Actually, if you read the study abstract, it says: “LLM users spent MORE time per case (mean difference = 119.3 s, 95% CI = 17.4 to 221.2, P = 0.02)”(emphasis mine)

While, the LLM augmented group did have better accuracy scores, it arguably made them less efficient!! And as Will Lockett points out, there are many other reasons we shouldn’t want experts like physicians relying too heavily on AI:

• These systems can experience outages that last hours, which could be devastating in a busy hospital (see the latest season of The Pitt)

• AI labs like Anthropic, OpenAI, and xAI are massively unprofitable, and if they go out of business, these tools could go away completely

• The systems require constant fine-tuning based on expert input; if their skills atrophy, it could ultimately reduce the performance of the models