There is still some room for improvement where trying to introduce LLMs into chaotic codebases and whatnot. LLMs continue to struggle there. Although so do humans, to be fair. Chaos is plain hard. However, in a lot of cases AI is already as good as it can get. "Won't improve much" is a fair take.
Technology has plenty of room to keep marching forward, but the next substantive improvement here will see it no longer be AI. It will become AGI.
“The citation seems to have been included in many lower quality papers—likely due to laziness and sloppiness rather than an intent to deceive.”
Yeah … no. If you use a citation and you didn’t read the article yourself then it is absolutely intentional deception, and it should be treated as such.
One helpful data point is that only about 20% of people over age 12 report any THC use at all in the prior year. Some surveys have even lower numbers, around 1 in 8, but let’s take the highest number for the sake of this comparison.
So the median THC level is 0%.
Having 40% of people register high enough levels of THC to pass an impairment threshold is a remarkably high number no matter how you look at it.
I think there is definable a connection between cannabis use and auto accidents. It slows your reaction time and that’s a known factor with accidents. That said, substance use data is notoriously underreported[1] in surveys. So that 20% data point is not very helpful. Also 12-15 year olds are bringing that number down in your data and also can’t drive making it even less useful for comparison.
And substance use impairment is overreported. If the driver was impaired it's counted even if the impairment has no bearing on the accident. A drunk hits a red light runner--it's called alcohol even if he had no hope of avoiding it.
My understanding of stoned drivers is they tend to be too conservative--waiting for the stop sign to turn green etc. If that's accurate it could also mean stoned drivers are worse at avoiding the mistakes of others.
Say what?
reply