Simple Social Network

Darius Kazemi

1y ·
Public

·
friend.camp

My paraphrase of the below article: "LLM-generated content tweaked by humans is a good data source for LLMs. By 'good' we mean 'not as good as human-generated data but it's cheaper and probably good enough'. We're running a damn business here, what do you expect us to do, feed them the best possible data? Lol. Lmao."

If you're an "ML is consciousness" type person, this is treating your AIs like factory farmed livestock, feeding them their own waste product to save money

https://www.ft.com/content/053ee253-820e-453a-a1d5-0f24985258de

Craig P

@darius Great way to end up with AI kuru.

👨🏻‍✈️(Sky?)wanderer - Nicolas

@darius what’s the name for the AI version of prion? Because that’s how it happened.

Darius Kazemi

I gave a talk at FutureEverything 2016 (no video online afaik) where I said *if* we do manage to create conscious intelligent beings, I'm more afraid of us treating them as chattel slaves than I am of them enslaving us.

mhoye

@darius The Singularity crowd are out here believing that they're going to create this superintelligent being on a healthy diet of its own excrement, what they're going to get is a fountain of unintelligible epistemological-shitpost word-slurry, and you _have to know_ they're going to start drinking from their gibberish spigot and saying, I don't understand this at all it makes no sense, oh my god this machine must be _too smart_.

@darius us: oh so you're actually serious, you genuinely believe this is a new form of consciousness - so you must want it to be treated humanely and insist on its fundamental rights and freedoms as a sentient being?
them: absolutely not