大型(更大)语言模型:大而不倒?
原文英文,约600词,阅读约需2分钟。发表于: 。As AI spending balloons, new research published in Nature suggests that LLMs may become less reliable as they grow. The post Large(r) language models: Too big to fail? appeared first on IBM Blog.
OpenAI获得66亿美元投资扩展大型语言模型,Anthropic也在开发更大模型。研究显示,模型变大可能导致不可靠,人们误以为更强大模型的错误模式可预测。新模型在复杂任务上表现更好,但简单任务仍有缺陷。专家建议企业先用大模型,再优化为更小更高效的模型。