this post was submitted on 05 Feb 2025
103 points (99.0% liked)
Memes
4128 readers
105 users here now
Good memes, bad memes, unite towards a united front.
founded 5 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
it hasn't been confirmed, but it makes sense. in one of my conversations with deepseek, it explicitly told me, unprompted, that it was "chatgpt" and it was trained by openai. people told me it was just "hallucinating", but why would it hallucinate being trained specifically by openai? so yeah, they probably did the thing where they ask questions to chathpt to train their model
but as the saying goes, ladrão que rouba ladrão tem 100 anos de perdão
Even if Deepseek didn't explicitly train on ChatGPT output, there's so much of it pasted across the internet that just by scraping the web, Deepseek would've already faced that issue.
This distilling capability is super powerful. If any cutting-edge model can simply be distilled, there is literally no moat around AI. If OpenAI comes out with GPT-infinity, in a few months everyone else will just distill out an equivalent model. That means AI costs (and profits!) should drop to zero over time due to competition.
This matches both Jeremy Rifkin's concept of zero marginal cost, and Marx's predictions of the falling rate of profit.
The information wants to be free.