Part of what makes these models so dangerous is that as they become more “powerful” or “accurate”, it becomes more and more difficult for people to determine where the remaining inaccuracies lie. Anything using them as a source are then more at risk of propagating those inaccuracies which the model may feed on further down the line, reinforcing them.
Nevermind the fact that 100% is just statistically impossible, and they’ve clearly hit the point of diminishing returns some time ago so every 0.1% comes at increased cost and power. And, you know, any underlying biases.
Part of what makes these models so dangerous is that as they become more “powerful” or “accurate”, it becomes more and more difficult for people to determine where the remaining inaccuracies lie. Anything using them as a source are then more at risk of propagating those inaccuracies which the model may feed on further down the line, reinforcing them.
Nevermind the fact that 100% is just statistically impossible, and they’ve clearly hit the point of diminishing returns some time ago so every 0.1% comes at increased cost and power. And, you know, any underlying biases.
Just ridiculously unethical and dangerous.