Open-1B: the first model you don't have to trust

A new AI language model called Open-1B is being released with something no other lab has offered: verifiable proof of exactly how it was trained. Anyone can download a tool to replay every training step and confirm the results match published records, bit-for-bit, on their own hardware. This tackles a major trust problem with closed AI models, which can hide biases or backdoors with no way to detect them from the outside.