this post was submitted on 04 Aug 2024
35 points (100.0% liked)

TechTakes

1436 readers
120 users here now

Big brain tech dude got yet another clueless take over at HackerNews etc? Here's the place to vent. Orange site, VC foolishness, all welcome.

This is not debate club. Unless it’s amusing debate.

For actually-good tech, you want our NotAwfulTech community

founded 1 year ago
MODERATORS
 

The internet has run out of training data - cadence's weblog (personal blog) https://cadence.moe/blog/2024-05-28-the-internet-has-run-out-of-training-data #AI #AIHype @techtakes @fuck_ai

top 5 comments
sorted by: hot top controversial new old
[–] Soyweiser@awful.systems 21 points 3 months ago

"I know not with what GPT-5 will will reply with, but GPT-6 will reply with 'Unfortunately as an language model I Unfortunately as an language model I I I Unfortunately Unfortunately Unfortunately model model model'" Albert GPTstein.

[–] snooggums@midwest.social 16 points 3 months ago* (last edited 3 months ago) (1 children)

They could always use AI to create more content and then train on that!

It isn't like that causes issues in nature or anything.

[–] weirdwriter@tweesecake.social 14 points 3 months ago (1 children)

I hope they do so they spend more money and so investors will get pissed off at how much money they are spending. Investors are already getting pissed, which is good! https://pivot-to-ai.com/2024/08/03/ai-investors-are-getting-the-jitters/ @snooggums

[–] snooggums@midwest.social 10 points 3 months ago

They already know that trying to use AI output to train AI causes massive issues by exponentially increasing the inherent flaws.

[–] atrielienz@lemmy.world 6 points 3 months ago

This is the thing that kills me about the "AI LLM's will just keep getting better at ______" people. Potentially it can if you feed it nothing but new high quality data. But you need humans to generate that data and allow access to it. We are heading towards a point where people aren't going to do that because they want to be compensated and AI LLM's by and large don't want to pay. Because they don't want to divide the pot so to speak. They're in it to make money, not pay people. That's what this has always been about. It's why so many of the larger ones are so worried about making themselves profitable with a paid tier and so on. And the more data that is scraped that's not quality data the worse they become meaning the other companies who can afford to use them as a service and make them profitable won't.