I come to the end of another surprisingly insane week of work and finally look at the news, and what do I see? Multiple stories about the sex life and/or bladder integrity of Elon Musk (also, there is apparently a Glenn Greenwald sex tape floating around the internet, so take care out there).
Let’s think about literally anything else instead.
There’s a story by writer Noor Al-Sibai in Futurism today about a major problem with large language models (LLMs), the jumped-up autocomplete programs we’ve collectively decided to call “AI”. In order to keep progressing, LLMs need to be fed more and more training data. But they’ve chewed through most of the publicly available/easy-to-steal data out there, so AI programmers have to find new sources. First, they tried augmenting human-generated training data with LLM-generated data, or “synthetic” data, instead. They fed the machine what it shits out, in other words.
Mad Cow Disease for Chatbots (Open Thread)Post + Comments (120)