Microsoft and Nvidia create 105-layer, 530 billion parameter language model that needs 280 A100 GPUs, but it’s still biased

Tech giants have come up with the ‘most powerful monolithic transformer language model trained to date’, but it still suffers from bias.