ChatGPT
Desperate times call for desperate measures. So to goose the stock market and push up interest in select businesses, they're pushing another dud, ChatGPT, with grandiose claims.
At the same time they've released generators that produce pictures and artwork from key phrases. This sudden barrage is to give credibility to their shaky, questionable products. All part of the plan, and like a used car salesman, they hope you don't look too closely.
ChatGPT is a valiant attempt to cash in with a half-baked product. It works fine for what it is, but it's not some seer or intellect. Odd how things like GPS and Google Maps, which gives things a whole new perspective, aren't recognized or are merely taken for granted, when they're huge, useful developments, but nonsense gets full court press and outsize promotion.
Why don't they work on making what we already have, spectacular, like personal computers, internet and enhanced use of databases, like inventory databases. (For instance, why can't I tap into the inventory of my local store to see if they have what I'm looking for, in stock?)
The Wolfram site has a clear explanation of ChatGPT, but it doesn't seem complete.
ChatGPT is a clever type of neural net application. As such, it merely places one word, or group of words, after another, based on probability. It has been fed billions of samples based on human writing, and from there it has accumulated a pool of information to tell it how likely one word is to follow another! So it constructs its "chat" response based on the most probable responses it has accrued from other people's writings. This is adjusted so it doesn't always output the same text to the same question, but the result is, you have a sort of a parrot, without the intelligence of the actual bird.
We've brought it on ourselves, though. People talk such nonsense platitudes and pap, it's no wonder they're fooled by ChatGPT. Its mindless drivel and small talk is just like talking to a person. The exact same effect as when the first conversational program, Eliza, was introduced, and people were fooled by it, thinking it was human, and it was straight-up lame.
ChatGPT has its place as an interesting exercise in neural nets, but it's not going to give you anything original, only a sort of sophisticated boilerplate.
The problem with the Wolfram explanation, though, is that ChatGPT has been tweaked — programmed to provide only the "politically correct" responses — so there must have been other tampering going on whereby it doesn't follow only the rules outlined above.
Well, that's the thing about neural nets: they can be tweaked, their processing weighted to give more credence to certain things than others. That neuters the product, so it will have a chilling effect, reducing its user base. This already limited program is not even allowed to fulfill its potential. There will be a lot of hype and flurry around it for a while, then boredom will set in, as its limitations become very apparent.
But there's more, since that method still won't work for cases where someone wants an answer to a programming problem. There, where every symbol is significant, using a probabilistic response will just return nonsense. So this means there must be some other tactic in use, and it seems they're calling two different programs, "ChatGPT."
How the other works: The "problem solving" ChatGPT is a search engine that relies on fact that people have been doing all the leg-work, on sites like Reddit, where people pose questions, others answer, and the answers are graded, given "points" by users. This and other "answer" sites already have compiled databases of rated answers to programming questions and such, which ChatGPT can skim and regurgitate.
So this incarnation of the A.I., is a search engine/scraper, harvesting all these answers and re-issuing them as its "intelligent responses."
How we know for certain that this crock is a crock? Easy giveaways: Things like misspellings in its responses, like an instance where, "too," was spelled "to" of all things. That's all you need to know. An A.I. obviously couldn't make that mistake, it simply regurgitated a misspelled human response to the question from somewhere else.
Also, it can't always provide answers, and then comes up blank. Nothing in the database, it can't function.
In another fiasco, it "played chess," against top bot, Stockfish, and didn't know how to move its pieces correctly, moving its pawn like a queen, taking its own pieces, castling through pieces, ignoring being checkmated, etc. Sure, you want this "super-intelligence" to write your computer programs when it can't even learn or conform to the rules of chess.
Now for the vitally important piece of the puzzle. What about all these goons saying it wrote a computer program for them, or a sonnet, or cured their ailments? In cases of very specific things like that, they're doing what all hyping shills do: they lie. Some human manually programs some software code, or tweaks a sonnet, claims it's "A.I." then the legions of these shills — the "grassroots marketers," particularly certain YouTubers that suddenly appear — publicize it. (Or there are of course those who simply post lies about how ChatGPT wrote a program for them.) No magic.
This trick is a replay of something we've seen before, years ago, a little ruse pulled by a consulting company. It couldn't complete its networked accounting software in time, but it was contracted to a fixed delivery date, with big penalties if they buggered up. So they just hired a legion of data-input clerks to post and process data behind the scenes. The client would enter data, see a calculated result spit back up on the screen, and it looked like everything was dandy. One person entered an obvious error, that the hidden worker caught and corrected on the spot, and the client marveled. "Look at that, it even corrects typos!" "Oh, yes," said the consultant. "The system can be quite intelligent. But we don't guarantee it will catch all user errors!"
The always-amusing hucksters are out in full force for this ChatGPT operation. One stooge writes Coding Won’t Exist In 5 Years. This Is Why in the JavaScript in Plain English site, but provides no evidence to back it up. Well into the article, we see the actual intent in writing this screed: a puff-piece to promote ChatGPT, where he tries to support his thesis, with no technical explanation, and without a single example!
Comments