Google trained a trillion-parameter AI language model
For example, OpenAI’s GPT-3, one of the largest language models ever trained at 175 billion parameters, can make primitive analogies, generate recipes, and even complete basic code. They say that their 1.6-trillion-parameter model, which appears to be the largest of its size to date, achieved an up to 4 times speedup over the previous-largest Google-developed language model (T5-XXL). The Switch Transformer builds on mixture of experts, an AI model paradigm first proposed in the early ’90s. This being the case, the Switch Transformer led to gains in a number of downstream tasks. Unfortunately, the researchers’ work didn’t take into account the impact of these large language models in the real world.