https://market-ticker.org/akcs-www?post=255473
You can teach a teen-age human how to acceptably operate a motor vehicle in 20 hours; so says the DMV. That is sufficient for said human (remember, he or she puts at best five watts of power into this task when learning it) or the equivalent of roughly two cents worth of electricity, the "most refined" (highest) form of energy we currently have -- and when licensed that individual is 99.9% or better capable of operating said vehicle successfully. The person does not need to have seen a trailer across a highway to know its dangerous; the computer which had not in the past in fact ran a car directly under said trailer, instantly decapitating the owner. It didn't know because it had never seen that particular scenario before and despite all the claims of "inference" it could not infer the danger. Proof of this was found in the driver's seat -- minus his head. He believed the claims, which is why he was watching a DVD instead of the road (proof of that was found in the form of the portable DVD player in the passenger footwell, with said disk in it.)
Multiple firms are now pointing out what I've said was indisputably true for the last several years: Yes, they're spending a boatload of money on AI in the coding and similar space but it is not translating into more shipped code features nor into lower costs. In other words they're still paying the engineers and what they've done is damage their productivity per unit of cost!
