114
‘It’s destroyed me completely’: Kenyan moderators decry toll of training of AI models
(www.theguardian.com)
this post was submitted on 03 Aug 2023
114 points (100.0% liked)
Technology
37754 readers
475 users here now
A nice place to discuss rumors, happenings, innovations, and challenges in the technology sphere. We also welcome discussions on the intersections of technology and society. If it’s technological news or discussion of technology, it probably belongs here.
Remember the overriding ethos on Beehaw: Be(e) Nice. Each user you encounter here is a person, and should be treated with kindness (even if they’re wrong, or use a Linux distro you don’t like). Personal attacks will not be tolerated.
Subcommunities on Beehaw:
This community's icon was made by Aaron Schneider, under the CC-BY-NC-SA 4.0 license.
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
It utterly baffles me that people keep repeating this nonsense. Have you ever even bothered to try ChatGPT? At all? If you managed to make ChatGPT actually repeat existing text, congratulations, tell me how, since I never have been able to do that in months of using it. ChatGPT has no access and no way to reproduce the texts it was trained, the only thing it can successfully reproduce are shorts quotes or popular phrases ("May the force be with you") that are repeated all through pop culture. Everything beyond that it will give you as vague retelling at best. Or simply put: It literally can't "find" text, since there nothing it can search.
Same for the creativity, you can complain that the stories it writes aren't the most interesting ones or still suffer from lack of coherence when they get too long. But you can't complain that it doesn't get creative. You can throw literally any topic, item, person or whatever at it and it can weave it into a story. You can make it rhyme while doing so or turn it into haiku or talk like a pirate. And you can do so incrementally, ask it to change characters and locations in the story and it will rewrite it. And when you are out of ideas, you can ask it to come up with some.
AI discussion is starting to feel like talking to moon landing deniers, just repeating the same nonsense that has already been debunked a million times.
The difference between what a human mind does in transforming their nature and experiences through artistic expression and what the machine does by referencing values and expressing them in human language without any kind of understanding is very different. You are right that LLMs don't literally copy word for word what they find, and they certainly are sophisticated pieces of technology, but what they are expressing is more processed language or images than an act of artistic creation. Less culinary experience and more industrial sausage. They do not have intelligence and are incapable of producing art of any kind. This isn't to say they aren't a threat to commodified art in the marketplace because they very much are, but in terms of enrichment or even entertainment the machine is not capable of producing anything worthwhile unless the viewer is looking for something they don't have to look at for more than a moment or read with any serious interest of the contents. I'm interested in people using LLMs as a tool in their own artistic pursuits, but they have their own limitations as any tool does.
Give the AI a body with sense inputs, and allow those sense inputs to transform the "decider" value. That's a step in the direction of true creativity
A step closer to approximating the intelligence of a worm, perhaps. I once looked into where the line is on which anamalia were capable of operant conditioning, which I hypothesize may be the first purpose of a brain, and the line on our present taxonomic hierarchy is among worms (jellyfish do not have sufficient faculties for operant conditioning and are on the other side of the line). Sensory input being associated with decider values is still not as sophisticated as learning to be attracted to beneficial things and avoiding dangerous things because the machine does not have needs or desires to base its reactions on which would have to be trained into it by those with intelligence. I'm not saying it's impossible to artificially create a being like this, but in my estimation we are very far from it considering that we barely grasp how any brain works other than to be aware of their extreme complexity. Considering the degree of difference between a worm and a sentient human, we are much further from what we would consider a human level of intelligence.
Edit: Re-reading this it seems much more snippy than I intended and I'm not sure how to frame it to sound more neutral. I meant this as a neutral continuation of a discussion of an idea.
Yes, I use the tools every day and understand how they work. Failing to fully explain the mechanics of LLMs does not materially change the meaning of my original statement.
There's a reason that fan fiction is not regarded as true creative art that should be respected and discussed like other mediums: it;s not trying to do something new and original, the whole point is to re-combine and shuffle things around to sound and feel and look like the original work, just more, but not different in any real enough way to matter.
No true Scotsman
Neither are humans. Everything is a remix.
The only real advantage that humans have in this is that there are 7 billion of us and there is only one of ChatGPT. Everything ChatGPT produces ends up sounding very similar, since all of it is based on the same training data. With humans you get a lot more variety as each of them had their own unique slice of training data.