this post was submitted on 10 Jul 2023
351 points (91.7% liked)
Technology
59377 readers
3769 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
This is 100% true. LLMs, neural networks, markov chains, gradient descent, etc. etc. on down the line is nothing particularly new. They've collectively been studied academically for 30+ years. It's only recently that we've been able to throw huge amounts of data, computing capacity, and time to tweak said models to achieve results unthinkable 10-ish years ago.
There have been efficiencies, breakthroughs, tweaks, and changes over this time too, but that's just to be expected. But largely its just sheer raw size/scale that's just been achievable recently.
We all remember SmarterChild....right?
No, I have clearly forgotten: What was that?
AOL chatbot that did basic stuff.
Here you go.
https://www.vice.com/en/article/jpgpey/a-history-of-smarterchild
I do now!
I remember Tay
Oh flashbacks there. Completely forgot about this
Well LLMs and particularly GPT and its competitors rely on Transformers, which is a relatively recent theoretical development in the machine learning field. Of course it's based in prior research, and maybe there even is prior art buried in some obscure paper or 404 link, but if that's your measure then there is no "novel theoretical approach" for anything, ever.
I mean I'll grant that the available input data and compute for machine learning has increased exponentially, and that's certainly an obvious factor in the improved output quality. But that's not all there is to the current "AI" summer, general scientific progress played a non-minor part as well.
In summary, I disagree on data/compute scale being the deciding factor here, it's deep learning architecture IMHO. The former didn't change that much over the last half decade, the latter did.
Now as I stated in my first comment in these threads, I don’t know terribly much about the technical details behind current LLM’s and I’m basing my comments on my layman’s reading.
Could you elaborate on what you mean about the development of of deep learning architecture in recent years? I’m curious; I’m not trying to be argumentative.
Transformers. Fun fact, the T in GPT and BERT stands for "transformer". They are a neural network architecture that was first proposed in 2017 (or 2014 depending on how you want to measure). Their key novelty is the method of implementing an attention mechanism and a context window without recursion, which was the method most earlier NNs used for that.
The wiki page I linked above is admittedly a bit technical, this articles explanation might be a bit more friendly to the layperson.
Thanks for the reading material: I’m really not familiar with Transformers other than the most basic info. I’ll give it a read when I get a break from work.
Okay, I’m glad I’m not too far off the mark then (I’m not an AI expert/it’s not my field of study).
I think this also points to/is a great example of another worrying trend: the consolidation of computing power in the hands of a few large companies. Without even factoring in the development of true AI/whether that can or will happen anytime soon, the LLMs really show off the massive scale of both computational power consolidation AMD data harvesting by only a very few entities. I’m guessing I’m not alone here in finding that increasingly concerning, particularly since a lot of development is driving towards surveillance applications.
by that logic there was nothing novel about solid state transistors since they just did the same thing as vacuum tubes; no innovation there I guess. No new ideas came from finally having a way to pack cooler, less power hungry, smaller components together.