this post was submitted on 13 Aug 2023
1092 points (96.1% liked)

Technology

60082 readers
3879 users here now

This is a most excellent place for technology news and articles.


Our Rules


  1. Follow the lemmy.world rules.
  2. Only tech related content.
  3. Be excellent to each another!
  4. Mod approved content bots can post up to 10 articles per day.
  5. Threads asking for personal tech support may be deleted.
  6. Politics threads may be removed.
  7. No memes allowed as posts, OK to post as comments.
  8. Only approved bots from the list below, to ask if your bot can be added please contact us.
  9. Check for duplicates before posting, duplicates may be removed

Approved Bots


founded 2 years ago
MODERATORS
 

As the AI market continues to balloon, experts are warning that its VC-driven rise is eerily similar to that of the dot com bubble.

you are viewing a single comment's thread
view the rest of the comments
[–] [email protected] 11 points 1 year ago (2 children)

I just want to make the distinction, that AI like this literally are black boxes. We (currently) have no ability to know why it chose the word it did for example. You train it, and under the hood you can't actually read out the logic tree of why each word was chosen. That's a major pitfall of AI development, its very hard to know how the AI arrived at a decision. You might know it's right, or it's wrong....but how did the AI decide this?

At a very technical level we understand HOW it makes decisions, we do not actively understand every decision it makes (it's simply beyond our ability currently, from what I know)

example: https://theconversation.com/what-is-a-black-box-a-computer-scientist-explains-what-it-means-when-the-inner-workings-of-ais-are-hidden-203888

[–] [email protected] 9 points 1 year ago (1 children)

You train it, and under the hood you can’t actually read out the logic tree of why each word was chosen.

Of course you can, you can look at every single activation and weight in the network. It's tremendously hard to predict what the model will do, but once you have an output it's quite easy to see how it came to be. How could it be bloody otherwise you calculated all that stuff to get the output, the only thing you have to do is to prune off the non-activated pathways. That kind of asymmetry is in the nature of all non-linear systems, a very similar thing applies to double pendulums: Once you observed it moving in a certain way it's easy to say "oh yes the initial conditions must have looked like this".

What's quite a bit harder to do for the likes of ChatGPT compared to double pendulums is to see where they possibly can swing. That's due to LLMs having a fuckton more degrees of freedom than two.

[–] BackupRainDancer 4 points 1 year ago* (last edited 1 year ago)

I don't disagree with anything you said but wanted to just weigh in on the more degrees of freedom.

One major thing to consider is that unless we have 24/7 sensor recording with AI out in the real world and a continuous monitoring of sensor/equipment health, we're not going to have the "real" data that the AI triggered on.

Version and model updates will also likely continue to cause drift unless managed through some sort of central distribution service.

Any large Corp will have this organization and review or are in the process of figuring it out. Small NFT/Crypto bros that jump to AI will not.

IMO the space will either head towards larger AI ensembles that tries to understand where an exact rubric is applied vs more AGI human reasoning. Or we'll have to rethink the nuances of our train test and how humans use language to interact with others vs understand the world (we all speak the same language as someone else but there's still a ton of inefficiency)

[–] Freesoftwareenjoyer 2 points 1 year ago

You can observe what it does and understand its biases. If you don't like it, you can change it by training it.