Pinned post
we've just released our first foundation model - open-1b. unlike models by other AI labs, which require that you trust their training process, open-1b is fully auditable and replayable.
you can take any checkpoint, re-run it using our custom kernels, and get a bit-identical result across nvidia and apple hardware. it uses compiler, kernel, and training technology we've been building for years.
if the world is going to offload more of its intelligence to machines --> you should be able to trust them
you can take any checkpoint, re-run it using our custom kernels, and get a bit-identical result across nvidia and apple hardware. it uses compiler, kernel, and training technology we've been building for years.
if the world is going to offload more of its intelligence to machines --> you should be able to trust them
open-1b: the first model you don’t have to trustView quoted post on X
Auditable training is the best defense against the future of AI we’re being warned about
open-1b is a milestone toward verifiable AI, a goal that is absolutely necessary for the future of intelligence
gensyn.ai/news/introduci…