Latest / Logosofia no Dia a Dia
Can Rival Labs Poison AI Training Data?
Your AI's recommendations are only as trustworthy as its training data. We explore the mechanics of data poisoning — how a rival lab could inject disinformation into a model's corpus, why a few hundred documents can have outsized influence, and why no one has been caught doing it yet. From Common Crawl to Anthropic's research, this episode breaks down a credible threat that could undermine trust in AI recommendations. Episode 5113 — myweirdprompts.com/n/5113 (permanent short link: myweirdprompts.com/070482)
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.