精选理由
医疗AI新模型参数1万亿，推理比Claude Opus便宜20-100倍，性能还更强，值得一看。
该模型规模达1万亿参数，专为医疗领域设计，采用递归自我改进训练。在自有基准上，模型性能优于Opus，与Fable相当。推理成本仅为竞品的1/20至1/100。基准详情和运作方式已公开。
原文 · Ate-a-Pi
A 1-trillion-parameter model purpose-built for hea…
A 1-trillion-parameter model purpose-built for healthcare (trained using recursive self-improvement).
They say the model outperforms Opus and is comparable to Fable on their benchmark, while inference costs are 20–100x cheaper.
Here is a link to the benchmark and how it works: https://t.co/Utu6A1ttmc