Using an AI generation tool does not train it. Training is completely distinct from inference. In fact that's why those things could never be conscious, they can't actually learn from experience lol.
0 Boosts
0 Likes
@lanodan@queer.hacktivis.me @chjara@akko.tuxcrafting.xyz Many chatbots are stateful, and agent frameworks are specifically designed to manage a large number of states... which may be confused with "training" by naive users.
@niconiconi @chjara Oh my base aren't users, I take those as dumbasses falling for the Eliza effect.
My base are interviews with LLM makers, there's been quite few since.
My base are interviews with LLM makers, there's been quite few since.
@chjara Plus I wouldn't be surprised that when you give a URL, your machine serves as proxy, cheap way to get a residential proxy which they could even claim as legitimate.
@lanodan I don't know if using it as a chatbot actually provides that much learning material, that wouldn't just reinforce whatever instead of providing useful learning. There's enough human to human conversations on the internet to scrape and use anyway.
@chjara I don't take it as straight up raw material, but more as raw feedback for adjusments.
Similar to like users dropping email in spam folder vs. archive/replying. Even though we have decades of email archives.
Similar to like users dropping email in spam folder vs. archive/replying. Even though we have decades of email archives.
But the AI companies do inspect the logs, so effectively still providing them material.