Lucy Farnik

🚨NEW PAPER ALERT 🚨 SAEs can give us insight into the representations of LLMs. But what about the LLMs' computations? If we want to understand LLMs, we don't just need sparse SAE activations, but also a sparse computational graph connecting them. So how do we get them? A 🧵

"I'm at my limit" emotional or claude?

@So8res “X is a problem but Y is worse and I don’t know how to address X without exacerbating Y” seems like a coherent position?

AI execs when talking about the danger vs the exact same AI execs talking about how we should respond:

@zudasworld @ESYudkowsky I'm confused, if an LLM thinks that you're wrong about something, do you want it to push back or do you want it to be sycophantic? I want the former, I explicitly have that in my system prompt, and Claude 4.5 has been much better than most models at following that instruction.

The most important news from today's reshuffle: