How an artificial intelligence (as in large language model based generative AI) could be better for information access and retrieval than an encyclopedia with a clean classification model and a search engine?
If we add a step of processing – where a genAI “digests” perfectly structured data and tries, as bad as it can, to regurgitate things it doesn’t understand – aren’t we just adding noise?
I’m talking about the specific use-case of “draw me a picture explaining how a pressure regulator works”, or “can you explain to me how to code a recursive pattern matching algorithm, please”.
I also understand how it can help people who do not want or cannot make the effort to learn an encyclopedia’s classification plan, or how a search engine’s syntax work.
But on a fundamental level, aren’t we just adding an incontrolable step of noise injection in a decent time-tested information flow?
Looking at my ChatGPT “random questions” tab and the things I’ve asked from it, much of it are the kind of things you probably couldn’t look up on encyclopedia.
For example:
“Is a slight drop in the engine rpm when shifting from neutral to 1st gear while holding down the clutch pedal a sign of worn out clutch”?
Or:
“What’s the difference between Mirka’s red and yellow sandpaper?”
Now I want to know the answer to the clutch question ☺️
If it actually worked reliably enough, it would be like having a dedicated, knowledgeable, and infinitely patient tutor that you can ask questions to and interactively explore a subject with who can adapt their explanations specifically to your way of thinking. i.e. it would understand not just the subject matter but also you. That would help facilitate knowledge transfer and could reduce the tedium of trying to make sense of something that’s not explained well enough for you to understand (as written) with your current background knowledge but which you are capable of understanding.
But on a fundamental level, aren’t we just adding an incontrolable step of noise injection in a decent time-tested information flow?
Yes.
It literally cannot
One of the ways I’ve found it to be useful so far that it can contextualise knowledge for you.
AI today in LLMs are quite juvenile and quite bonkers in image generation. They will probably get better like all information technology do (anyone remember the mobile phone? It went from popular, bad and expensive to a 20€ perfectly working door-stop).
So to answer your question, imagine an AI functioning like a personal teacher, so that when you see that pressure regulator valve, you can ask for why it works, what happens if the gas is not isotropic? How’s the reaction to pressure changes? Where is it used? Show me a simulation of it in a real world situation. Calculate which one to get ad this specific replacement. Can you 3D print one? Why was it used on steam engines, or was it? Thousands of informations that won’t fit on one page, that can be explained to you at your level too, if the teacher is smart enough.
I mean, that could be quite neat IMO.
To me the value has come mostly from “ok, so it sounds to me you are saying that…” and the ability to confirm that I haven’t misunderstood something (of course with current LLMs both the original answer and the verification have to be taken with a heaping of salt). And the ability to adapt it on the go to a concrete example. So, kind of like a having a teacher or an expert friend, and not just search engine.
Like the last time I relied heavily on a LLM to help/teach me with something it was to explain the PC boot process and BIOS/UEFI to me, and how it applied step by step on how successfully deal with USB and bootloader issues on an “eccentric” HP laptop when installing Linux. The combination of explaining and doing and answering questions was way better than an encyclopedia. No doubt it could have been done with blog posts and textbooks, and I did have to make “educated guesses” on occasion, but all in all it was a great experience.
What’s understanding? Isn’t understanding just a consequence of neurons communicating with each other? This case LLMs with deep learning can understand things.
Yeah no
