> You strike me as someone who thinks like me before I started using LLMs.
Cool, except I've been using LLMs since their inception, and they're now part of my daily workflow, for better or worse. My opinion is based purely on my experience as a user over the last few years.
Besides, my point in that paragraph was specifically about the ability to judge accuracy in information and to produce accurate information, not whether AI can outperform humans in other tasks. Modern AI systems fail miserably at this, which might be acceptable for casual use, but when human lives depend on that accuracy, such as in the healthcare industry, the process of integrating them should be gradual, with a lot of oversight and regulation. We've done that in the automotive industry, for obvious reasons, and the same should be done in others.
> I think you hold doctors in pretty high esteem and I'm not sure how many other people hold doctors in as high esteem as you do.
We hold the medical profession, not individual doctors, in high esteem for good reasons. It is critical to society, and requires intense education for many years. At the end of the day, doctors are as fallible as any human, but they're required to be licensed, and we have regulations in place for medical malpractice. The idea that a computer system based on probabilities and randomness, that hallucinates most of the time, will be more reliable than the average human doctor is ludicrous, regardless of what some benchmark says.
I'm not against using AI in healthcare. Pattern matching to detect early stages of cancer is a very good use case that's been helping radiologists for many years now. But that's the key word: _helping_. My concern is that when these systems are used for medical diagnosis and advice, they will instead be relied on much more than they should be.