What likely resonates is AI really does feel like a science experiment. There is clearly real value here. The problem is that the economic value has yet to catch up with the technological value. And yet the claims coming from AI companies have the unmistakable energy of a state-fair entrepreneur standing beside a suspicious knife yelling, “You have never seen anything like this before, it slices it dices...!”

I think Ed goes too far when he compares LLMs to garbage. He’s tried them, had a handful of bad experiences, and apparently decided the entire technology belongs in the round file. But much of what humans do is essentially trial and error with better PR: apply some logic, see what happens, adjust, try again, and continue until you eventually solve the problem.

If you can get an LLM to reliably do that, you can solve certain classes of problems dramatically faster.

Comments URL: https://news.ycombinator.com/item?id=45926224

Points: 262

# Comments: 155

All of the tooling, and processes, documentation, and code modularization that it takes us mortals to build complex things - this guy likely does in his sleep without it, until there's a problem he'll face where he can't anymore. He hasn't learned the curse of the gifted. You might have met this guy in high school - didn't need to study for his tests, aced most of his classes, then he specialized in college and ran into trouble because he never bothered to build such rigor into what he set out to achieve, it came naturally - and the longer that problem occurs the harder it becomes to learn the kinds of things that us mere mortals require.

My advice, if you can't manage him in house, manage him out

We still break problems down when solving problems that can apply machine learning. There's no single "drive the car" neural net, rather the task of driving a car has been broken down into subcomponents, sign detection, pedestrian detection/object in front of the car detection - and then there's logic that encapsulates these classifiers using them as inputs to determine how best to steer and power the cars drive wheels.

Its a bit far fetched to believe programming has fundamentally changed, at least not yet

https://www.tesla.com/videos/autopilot-self-driving-hardware...

If you use this to market your car without any caveats, don't be dismissal of some people who might read into that your car is capable of self-driving, when in reality its clearly absent of anything of the sort.

Its foolish to oversell thru marketing the capabilities of AP, as it's fairly easy to conflate reliability with observed behavior.

As long as the system mimics enough of the capabilities of full autonomy and it's oversold as such customers will underestimate the risks and become victims of an unfinished product

My concern is that it looks like Tesla is 90% of the way there to full autonomy and the way the feature is marketing will lull even engineers who know more about how these systems work into a false sense of security and end up dying as a result -- they'll trust a system that shouldn't be trusted. There isn't a good system for detecting a lack of focus especially when it won't take more than a few milliseconds to go from good to tragic.

Some number of people, N are willing to risk their lives to use autonomous vehicles they'll die as a result. It should be just as clear to person using autopilot the risks involved not misled with marketing fluff that doesn't come close to reality. Martyrs not victims