Bytedance argues most AI training phases are unnecessary and backs it up with a competitive open model
Bytedance and several universities have released Axion 1.0, an open AI model that performs competitively with top commercial offerings while using only three training phases. The developers claim this efficiency is achieved through highly curated training data rather than the extensive, multi-stage compute processes typical of modern large language models. This approach challenges the prevailing industry strategy that relies on scaling training steps to improve performance.
Covered by 1 source
- TThe Decoder↗Jonathan Kemper1d ago