this post was submitted on 28 Jan 2025
94 points (99.0% liked)
Asklemmy
44624 readers
1184 users here now
A loosely moderated place to ask open-ended questions
If your post meets the following criteria, it's welcome here!
- Open-ended question
- Not offensive: at this point, we do not have the bandwidth to moderate overtly political discussions. Assume best intent and be excellent to each other.
- Not regarding using or support for Lemmy: context, see the list of support communities and tools for finding communities below
- Not ad nauseam inducing: please make sure it is a question that would be new to most members
- An actual topic of discussion
Looking for support?
Looking for a community?
- Lemmyverse: community search
- sub.rehab: maps old subreddits to fediverse options, marks official as such
- [email protected]: a community for finding communities
~Icon~ ~by~ ~@Double_[email protected]~
founded 5 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Don't believe the hype: LLMs are not AI. Not even close. They are in fact, much closer to pattern recognition models. Fundamentally, our brains are able to 'understand' any query posed to it. Only problem is we don't know what 'understanding' even means. How can we then even judge if some model is capable of understanding, or is the output just something that is statistically most likely?
Second, can AI even know what a human experience is like? We cannot give AI inputs in the exact form we receive them in. In fact, we cannot input the sensations of touch, flavor and smell to AI at all. So, AI as of yet cannot tell you how a freshly baked bread smells like or feels like, for example. Human experience is still our domain. That means our inspirations are intact and AI cannot create works of art that feel truly human.
Finally, AI by default has no concept of truth or false. It takes every statement in it's training data as true, unless, they are labelled individually by hand. Of course, such an approach doesn't scale well for petabytes of text data. So, LLMs tend to hallucinate stuff because again it is only giving out text that is only statistically most likely, given the input.
In short, we still don't have many pieces of puzzle that is true AI. We know it is possible because we exist, but that's about it. Sure, AI is doing better than humans in specific cases, but they nowhere close humans in understanding and reasoning.
that didn't really answer my question
I guess you are right. Think of it this way, LLMs are doing great at solving specific sets of problems. Now, people in charge of the money think that LLMs are the closest thing to an intelligent agents. All they have to do is reduce the hallucinations and make it more accurate by adding more data and/or tweaking the model.
Our current incentive structure reward results over everything else. That is the primary reason for this AI race. There are people who falsely believe that by throwing money at LLMs they can make it better and eventually reach true AGI. Then, there are others who are misleading the money men, even when they know the truth.
But, just because something is doing great at some limited benchmark doesn't mean that model can generalise it to all the infinite situations. Again look at my og comment for why it is so. Intelligence is multi-faceted and multi dimensional.
This is unlike space race in one primary way. In space race, we understood the principles for going to space well enough since the time of Newton. All we had to do was engineer the rocket. For example, we knew that we have to find the fuel that can generate maximum thrust per kg of fuel oxygen mixture burnt. The only question was what form it would. Now you could just have many teams look for many different fuels to answer this question. It is scalable. Space race was an engineering question.
Meanwhile, AI is a question of science. We don't understand the concept of intelligence itself very well. Focussing on LLMs solely is a mistake because the progress here might not even translate well and maybe even harm the larger AI research.
There are in scientific community who believe that we might never be able to understand intelligence because to understand it a higher level of intelligence is needed. Again, not saying it is true. Just that there are many ideas and viewpoints present with regards to AI and intelligence in general.
LLMs not being able to tell us what bread tastes like has nothing to do with intelligence. it's a qualia. I think you meant it cannot KNOW what bread tastes like... although I still don't understand why you'd think that's a requirement for intelligence
That’s all well and good—that LLMs aren’t AGI—but not really what’s being asked.