Daniel Francis and Jeff Huber compare pre-2023 text to "low-background steel" free of synthetic contamination
Pure data prevents model collapse from recursive synthetic outputs.
Positive users praised the founder's comparison of pre-GPT-3.5 books to pre-nuclear steel as insightful, while negative users lamented AI contamination preventing future original literary works.
No Digg Deeper questions have been answered for this story yet.
Most Activity
@growing_daniel
follow me for bangers 2.5 years ahead of schedule
gotta love those pre-2023 low-background tokens
https://en.wikipedia.org/wiki/Low-background_steel
@growing_daniel http://Book.jfg.world
@growing_daniel I can't believe we may never have another pre-gpt Twilight, or any other such great literary works where the writer hand-crafted slop with such loving artisanal devotion
@growing_daniel That’s right
@growing_daniel We can’t even write new books either because it’s gotten in our minds as well
@mauddweeb @growing_daniel Can you explain the steel thing
@coleheggi @growing_daniel Essentially all the steel is more radioactive than we want it to be now
@growing_daniel
@growing_daniel What about my 2023 Wikipedia extract
@growing_daniel banger
@growing_daniel 😂
@growing_daniel OG human slop is going priceless
@growing_daniel they taught you what to do when stuck.
LLMs don't help there.
getting stuck never got cheaper or easier.
@growing_daniel Ain’t that the truth
@growing_daniel Another banger
@growing_daniel Feed it to the machine.
@mauddweeb @growing_daniel 🤯