Today we worked on comments (some were toughies) from 8 readers/reviewers of our LLM architectural risk analysis (ARA) draft. BIML plans to release this work 1.24.24
"Their AI tools divert traffic that would otherwise go to the Times’ web properties, depriving the company of advertising, licensing and subscription revenue, the suit said."
Wonder why search engines are now just "telling you the answer" instead of giving you a URL? Attention capture. #MLsec meets surveillance capitalism.
In my view, the fiction authors have a much weaker case WRT fair use. When someone is looking up a news story to find out what happened, they don't care about the reporter...they care about the institution. So attention hijacking is an obvious threat. #MLsec
ML systems do not "know" anything. They simply predict the next token. And training token gathering may have stopped in 2021. Time is frozen in many generative ML systems. #MLsec
ML models are certainly biased in many well-known examples. But the bias is racism, sexism, and xenophobia, not whatever these "conservatives" believe. #MLsec