Hacker News Reader

I read (probably on HN) where someone suggested that pre-LLM data will become the 'low background steel' (pre-Nuclear-age steel with lower levels of internal radiation) of machine learning.

It seems like provenance/traceability of data sources will become important, both for the builders and users of LLMs.

View on HN

There's a no-longer-very-speculative bit of Sci-Fi waiting here about whole societies "forking" based upon the curation of their training data. Arguably social-media + bots + paid state actors means we're already there.

Isn't "societies forking based upon curation of their training data" practically what culture is?

Now, I agree that having scalable generative AI and high-throughput-high-virality mass media (social networks) brings new unspeakeable horrors like feedback loops to the mix. Interesting times indeed.

can you point to an actual paid state actor. the term seems like mostly bullshit.

Reply on news.ycombinator.com