Meet BloombergGPT: A Large Language Model With 50 Billion Parameters That Has Been Trained on a Variety of Financial Data
By Aneesh Tickoo, MarkTechPost
The 2020 release of GPT-3 served as a compelling example of the advantages of training extremely large auto-regressive language models. The GPT-3 model has 175 billion parameters—a 100-fold increase over the GPT-2 model—performed exceptionally well on various current LLM tasks, including reading comprehension, answering open-ended questions, and code development. Many additional models have reproduced this performance. Moreover, data shows that huge models display emergent behaviours because their size permits them to gain skills unavailable to smaller models. A famous example of emergent behaviour is the capacity to accomplish tasks with few-shot prompting, where a model can learn a task from just a few examples. When the number of language models increases, this ability increases beyond random.