God do I hate it when people allude to the Hitchhiker's Guide to the Galaxy without ever having read the books or even listened to the original.

Don't forget your towel.

How to have fun with AI for the German speaking folks:

Search "Was macht Michael Ende heute?" in google.

Have a good laugh and maybe be a little sad.

Why Lukas Pravda one of Exscientia pharmacy employees old friend of Michal"LEOTCK"Bonk would be accused of scam by one of employees of Kopernik Observatory & Science Center Marcel Mazur in LinkedIn is it because of so-called Artificial Intelligence/AI Drug? And why police locally or even internationally in any country keeps ignoring that such case?

Who claims so? Let me guess: the AI itself?

Anyone who've seen the movie "Upgrade" KNOWS full well that AI does not always really help. Sure they act like they are there to benefit us, but in reality they have very sinister plans.

So I wouldn't believe an AI if I were you! Who knows, maybe I even am you!

Could we please stop calling it "AI"? That abbreviation "AI" is a marketing term.

It is, and always was, a "pattern match algorithm". There is no intelligence to it. All it does is inhale input, calculate statistical distances of words or pixels by matching how often those appear next to each other, and associate them with search terms.

PMAs can be trained to tell the pictures of a cat from pictures of a dog. And you can reverse it: Tell it to generate an image that is statistically the most typical dog it knows.

But, ask it anything it hasn't been trained on, like "draw me an elephant", and you get (surprise) another dog.

Or, give it an elephant and ask what this is, and it will tell you: "This is a Norwegian Forest Cat."

This cannot be fixed. Because of the very nature of PMAs as such, they only deal in probabilities based on how often they have seen a certain input during training.

This also means, that PMA models will degrade over time and become entirely useless after a while. Because, you cannot give it random data and expect it to "learn" anything. You need to invest time, carefully select and add weighting info to data, to ensure the training data is accurate. If you don't, you skew your probabilities.

Classic example is "AI" in garbage collection plants, that was meant to sort garbage, by--for example--removing non-recycleable materials from a recycling production line. Except, given the right incentive, people over time became too good at presorting their garbage at home. So, when the "AI" continued to "train" with the input it saw, its probability-model degraded and broke down. Eventually, it would either stop sorting garbage all together, or do it at random. Resulting in the ridiculous outcome, that human workers had to be put back into line, to carefully introduce a representative mixture of "wrong" garbage back into the system, to keep the "AI" from unlearning what that "wrong" trash looked like.

Keep feeding any model irrelevant data, and after a couple of iterations, it breaks down, returning only incoherent BS - or nothing at all.

We have seen this a couple of years back, when somebody proudly introduced an "AI" that could beat humans in rock-paper-scissors. Now, that "AI" doesn't work anymore. What happened? Well, humans happened. Humans fed the "AI" A) its own output, B) randomized input, C) thought about what they would pick, then did the opposite. All of A+B+C destroyed the PMAs prediction model and rendered it useless within days.

Which, of course, is not really much of a risk for current "AI" that is being released, because, thanks to human greed and the resulting lack of properly vetted training data, those PMAs are useless to begin with.

What makes them dangerous though is the security risks they pose. Because, as users have repeatedly demonstrated, "AI assistants" can be easily exploited to do things on behalf of their owner, by injecting commands in unchecked input. We have seen an explosion of attack vectors, where people have carried out successful attacks by tricking an "AI assistant" to expose system information, or open the door for them. I.e. I recently read about a successful attack, where an AI assistant was told in an e-mail, to send the attacker all the users contacts, and then delete the compromising e-mail with no trace it was ever there.

https://www.youtube.com/watch?app=desktop&v=aJlMjltpUxo
Is there really such thing as AI Drugs related to the Artificial Intelligence or still not yet?

Well excuuuuse me mr smartypants!

So what did you like about that Steven Spielberg's movie, "P.M.A." that he made back in 2001?

AI exists since the 1950s and since the 1950s people, even some experts, have completely unrealistic expectations about AI. *shrug*

Its like those people who say the climate "change" is happening and it may wipe out mankind. Yes the climate collapse is indeed happening and it may, in worst case scenarios, kill billions of people, and wipe whole countries (Bangladesh, Netherlands) off the map. But it is obvious mankind as a whole will find ways to survive it.

Likewise AI is a very useful technology. But its far from a magic trick that can just solve any problem. It just can solve a specific type of problems better than other solutions.

Exactly. Thank you!

It jumped all over the place with an 'adventure' and a nonsensical ending involving aliens. As a thought experiment for how AI (self aware machines) are treated was interesting, but those are not the same things being pushed right now.

It's programming, give it complex enough programming and it can seem alive enough, within reason. But usually they are better specialized, like for games. Chess AI when i wrote one, simply looks at a board configuration and assigns points based on what i think is the best strategies, and then picks the best among them. Not very intelligent, more brute force.

Did a C-robots AI, where i tell it what areas to scan when to move and where and when to aim and shoot at something.

Game AI, often is 'go towards enemy , attack when within range' with no sense of walls or anything; Though RPGMaker AI is more literal RNG. 30% attack normal, 30% strong attack, 10% attack magic, 10% runaway 20% do nothing. Then 'who' it attacks (if you have a party) will be some percentage, say 25% to each of the front positions, and 5% of any of the back positions (if able)

Current 'models' is just lazy programming. Throw spaghetti at the wall have a bunch of nodes that may do or decide something; Try and keep what sticks and is closest to what you want and repeat millions of times. literally you don't know what it is doing so there's no checking it for bugs and flaws. For some artistic things this may work. For anything else, it's a headache and failure waiting to happen.

Until it can be self aware and contemplate and actually articulate it's own thoughts without regurgitating what it has gotten as input, it will just be a tool and likely inferior product.

Didn't someone one put up a million buck reward for an AI that was convincingly "100% human"?
If so, has anyone tried claiming it?

I even saw John Rambo as a’la Rabbit in JJ2 Multiplayer Discord Server. Probably AI Generated Image.

I believe that's just the Turing test. Back in 2010-2015 i remember getting a lot of AIM messages from random people, bots of course. Many were using scripts, but some i think were evolving... well evolving is the wrong word, i suspect a real person was using certain chat conversation bits, then after it was complex enough of a tree had it use 'answers' from humans, so you would get a totally believable answer.... sometimes... But if you asked the same question multiple times you started getting very weird answers. The easiest way to detect if it was a bot, was to ask it's age.... multiple times. I would get 14, 17, 45, 999, 0... etc... Though i was the kind to give very unique replies and watch the response. If the response wasn't WTF!!! i knew it was a bot.

Naturally some would skim a message and have an auto response. If 'bot' was used anywhere in one message i got a reply with 'ha ha you're silly, I'm not a bot' or something, so i ask another question with 'bot' in it somewhere and got the identical response. Without prying it would have been believable. With prying, it wasn't hard to detect. Same with current 'ChatGPT' where it gives nonsense answers, and seems to get more unhinged after 10 messages in a session.

Mmmm... Can't say i'm much a cook, but i cook out of necessity. Pre-processed food, frozen dinners, fast food, restaurants and snack foods are too expensive for what they give. In many ways being forced to cook for myself i've gotten quite good at decent cheap cooking from raw ingredients, and seasonings.

You can only get so lazy for so long. Current prices of things are making some people be forced to do their own cooking, something we've always done ourselves in the past. And.... to some degree, it's comforting. I don't ever follow a recipe book, instead i throw in what seems to be the right amount of meat, veggies, seasonings, and then touch it up near the end and serve. Though judging and coming up with basic rules takes practice and success and failure before you get good. I've only been really cooking for myself since i moved in 2018...

Just saying, currently you can't really be lazy. Either it's too expensive, or you save money by making your own food. Assuming basic raw food doesn't skyrocket.... (which it shouldn't, though meat and cheese are more expensive than i like)