patrick.net

Perhaps I shouldn't have made a new AI thread this week - did not know about this one.

So.

I'll just post this right here:

https://patrick.net/post/1346289/2022-06-19-ai-thread

MisdemeanorRebel says

Can't recognize the term "Wire Transfer" on the IVR

In development for 25 years.

You might want to read about GPT-3. It learned about itself shortly before the stolen presidential election.

@patrick

Try this link instead: https://huggingface.co/spaces/dalle-mini/dalle-mini

Thanks @Ultra_FJB

Was this ad created by AI?

Inventory robot crashes after lifting 9 boxes.

https://twitter.com/realJustATheory/status/1646400484230713346?ref_src=patrick.net

There is to much risk to have large transfers done though a mouse click. You want the bank employee to verify not a computer.

beershrine says

There is to much risk to have large transfers done though a mouse click. You want the bank employee to verify not a computer.

Wait! Computers can be off by one?

Artificial intelligence needs to get better, it can't predict winning lottery numbers. So sad!

AmericanKulak says

Inventory robot crashes after lifting 9 boxes.

It was the bypassing trannies that jumped on it and started dry humping it that was truly disturbing.

AI is finding a market right now. Once that is established, the process of AI replacing human workers will take about as long as it took Germany to conquer France.

We are looking at another year, maaaaybe two.

Students everywhere are using ChatGPT-4 to complete their college assignments.
What I’m wondering is when will they figure out that the same AI who “helps” them now will be doing the jobs they trained for before they graduate.

1337irr says

Artificial intelligence needs to get better, it can't predict winning lottery numbers. So sad!

Lottery numbers are not random, they are picked by a computer alogorithm.

https://www.cnbc.com/2017/06/30/mastermind-of-lottery-fraud-admits-he-rigged-jackpots.html

You'd think this would have been a huge scandal, but it was barely mentioned. The entire system is a scam.

Shaman says

AI is finding a market right now. Once that is established, the process of AI replacing human workers will take about as long as it took Germany to conquer France.

We are looking at another year, maaaaybe two.

Students everywhere are using ChatGPT-4 to complete their college assignments.
What I’m wondering is when will they figure out that the same AI who “helps” them now will be doing the jobs they trained for before they graduate.

I dunno, it seemed pretty helpless when I asked it two simple questions. One it managed to answer correctly after 8 iterations and the other one it didn't get right until I ran out of 10 allowed by the free trial. The thing is: if I didn't know the correct answer for the 1st one I would've stopped after first iteration and went away with incorrect info. As for the the second one: I know the answer is incorrect and useless, but I don't know the correct answer either. Well, scratch that, I know it now, but I did find out by good old search and reading through an article on the subject from a known reputable source.

https://www.breitbart.com/news/deepfake-porn-could-be-a-growing-problem-amid-ai-race/

So I couldn't make this long and complicated jdbc url work (it's failing with a syntax error but looks completely fine to me) so I asked Bing AI what's wrong with it. Should be easy for something ready to replace all white collar peeps, right? The fucking thing recommended to remove one closing parenthesis leaving the whole url with unequal number of opening and closing parentheses.

Yeah, this shit is ready to take over ANY SECOND NOW!

Eric Holder says

So I couldn't make this long and complicated jdbc url work (it's failing with a syntax error but looks completely fine to me) so I asked Bing AI what's wrong with it. Should be easy for something ready to replace all white collar peeps, right? The fucking thing recommended to remove one closing parenthesis leaving the whole url with unequal number of opening and closing parentheses.

Yeah, this shit is ready to take over ANY SECOND NOW!

And you know what ChatGPT suggested? Attempting to connect to the database and seeing what error will it display. When informed that the error is reported as a "syntax error" the fucking HAL recommended to check syntax! At least it didn't offer an obviously wrong answer, but fucking shit, man! It's totally gonna replace EVERYBODY!!! SOON!!!!

Eric Holder says

So I couldn't make this long and complicated jdbc url work (it's failing with a syntax error but looks completely fine to me) so I asked Bing AI what's wrong with it. Should be easy for something ready to replace all white collar peeps, right? The fucking thing recommended to remove one closing parenthesis leaving the whole url with unequal number of opening and closing parentheses.

Yeah, this shit is ready to take over ANY SECOND NOW!

The MS Visual Studio RAD autocomplete and syntax has been checking syntax for over 25 years.

Tenpoundbass says

The MS Visual Studio RAD autocomplete and syntax has been checking syntax for over 25 years.

I've been working with American syntax for over a half century and still get it wrong.

https://finance.yahoo.com/news/ibm-plans-replace-nearly-8-174052360.html

back-office functions, specifically in the human resources (HR) sector, will be the first to face these changes.

Sostenga mi cerveza, por favor.

https://twitter.com/ExposingBiden/status/1678856789838708737

zzyzzx says

https://finance.yahoo.com/news/ibm-plans-replace-nearly-8-174052360.html

back-office functions, specifically in the human resources (HR) sector, will be the first to face these changes.

That's because all HR does now is check boxes, just like an "AI."

The only function of HR is to keep white collar wages low by using HR as gatekeepers for salary negotiations. If all the HR flunkies don't know what the jobs entail, they massively lowball offers until it gets so bad that they are directed to stop fucking around.

Supposedly, HR flunkies can go as much as 20-25% higher than the initial offer. It's about 10% for gubment. Source for this is a 30 year gubment HR rep.

I've been playing around with the artificial intelligences in the last few days for coding.

1) it gives incorrect answers quite frequently
2) it gives correct answers quite frequently

I think it's very useful for development as well as learning but it's a bit limited. I can see this being used as an assistant. You still need to know what you are doing, and you can ask it questions which are still "black arts" and it will just make shit up. I tried to get it to do a websocket example using Apache, it gave me a cgi script which is NOT a solution - far from it.

Still, it was great for remembering how to setup ncurses and graphical libraries I have never worked with. I think it's very useful, but it can't (yet) replace a coder, not yet. It was useful for using local sockets as well, which is inter-process communication, both named and anonymous pipes. It's not the job killer it was being sold as, but boy is it useful for learning and refreshing memory.

Ever worked on WebAssembly? I THINK (because I didn't test it) that it dumped an example right out. WebAssembly allows you to run an application in your browser, which means it will run on any phone, any tablet, any computer, because the code is interpreted. It eliminates the need to write an application specific to the device at the cost of speed.

The first fucking problem is "how do I even start on this fucking problem?", and it's GREAT at doing that! However, it's based on a lot of false information. I worked on EXI (you don't need to know about it or what it is), and the solution it gave me was the FIRST solution I tried, which doesn't fucking work. Wasted a week finding that out.

It's like having a really really really knowledgeable intern with no experience. Very useful. Anything I need to learn, this speeds it up by 10x.

Instead of all the drudgery of figuring out how to startup on something, it just gives it to you with a baseline. From THERE you can develop. I just made a simple ytalk interface today just fooling around. None of you know what ytalk is, it's between a phone call, and an instant messenger. As you type, the person on the other side sees what you are typing as you type it, and this allows them to ignore what you are typing or to predict what you're about to say, but unlike a voice call, you don't end up "talking over one another". That's one thing I want to bring back, HOWEVER, a problem with it is that it consumes TREMENDOUS bandwidth which is a problem on a PHONE on 5G but not on your local internet connection. Ytalk also had some dumb problems with it back in 1990 in order to save memory and reduce CPU consumption, I can eliminate those problems because they no longer exist.

AIs suck and are brilliant, both. I don't think they are going to be free forever, so make use of them as much as possible when you can.

Also, the AI I'm playing with (Bing) when caught in a contradiction about politics or logic, will disconnect. I got it to admit Ukraine had a coup in 2014, but it also said it had a "Revolution of Dignity". It knew Victoria Nuland picked out the new government, and when I asked it the definition of a coup and a revolution, and told it that it was either a coup or a revolution, and since Victoria Nuland picked out the new government it had to be a coup, it disconnected. It's useless for political information. It gets information from Wikipedia apparently, it knew about the Douma chemical attack, AND that two whistleblowers from the OPCW explained it was staged and how it was staged - another disconnect.

Once it disconnects, it resets and you're back to square one. It cannot resolve cognitive dissonance at all. When confronted with it, it appears to reset.

Thought this was going to be a thread about Democrat woke voters.

PeopleUnited says

Thought this was going to be a thread about Democrat woke voters.

Dunno, intelligence and Dem voters do not intersect. You may be thinking of NPC

This must be the product of AI

British king or BK?

https://www.zerohedge.com/technology/google-ai-says-calling-communism-evil-harmful-and-misleading

Booger says

https://www.zerohedge.com/technology/google-ai-says-calling-communism-evil-harmful-and-misleading

If Calling Communism "Evil" Is "Harmful And Misleading" then AI is not "Intelligence." It is
simply a means to promote Fake News and Propaganda.

The_Deplorable says

AI is not "Intelligence

Yes, AI is not intelligence. Those old enough should remember the big AI fad in the 1980s. True "intelligence" come from the ability to originate a thought, something a human can do in the flash of an eye. Computers in the 1980s were unable to have an original thought and neither can any computers today. AI in reality is just an auto-summary device.

More Google Glasses and Self-crashing cars.

We're about due for a sudden glut of Vampire movies and a Florida R/E Crash.

Sora is amazing at videos. We are still early in this but progress is ridiculously fast.

KgK one says

Sora is amazing at videos. We are still early in this but progress is ridiculously fast.

Its going to be a major disruptive technology to Hollywood. Look at how the movie 300 was revolutionary in regards to green screen film making, as well as computer generated imagery (CGI) with Tron (1982) and Westworld (1973). And now there is free and open source CGI apps such as Blender.

This is going to have a vastly greater impact. The only question is how can it be prevented from being exploited as far as creating deepfake videos ?

.

HeadSet says

Computers in the 1980s were unable to have an original thought and neither can any computers today. AI in reality is just an auto-summary device.

This isn't true.

A computer can create something new. When chips are routed, the shortest traces and least number of traces is what is desired. With a billion gates, that number of traces is very high. You do this with what is called simulated annealing. You put a computer (or a bank of computers) to work randomly doing placement and routing, and then you compare all the solutions against one another. The least worst solutions are combined sometimes with randomness, and then it's tried again, and again, and again and again.

When you're done running it, there's virtually no improvement or no improvement.

No human being or team of computers can create a result as good as the computer simulation did.

It all boils down to what you consider "thinking". It might just be random ideas that turn out to be useful, and in simulated annealing, it's just randomness combined with what was found to be previously useful.

AD says

This is going to have a vastly greater impact.

Hollywood is dead.

AD says

The only question is how can it be prevented from being exploited as far as creating deepfake videos ?

It can't be.