Hacker News Reader

Comment by zb3 | Hacker News Reader

The page doesn't load, so I'll just add that we need AI that can clean up after AI.. but this might be difficult, because AI accepting AI output as input seems to be more likely to get "infected" and return the same AI crap.

View on HN

The recursive self improving AI is the pipe dream.

until it works

Only the naive trust the tech over-lords.

yes… just one more data center… one more model… one million more gallons of water… it’s going to work any second now, I swear… we definitely can get human level output without human level intelligence… i promise…

im not a sam altman hypebeast. My point is we are so early. Nobody thought todays LLM's were possible 3 years ago.

My use of LLMs has declined not increased. Some of us are way ahead of the curve of understanding what these things can and cant do, and will not do due to fundamental issues the technology faces.

Model output is already superhuman in many significant respects, which raises the question: what are you talking about?

Superhuman my arse.

Any recent LLM can converse in more natural languages than you, can write code in more programming languages than you and do it much quicker than you, can find bugs quicker than you, can extract useful information from large amounts of text quicker than you, can produce a more balanced take on many issues than you. The list goes on. Your denial of this is either ignorance or a coping mechanism. Either way, I recommend learning more about the subject and getting meaningful experience with it in order to address those shortcomings.

not to beat a dead horse but… they can’t do any of this. they can do about 20-90% of this and then hallucinate the rest, rendering the entire thing useless

I’m really not sure what your angle is here in ignoring that basic fact

https://archive.ph/qb2h8

Reply on news.ycombinator.com