On AlexaTM 20B: Few-Shot Learning Using a Large-Scale Multilingual Seq2Seq Model

Very impressive work by Amazon! Just finished reading it. This research is successfully questioning the trend towards ever larger language models. It also confirms that bidirectional training is better for many tasks than unidirectional.

[2208.01448] AlexaTM 20B: Few-Shot Learning Using a Large-Scale Multilingual Seq2Seq Model