Today we’re releasing Inkling 2 Mini, a looped transformer, with open weights.
A big reason we're doing this is to let the AI safety community study looping, especially how loops affect chain-of-thought monitorability and how to interpret the computation inside loops. 🧵