Google DeepMind Diagnostic LLM Exceeds Human Doctor Accuracy (59% vs. 34%)

Summary substack article: https://aimodels.substack.com/p/googles-new-llm-doctor-is-ri...

Original paper: https://arxiv.org/abs/2401.05654

After seeing Googles recent contortions to make Gemini look like something that’s not an order of magnitude more expensive for weaker results … not sure why anyone should trust anything they say that can’t be validated directly.

Code, weights or API or it’s not real.