re talking in this thread about this wider systemic issue, not just what suits you personally regardless of how much it gaslights you, but if that’s all you care about then you do you I guess ¯\_(ツ)_/¯
cybersandwich@lemmy.world 11 months ago
Did you read my last little bit there? I said it depends on the information you are looking for. I can paste error output from my terminal into Google and try to find an answer or I can paste it into chatgpt and be, at the very least pointed in the right direction almost immediately, or even given the answer right away vs getting a stackoverflow link and parsing the responses and comments and following secondary and tiertiary links.
I absolutely understand the stochastic parrot conundrum with LLMs. They have significant drawbacks and they are far from perfect, but then neither is are Google search results. There is still a level of skepticism you have to apply.
One of the biggest mistakes people make is the idea that LLMs and websearching is a zero sum affair. They don’t replace each other. They compliment each other. Imo, google is messing up with their “ai” integration into Google search. It sets the expectation that it is an equivalent function.