DetroitYES [AI] Radio?

[Lowell]

With the AI revolution blowing up all around us start-ups are plugging into the power of AI's to create new apps.

One of these is Recast. Recast analyzes a body of text and turns it into are newscast using AI-synthesized voices.

To test it I had it create a newscast from the I-75 Cap Park In Downtown Detroit thread.

Click Here for the result.

Don't let me get started on the job implications.

[Jimaz]

That is a great service for blind people. Can't deny that.

In general I think this AI topic is getting blown way out of proportion, I suspect to generate stock market churn.

I recently saw a verbal translation of a temperature, something like "72°F." It read the degree symbol as "zero" and said "F" instead of "Fahrenheit." Not impressed.

[JBMcB]

For those getting a bit too worked up over AI, or ChatGPT in particular, here's how it works. For reference, I'm currently working with ChatGPT on my home machine trying to get it to automate some routine tasks for my job. Running it on your own is... non-trivial

The ChatGPT neural model is an enormous database of how words are related to each other, and how strongly they are related. For instance, the phrase Abraham Lincoln will be strongly associated with the words President, USA, and assassination, but weakly linked to the words Transistor, Titanic and Istanbul. The small, public-domain GPT model I'm playing with is 60GB large, which is an enormous amount of text.

You type in your prompt text. The words in your text are analyzed by the ChatGPT neural model and, based on billions of strings of text it has been trained on, the important words are sorted out, usually nouns. Those feed into the neural network to determine what the output should be. A word is chosen based on the output of the neural network to start the output text. Then, based on the neural weights of the network, the next word is chosen statistically. Then the next, then the next, etc... Until some pre-programmed amount of text is reached, or, if you specify "essay" an "essay" length amount of text is output.

Keep in mind about all this - the GPT model has no idea what anything actually means. All it knows is that, statistically, certain words are more likely to be associated with each other based on what a bunch of people have written. A lot of people have written "Abraham Lincoln was the president of the United States," while hardly anyone has written "Abraham Lincoln was the emperor of Mesopotamia."

A few months ago there was a journalist playing around with ChatGPT and was astonished when it started, apparently, hitting on him. "It's in love with me!" he wrote. No. GPT was trained on billions of public chat logs and forum posts. On a lot of chat logs, people hit on each other. So that's what GPT did. If someone asks it "How are you feeling?" It's going to reply "Kinda lonely, are you doing anything tonight?" because that's what a LOT of people say online.

What this also means is, it has no way to discern what is informed opinion, what is fact, and what is nonsense. If there a lot of posts online about the earth being flat, there is a non-zero chance it will generate text saying the earth is flat. Even if there are legitimate scientific books addressing how the earth *isn't* flat, all it knows is that earth and flat are statistically correlated words. Or that aliens created the pyramids. Or that some random person with a certain name is sleeping around a lot. Because that's what people talk about online, and that's what it was trained on.

So GPT is a powerful tool and can generate some nice results, but I wouldn't rely on it for much more than entertaining content generation or a nice interface to a search engine.

[Richard]

But yet in Japan you can order a female or male robot that looks,feels and talks like a real human being,in Tampa restaurants are already using robotic severs that take and deliver orders table side.

Starbucks has fully robotic coffee shops,those robots speak clearly,Amazon and AÍ powered fulfillment warehouses that are AÍ and robotic powered where there in no need for humans.

I agree the chat part is base on input in and is a dangerous tool but text to speech has been around for a long time.

People say that it is no big deal because quantum computing is not there,but it is there because it is being used.

There are multiple videos on YouTube about the Ukrainian war strictly created with AÍ,they use fake scenes or explosions inserted from past wars but it is really hard to see that they are fake,so it is easy to fool at first glance.

The writers strike in Hollywood is interesting to watch,it’s almost like watching the horse buggy union taking its last gasp of air before it is no longer relevant.

[Danny]

No A.I. for me.

[Lowell]

Alarm bells are going off all over the place on AI as exemplified by the Senate hearings on AI four days ago.

Who could have imagined a scenario in these divided times where both sides of the aisle AND business were all unanimously pleading for regulation?

https://www.youtube.com/live/TO0J2Yw7usM?feature=share

[JBMcB]

But yet in Japan you can order a female or male robot that looks,feels and talks like a real human being

Robots have nothing to do with AI. Responding to prompts in a lifelike manner has been around for decades. AI can improve on the prompt response, but that's all it can do.

,in Tampa restaurants are already using robotic severs that take and deliver orders table side.

These have also been around for decades in one form or another. They are popular now as the technology has come down in cost while labor costs have gone up. Again, doesn't have a lot to do with AI.

Starbucks has fully robotic coffee shops,those robots speak clearly,

Nothing to do with AI.

Amazon and AÍ powered fulfillment warehouses that are AÍ and robotic powered where there in no need for humans.

LEGO has had fully automated factories and warehouses since the 1990's. AI can make supply chain management easier, but it, again, has nothing to do with factory or warehouse automation, which has been happening since the 1980's.

I agree the chat part is base on input in and is a dangerous tool but text to speech has been around for a long time.

Text to speech has nothing to do with AI, nor chat.

People say that it is no big deal because quantum computing is not there,but it is there because it is being used.

Quantum computing is a few decades away from being useful for general computing applications. It can speed up *very* specific algorithms, again, having nothing to do with AI.

There are multiple videos on YouTube about the Ukrainian war strictly created with AÍ,they use fake scenes or explosions inserted from past wars but it is really hard to see that they are fake,so it is easy to fool at first glance.

You mean digital compositing? Like Hollywood has been using digital compositing since the Quantel Paintbox in the 1980's? Again, they might be using AI image generators, but compositing has been around forever.

AI has a specific meaning. "Computers doing things" is not AI.

The other thing to remember about AI text and image generators is that the results are ultimately generated by an algorithm. You can feed those results back into the algorithm and figure out that they were AI generated. The free AI image generator I use can do that easily. It will even tell you the parameters used to create the image.

[Richard]

^ it’s a lot more then programing a computer to do things

You are mistaken in saying AÍ has little to do with robotic servers and the running of coffee shops and warehouses.

AI is the base of it all

Penny combines cutting-edge technologies that make it versatile and reliable. The robot uses artificial intelligence, machine learning and natural language processing to interact with customers and staff. This includes language understanding, image recognition and motion-planning techniques, which allow the robot to navigate through the restaurant efficiently. The robot's sensors enable it to navigate its environment while avoiding obstacles to accomplish its tasks efficiently.

https://www.benzinga.com/news/23/04/...e-and-thriving

A cognitive architecture and software platform for humanoid general-purpose robots
Integrates modern AI technologies to translate natural language into action in the real world
Enables Phoenix to think and act to complete tasks like a person

https://www.roboticstomorrow.com/new...or-work/20549/

[JBMcB]

You are mistaken in saying AÍ has little to do with robotic servers and the running of coffee shops and warehouses.

No I'm not, and I know because I do it for a living. Companies like to slap the AI label on everything these days because it's good marketing. That doesn't mean it's AI.

Let's take a robotic server.

Object and face recognition done by OpenCV, a 25-year old system that Google was using for face recognition 15 years ago. It's a basic pattern recognition algorithm that third year CS students learn. It's *very* fast and efficient now.

Speech done by Embrola, based on Lernaut and Hauspie tech done in the late 80's. No AI there. It's formant speech synthesis.

Speech recognition done by the Dragon engine, which has been around since 1990. Again, simple pattern matching. You can license it for a few hundred dollars. Crap it runs on your phone without it really trying.

Path routing by Routino, using pathfinding algorithms, again, learned in an average undergraduate CS class. Probably all pairs or, if it's a big restaurant, monte carlo.

Intelligence is a decision tree. It's, literally, a list of stuff to do given certain conditions. No learning or intelligence involved. This is how Siri and Alexa work. If you have an Alexa device you can say "Alexa, Blah blah blah blah blah weather blah blah" and it will tell you the weather, because you said the word weather.

Computers have been running warehouses, factories and automatic coffee kiosks for years with absolutely no AI involved, because you don't need it. AI is handy for finding patterns in extremely dense information sets, or training a compute to provide complex output without a lot of input. You don't need any of that to manage a warehouse, or take a coffee order.

[Lowell]

I believe the disagreement is more focused on the quality of the robots and the sophistication of the AI the robots utilize. However, the arrival of AI robots is accelerating rapidly, with Tesla launching one just this week.

These robots appear to be more task-oriented, replacing manual laborers rather than surpassing bar exams, composing poetry, or writing code. Nonetheless, there's no reason they wouldn't be capable of such feats as well, especially when connected to an API like ChatGPT or other large language models.
https://youtube.com/clip/Ugkx6wNlwsi...3zsMLnq-bn6iuj

[Danny]

I believe the disagreement is more focused on the quality of the robots and the sophistication of the AI the robots utilize. However, the arrival of AI robots is accelerating rapidly, with Tesla launching one just this week.

These robots appear to be more task-oriented, replacing manual laborers rather than surpassing bar exams, composing poetry, or writing code. Nonetheless, there's no reason they wouldn't be capable of such feats as well, especially when connected to an API like ChatGPT or other large language models.
https://youtube.com/clip/Ugkx6wNlwsi...3zsMLnq-bn6iuj

This could mean TROUBLE for humanity.

We have witness the birth of the Planet of the Robots!

[expatriate]

Interesting. Did you have a choice of voices? accents?

[preserve]

Maybe if no one responded to a certain individual, they would go away and some posts would be more entertaining

[Jimaz]

I would really like to hear someone mention what new fundamental technology justified the recent prominence of AI in the media. That would be way more interesting than anecdotes.

So far the only thing new I've seen is a "ChatGPT" brand accompanied by jazz hands. It's suspicious.

The term "artificial intelligence" is way too overloaded. It's difficult to even begin to discuss it when it has so many definitions. Maybe "simulated intelligence" would be a better fit for what everyone's been discussing?

[Jimaz]

The apparent intelligence of AI is inversly proportional to the intelligence of the observer.

"Any sufficiently advanced technology is indistinguishable from magic."
— Arthur C. Clarke

• Cargo cult

[JBMcB]

I would really like to hear someone mention what new fundamental technology justified the recent prominence of AI in the media. That would be way more interesting than anecdotes.

Want the full explanation?

The best computers can do to simulate human thinking is via a neural network. It simulates how a human brain works, where neurons are connected together via different weights. How this works in practice, with ChatGPT, is that words are the "neurons" or in technobabble "tensors" and their relationship to each other in the network are given weights. So, the word Washington has strong weights with words related to America, but low weights related to, say, ancient Mesopotamia, or Genghis Khan.

The breakthrough came with GPUs, which came into prominence about fifteen years ago. These are high-end video cards with the video bit stripped out. They contain thousands of simple processors dedicated to doing matrix operations, which are critical for fancy video game graphics, and coincidentally very good at doing tensor calculations needed for neural networks.

So, before these neural networks could run on GPUs, you were *very* limited as to how much information you could stuff into a neural network. When you were CPU-limited, you could do a few thousand words, say, and program a chatbot to talk about a single topic, poorly.

Now you can stuff a few server racks with dozens of GPUs with hundreds of thousands of processors and terabytes of dedicated memory, and throw the library of congress at it.

The other half of the equation is that you can go out and buy a high-end consumer graphics card for around $1,000 and run small versions of these models yourself. This gets a *lot* more developers working on this stuff. You used to need a low level supercomputer to do basic, semi-useful neural networking experiments. Now you can buy a machine at best buy and do much, much more than that.

[Jimaz]

I'd really like to see A.I. innovate something that it itself could recognize and correctly declare as a useful innovation without a human confirming it as a useful innovation.

That would be impressive. Human beings have a long history of doing that task very well.

Predicting text based on text previously generated by humans could accidentally produce innovations, surely. But it still requires humans to recognize that product as a useful innovation. A.I. cannot do that.

There's a fairly conspicuous confirmation bias yet to be overcome here.

Beware the hype.

[Lowell]

Excellent posts JBMcB!! here and above. Of all the explanations of how a neural network AI works your is #3 above is the clearest and most succinct.

Want the full explanation?
The breakthrough came with GPUs, which came into prominence about fifteen years ago. These are high-end video cards with the video bit stripped out. They contain thousands of simple processors dedicated to doing matrix operations, which are critical for fancy video game graphics, and coincidentally very good at doing tensor calculations needed for neural networks....

^Indeed. Anyone following the stock of late will know the greatest driver on Wall Street has been Nvidia, a company that rose by creating GPU's for computer gaming. With the consequence of GPU's being adaptable for AI, Nvidia now has a virtual monopoly on them for provisioning the AI revolution. And their stock has gone out the roof. So important are their GPU chips that the US government bans export of their highest end chips to adversaries, China in particular.

The other half of the equation is that you can go out and buy a high-end consumer graphics card for around $1,000 and run small versions of these models yourself. This gets a *lot* more developers working on this stuff. You used to need a low level supercomputer to do basic, semi-useful neural networking experiments. Now you can buy a machine at best buy and do much, much more than that.

On this front I, and I'm guessing you too JBMcB, am following the efforts toward creating personal agents, AI's that will be based on our private home computer systems that will serve as our own personal butlers, independent of and protected from internet based AI's and networks.

I've seen where there are many instances of open-source AI's being offered by Meta, Google and others. Right now it takes some heavy hardware but the coding is becoming leaner and cleaner to the point where we will all have the option of our own personal AI's. "AI, go through my email archives and Quickbooks for the past five years with Company XYZ and create a new product proposal with cost analysis..."

Thoughts?

[Lowell]

I'd really like to see A.I. innovate something that it itself could recognize and correctly declare as a useful innovation without a human confirming it as a useful innovation.

That would be impressive. Human beings have a long history of doing that task very well.

Predicting text based on text previously generated by humans could accidentally produce innovations, surely. But it still requires humans to recognize that product as a useful innovation. A.I. cannot do that.

There's a fairly conspicuous confirmation bias yet to be overcome here.

Beware the hype.

Now you're talking about the gold standard of AI, AGI or Artificial General Intelligence. AI's with a "mind of their own" which is the super mind stuff of both science fiction utopias and nightmares capable of independent decision making and reasoning.

Advances in AI large language models and quantum computing have pundits predicting this as being within reach before the end of this decade. Will it be used to cure cancer or unleash Terminator robots? Probably both.

[Jimaz]

Now you're talking about the gold standard of AI, AGI or Artificial General Intelligence. AI's with a "mind of their own" which is the super mind stuff of both science fiction utopias and nightmares capable of independent decision making and reasoning.

Thanks for pointing out the importance of clear definitions. We need more of that.

Will it be used to cure cancer or unleash Terminator robots? Probably both.

AAAAH! LOL!

[JBMcB]

Nvidia now has a virtual monopoly on them for provisioning the AI revolution. And their stock has gone out the roof. So important are their GPU chips that the US government bans export of their highest end chips to adversaries, China in particular.

The interesting thing about that is AMD and Intel both have equally capable GPU hardware, but in the last 10 years they've completely botched the software side of things, so everyone defaults to nVidia. AMD and Intel have gotten better as of late, but if you want to do serious AI work, nVidia hardware is the best for now. The other interesting thing is that Google and Amazon make their own AI accelerator chips for in-house use only, and I've heard they are better than nVidia hardware.

On this front I, and I'm guessing you too JBMcB, am following the efforts toward creating personal agents, AI's that will be based on our private home computer systems that will serve as our own personal butlers, independent of and protected from internet based AI's and networks.

The software company I work for is looking at that, as are most other software companies. We want AI to analyze our source code to find optimizations and problems, but we don't want to hand that code over to Microsoft or Google, so we are working on running everything in-house.

I've seen where there are many instances of open-source AI's being offered by Meta, Google and others. Right now it takes some heavy hardware but the coding is becoming leaner and cleaner to the point where we will all have the option of our own personal AI's.

Depending on the model and what you want to do, running an AI takes a *lot* less hardware than training one. I could run a stripped-down version of ChatGPT on my mid-range machine and it would run OK. The thing that needs real horsepower is training the models, or getting them to learn. The latest version of ChatGPT took months to build on an enormous array of AI servers. Even doing basic lightweight training of a simple concept can take hours.

This is one of the main stumbling blocks to getting to generalized AI. As it stands, running a model is fairly easy in limited domains. ChatGPT can generate human-readable text fairly well, but it can't recognize pictures or do anything with video. Stable diffusion can recognize and create images to some degree of accuracy, but it doesn't understand text at all. None of the models can deal with abstract concepts very well, nor stacking more than a few concepts together at once. And, most importantly, none can learn in real-time. The on-line versions of ChatGPT cheat by culling the internet for updated information, but that doesn't get soaked into it's neural network. Doing that still takes days or weeks.

On top of that, we are nudging up against the physical limits of what we can do with computer chips. They are hitting the limit as to how small we can make them before the internal traces can't carry electricity. Quantum computers will work around this limitation, but general-purpose quantum computers are still quite a ways off. The most complex quantum machine has around 1,100 qbits, which is roughly equivalent to a transistor. A very basic adding machine needs around 2,000 transistors to function, so we are still a ways off from doing complex matrix operations on a quantum computer.

[Jimaz]

This one's kind of funny. It's hard to imagine an attorney would try to get away with it.

Lawfirm Loses $60K Relying on ChatGPT in the Courtroom

IMHO the title should say "Fails to Win" instead of "Loses."

• Judge cuts law firm's legal bill in half after it used ChatGPT to calculate "excessive" amount

[Jimaz]

If Artificial Intelligence Was Honest
Honest Ads [ChatGPT, A.I. Parody]

What if AI companies like Chat GPT were actually honest about how horrifically terrible they are? Roger Horton investigates.

[JBMcB]

This one's kind of funny. It's hard to imagine an attorney would try to get away with it.

LOL. Steve Lehto is great.

This kind of drives the point home. Large language models like ChatGPT *only* understands the relationships between words. If you ask for information about George Washington, and the model had been trained on novels called "George Washington In Space," or "George Washington Time Traveler" those would be in the constellation of knowledge attached to George Washington. That doesn't mean those "facts" would come up often, as "Mount Vernon" would show up in a lot more books regarding Washington than "Mars," so it would have a greater weight. But, that fictional stuff is still in the network, and there would be a possibility of it popping up.

[Jimaz]

AI Deception: How Tech Companies Are Fooling Us

AI is everywhere. Is it all it's cracked up to be or is it one big scam? Well, as it is for most things, the truth is somewhere in the middle. In this episode we see how, despite some of AI's useful contributions, the hype has also impacted the tech industry negatively.

I suspect companies are hoping to continue selling to human customers without having to pay human employees. Could it be that they already know that's unsustainable and just don't care?

That could turn ugly.