Meanwhile on Threads, someone is warming up a bot brigade attack on my book. So far it’s low-follower fake accounts repeating the same AI-generated scripts, expressing mild disappointment in my (highly-reviewed NYT-bestselling) book. Pretty low budget attack (mildly disappointed bots) so far. 🍿

Show Post

*ralph wiggum school bus dot gif* 🫠

Show Post

If you spend any amount of time using coding agents, it becomes obvious that LLMs are incapable of consistently maintaining attribution of any text in their context window to the correct author, including random prompts from its own system prompt.

Show Post

Given the way that LLM and LLM guardrails inherently work as statistical text completion generators, it feels weird to classify this as a security vulnerability so much as just how these things operate. Which making using them in security critical contexts a choice.

Show Post

i didn’t until i started sponsoring fundraising campaigns! i despise meta, but i’m not going to dictate Best Communication Practices to someone undergoing genocide

Show Post