Even with much better AI than today*, a human doctor is still a second opinion — and in a better way that just asking the AI twice or even asking two different models.
Hosted AI with specific version numbers are, by corporate preference, memetic monocultures: even when they're generally better, the failures are correlated.
* despite how generally impressed I am by SOTA models, I still suspect this story has been cherry-picking successes
Hosted AI with specific version numbers are, by corporate preference, memetic monocultures: even when they're generally better, the failures are correlated.
* despite how generally impressed I am by SOTA models, I still suspect this story has been cherry-picking successes