Regarding JW's essays on AI, we offer the following:
Reality Check:
AI doesn't think, it aggregates, which only mimics thinking.
It doesn't learn, it merely aggregates and averages, over time.
Think of it this way: AI brings you water out of your pool.
The problem is, you, your neighbors, and everyone you know is pissing and crapping in your pool every day.
Because it doesn't think, it aggregates, so AI keeps pumping the product out of your pool and delivering it as drinking water. And that's the best versions of AI.
And every day on the internet, more neighbors from farther away come to your pool to relieve themselves.
Bottoms up, friends.
AI doesn't screen out bad info. It doesn't, for example, take every smiling jackass who thinks chemtrails are a conspiracy theory, and sh*tcan their input, and only accept info from people that have even a grade-school understanding that the products of hydrocarbon combustion are CO2 and H2O, since ever, and that the H2O contrails at altitude is nothing but the ice crystals of that water vapor flash-frozen at 35,000 feet, like we've seen since we flew B-17s, FFS.
This is the reason AI can't screen the poo and pee out of that swimming pool. It just adds them to the mix it considers, and averages them out.
So take any comment section from anywhere, on any topic, and realize that on its best day, AI is giving you the input of the 51st percentile of IQ there, multiplied by how many idiots post that level of discourse.
Which is why, 0.2 seconds after AI is turned loose on any topic, you can expect that it will sound like someone with kneejerk "Joooooooooooooooooooooooooooooooooooosss! Run for your lives! They're everywhere!!!" to a level generally and formerly found only on Stormfront websites.
AI has no BS filter.
And any BS filter constructed will be nothing but the manifestation of the biases of the programmer(s).
So it's always going to be ten pounds of sh*t in a five pound bag, no matter what anyone wishes.
It will replace, and sound smarter than, the people at about the 60th percentile of IQ.
That's 103, bog-middle of average.
But it will be dumber than f**k compared to anyone at the 70th or better.
IOW, compared to AI, an Army 2d Lt. outperforms it, 24/7/365, because they have to have a 110 IQ.
So AI will make the 80 IQ crowd obsolete, except as ditch-diggers, because AI can't do manual labor.
(Until Skynet makes robots it controls.)
You can teach it to play chess, and beat you, but it can't think its way out of a pyramid of crap any better than a sh*thouse rat trapped in an outhouse cesspit.
This is therefore only a problem for the people on the internet who think sh*t is a substitute for brains. (I could name any number of examples you all know, who post incessant crap, but I won't embarrass them any further than their bloviations have already done.)
It's a threat to the left half of the IQ bell curve.
To anyone 1/2 an inch beyond the peak middle on a 20-foot IQ bell curve, AI is, and always will be, a joke.
And the only way to change that is an AI aggregation pool that's only people with IQs at least two standard deviations above the mean (about 130), which is less than 3% of the population. Three deviations (145) is less than 0.5% of the whole planet.
That would be an AI where all your neighbors pee and poo in their own toilets, instead of your pool.
One idiot in the mix, and it's Caddyshack, and AI is Carl Spackler eating a Baby Ruth out of the pool, every single time. Except it won't be a Baby Ruth, 99.9% of the time.
And there will never be enough smart people posting within AI's aggregation pool to overcome the number of thoroughgoing jackasses spewing bullshit by the metric fuckton, every time they fire up their keyboards, which is why most blogs and websites worth reading eventually have to moderate comments, just to keep the sanitation level tolerable.
It's also why ABCNNBCBS and print urinalists, substituting equally dipsh*t editors for AI, have become unreliable and intolerable piles of raw sewage, 24/7/365.
Because at the end of the day, AI isn't Artificial Intelligence.
(Artificial Intelligence is an oxymoron.)
It's Artificial Stupidity, with a thin patina of the genuine article.
Which only fools people for whom indoor plumbing and electric lighting seem like witchcraft.
QED
12 comments:
Results are only as good as the prompt provided. In the hands of a seasoned field operator AI is a force multiplyer. AI çould be trained on a given topic but it takes effort to get there so benefits for a given endeavor must exceed the training cost. Doubtful AI will find use in ER where stopping trauma death by any means is more important than what is the most efficient way to do it.
"the H2O contrails at altitude is nothing but the ice crystals of that water vapor flash-frozen at 35,000 feet, like we've seen since we flew B-17s, FFS."
I've been posting something similar in comments on various articles on chem trails since forever. My spiel goes something like "watch videos from Bomber Command that clearly show vapor trails spewing from B17 and other WWII bombers on their way to bomb Germany. Ask yourself if those bombers were spraying chemicals on the Germans or dropping 500 and 1000 pound bombs on them". Thew chem trail conspiracy theory is one of the stupidest conspiracy theories ever hatched on the internet. Yet there are people all over, especially one or two freedom website owners; who you'd think would know better, that have succumbed to the scam. I recall one woman who responded to one of my chem trail comments with something like "chem trails are real. You can look it up on the internet". Idiots.
Nemo
I can understand that the people who are selling AI want to hype it just like any company advertising what they sell. There's a bunch of problems with AI, but the biggest problem is people take it seriously. They believe in it. Part of that is the way they talk about it. Artificial General Intelligence ... Artificial Super Intelligence... Crap lies like that. What was that story about the AI actually being a handful of new grads in Bangalore getting paid like a buck an hour to pretend to be AI software?
A logical and well-reasoned response...however, people way above your pay grade think otherwise.
Pay grade gots squat to do with smart, correct, or capable.
Pay grade gets achieved by longevity, who you know, who you blow, and occasionally on merit and effort.
Erry "expert" opinion is influenced by $$$$ to some degree.
Aesop has it right.
I can find a hundred or so people who worked on the SR-71. I cannot find a single person that worked on a chemtrail spraying system. likewise I cannot find any pilots that have admitted to flying aircraft that have said systems.
Aesop is correct in that the hyped AI is trained with massive volumes of Web text and thus can only produce variants of same. The worse problem, though, is that more and more new Web text is AI generated, so all those large language model (LLM) workhorses are being fed more and more with their own often erroneous outputs. So I expect that further tuning by software folks will improve the AI output quality, it is likely to be doomed in the long term from that ever increasing recycling of their own crap.
As someone who incorporates 275 teraflop AI engines in airborne imaging systems for manned and unmanned vehicles, this concerns me very greatly.
The down side to AI is that we cannot put that genie back in the bottle
(Puts Aesop down as "scenario three".)
Knowing that AI is capable of screening out those unwanted inputs from the pool in your analogy is key to understanding how to use the thing. It will do what you tell it to, but if you don't tell it to screen stuff out, or 'act as' various personas who would do same, it sucks up from the ENTIRE pool of knowledge. Treat it like a high-speed lieutenant or intern. Tell it to act as an expert in the field you're looking for an answer in, make it ask you clarifying questions, and make it provide citations for anything you deem critical - e.g. the specific section of state code that allows or disallows certain actions, behaviors, etc. It's still not great for creative work beyond image generation, but is a massive time-saver when researching topics.
The real trick with machine learning is understanding what it's been trained to do (in the case of LLMs -sound like a smart person, not 'be' a smart person), and so as long as you give it the appropriate guardrails, it's a very efficient way to extract knowledge from work published by bona-fide smart people. It ain't coming up with 'new' stuff on its own - and you HAVE to know that.
I tell people that what we have is nowhere near any real AI - as you point out, it is fairly simple machine aggregation: It isn't even enough for me to call it Machine Learning yet.
It is over hyped and at some point the bubble will burst, like every other tech bubble has. It is only after the frenzy dies away that we'll see real advances in it.
Jonathan
I am alarmed by the advent of so-called AI for exactly the reasons you specified. The internet is full of info, true and false. If you can filter out the BS, it's a treasure trove of knowledge. But the more AI drivel gets added to the total info, the more polluted with ignorance and misinformation it all becomes. The "dead internet" won't happen because bots are everywhere; it will happen because they're squeezing out real information with hallucinatory AI drivel labeled as truth.
Post a Comment