this post was submitted on 22 Apr 2024
72 points (84.6% liked)
Asklemmy
43947 readers
638 users here now
A loosely moderated place to ask open-ended questions
Search asklemmy ๐
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
- !lemmy411@lemmy.ca: a community for finding communities
~Icon~ ~by~ ~@Double_A@discuss.tchncs.de~
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
The current AI boom is all based on a single paper from about 7 years ago, and has been achieved by just throwing more and more computing power at it. There has been basically no meaningful architecture improvements in that time and we are already seeing substantial fall off from throwing more power at the problem. I don't think its a given at all that we are close to the kind of disruption you are predicting.
I don't understand this deliberately pessimistic perspective I keep seeing around AI development that stubbornly ignores every other technological development in history. Even just considering the singular transformer architecture, we're still seeing significant and novel improvement. In just a couple years we've watched the technology go from basic predictive text to high quality image and even video generation, now to real time robotics control.
The transformer architecture is incredibly powerful and flexible. The notion that the basic technology staying the same is an indication of stagnation is as ridiculous as if you said the same of transistors half a century ago. Most of the improvement we see in the near future will be through recursive and multi-modal applications, meta-architechtural developments that don't require the core technology to change at all.
I see AI as something that will go the way of VR or cryptocurrencies or self-driving cars, it won't fully go away but people will realize that it is not suitable for nearly the number of use cases or improving as quickly as it was claimed it would and will sort of forget about it in most of the areas where it is not really improving anything.
AI is currently being used in both the wars OP mentioned.
Its primary use is always going to be in Surveillance Capitalism. The idea we can get nice things from it is mainly a consolation prize.
I mean yes I can now get AI to draw me a picture or write me an editorial. But meanwhile the IDF can get AI to choose people to kill and use the Wheres Daddy AI program to tell them when someone is at home so they can deliberately bomb him with his family.
So yeah it isn't much for consumers but it's not going away for use on us.
I think those use cases show how particularly bad AI really is considering how many wrong targets they have been bombing and how many bad recommendations consumers still get.
"The internet has reached the peak of its usability and will never progress much past it's current level"
This is you in 1997.
I'm not saying AI can't be disruptive. I'm saying we aren't there. The steady progress you think you are seeing is bought with increased processing power, the science isn't advancing steadily, it advances in unpredictable jumps. Because the performance gained with processing power is reaching its peak, we'll need at least another one of those unpredictable jumps for it to get to a state that will do what the comment I was responding to was claiming. It could be another 50 years before that happens, or it could be tomorrow.
Was there actually a statistical argument for that? IIRC the main argument was most people wouldn't have a use for it, in the guy's opinion.
There's stats for this. It's not certain, but "we're nearly at peak LLM" has become a reasonable guess in the last few months.
I think we need to hit the wall and start over with what we learned a few time over to really progress.
Which paper is that?
https://arxiv.org/pdf/1706.03762.pdf
Thanks for source