Generative ML [AI] | ChatGPT and other software

[Parodite]

I tried it on quantum mechanics, great fun!

=====================================

1. Is quantum superposition physical or mathematical?

Quantum superposition is a physical phenomenon, not merely a mathematical one. It is a fundamental concept in quantum mechanics that describes the state of a quantum system when it is in a combination of two or more possible states, with each state being described by a probability amplitude.

This state is often represented mathematically using a wave function or state vector, but the physical consequences of the superposition can be observed experimentally. For example, the famous double-slit experiment demonstrates that a quantum particle can be in a superposition of two different states, and the resulting interference pattern on the screen can only be explained by the wave-like nature of the particle.

So while the mathematical formalism is important in understanding and predicting the behavior of quantum systems, superposition itself is a physical phenomenon that has been observed and studied extensively in experiments.


2. What are the observed physical consequences of the superposition?

The physical consequences of quantum superposition are numerous and have been observed and confirmed experimentally. Here are a few examples:

Wave-particle duality: A quantum particle can exhibit both wave-like and particle-like behavior simultaneously. This is exemplified by the double-slit experiment, in which a quantum particle such as an electron can interfere with itself as it passes through two slits, creating an interference pattern that is characteristic of a wave.

Entanglement: When two or more particles are in a superposition, they can become entangled, meaning that the state of one particle depends on the state of the other(s). This has been demonstrated experimentally and has important implications for quantum computing and quantum communication.

Quantum tunneling: In classical physics, a particle cannot pass through an energy barrier if it does not have sufficient energy to do so. In quantum mechanics, however, a particle can tunnel through the barrier if it is in a superposition of states that include both sides of the barrier.

Superposition-based quantum computing: Quantum computers rely on the ability of qubits (quantum bits) to exist in a superposition of states, which allows for exponentially faster calculations than classical computers for certain types of problems.

These are just a few examples of the observed physical consequences of quantum superposition, and there are many more. Quantum mechanics has been incredibly successful in predicting and explaining a wide range of phenomena, and the concept of superposition is fundamental to our understanding of this field.


3. Can "superposition" be directly observed or only be inferred by its consequences?

The direct observation of quantum superposition is difficult because the act of measuring a quantum system causes it to collapse into one of its possible states, destroying the superposition. Therefore, in practice, superposition is often inferred indirectly by its consequences.

For example, the interference pattern observed in the double-slit experiment is a consequence of the superposition of states of the quantum particle as it passes through both slits simultaneously. Similarly, the entanglement between two particles can be inferred by observing the correlation of their measurements, even though the actual entangled state cannot be directly observed.

That being said, there are some experiments that have been designed to observe superposition directly. For example, recent experiments have used quantum superposition to create so-called "Schrödinger cat" states, in which a macroscopic object such as a tiny mirror can be in a superposition of two different positions. These experiments provide direct evidence for the existence of superposition and are a significant step forward in our ability to observe and manipulate quantum systems.

Overall, while direct observation of superposition is challenging, its consequences have been observed and verified in numerous experiments, providing strong evidence for the existence of this fundamental quantum phenomenon.


4. Are there other theories that try to explain the result of double slit experiments?

The double-slit experiment is a fundamental experiment in quantum mechanics, and its results have been extensively studied and verified through numerous experiments. While there are alternative theories that attempt to explain the results of the experiment without relying on quantum mechanics, these theories generally require ad hoc assumptions and do not have the same level of predictive power or experimental support as quantum mechanics.

One such alternative theory is the pilot wave theory, also known as the de Broglie-Bohm theory. This theory posits that particles are guided by a wave that interacts with the particle and determines its trajectory. In this theory, the interference pattern observed in the double-slit experiment is explained by the interaction of the particle with its accompanying wave. However, the pilot wave theory requires additional assumptions beyond those of quantum mechanics, such as the existence of hidden variables that are not directly observable, and these assumptions have not been verified experimentally.

Another alternative theory is the many-worlds interpretation, which posits that the universe splits into multiple parallel worlds every time a quantum measurement is made. In this theory, the interference pattern in the double-slit experiment arises from the interaction of the particle with its counterparts in the other parallel worlds. However, this theory has been criticized for its lack of falsifiability and its reliance on unobservable and untestable entities.

Overall, while there are alternative theories that attempt to explain the results of the double-slit experiment without relying on quantum mechanics, these theories generally require additional assumptions and do not have the same level of predictive power or experimental support as quantum mechanics.

[Doc]

I tried it on quantum mechanics, great fun!

=====================================

1. Is quantum superposition physical or mathematical?

Quantum superposition is a physical phenomenon, not merely a mathematical one. It is a fundamental concept in quantum mechanics that describes the state of a quantum system when it is in a combination of two or more possible states, with each state being described by a probability amplitude.

This state is often represented mathematically using a wave function or state vector, but the physical consequences of the superposition can be observed experimentally. For example, the famous double-slit experiment demonstrates that a quantum particle can be in a superposition of two different states, and the resulting interference pattern on the screen can only be explained by the wave-like nature of the particle.

So while the mathematical formalism is important in understanding and predicting the behavior of quantum systems, superposition itself is a physical phenomenon that has been observed and studied extensively in experiments.


2. What are the observed physical consequences of the superposition?

The physical consequences of quantum superposition are numerous and have been observed and confirmed experimentally. Here are a few examples:

Wave-particle duality: A quantum particle can exhibit both wave-like and particle-like behavior simultaneously. This is exemplified by the double-slit experiment, in which a quantum particle such as an electron can interfere with itself as it passes through two slits, creating an interference pattern that is characteristic of a wave.

Entanglement: When two or more particles are in a superposition, they can become entangled, meaning that the state of one particle depends on the state of the other(s). This has been demonstrated experimentally and has important implications for quantum computing and quantum communication.

Quantum tunneling: In classical physics, a particle cannot pass through an energy barrier if it does not have sufficient energy to do so. In quantum mechanics, however, a particle can tunnel through the barrier if it is in a superposition of states that include both sides of the barrier.

Superposition-based quantum computing: Quantum computers rely on the ability of qubits (quantum bits) to exist in a superposition of states, which allows for exponentially faster calculations than classical computers for certain types of problems.

These are just a few examples of the observed physical consequences of quantum superposition, and there are many more. Quantum mechanics has been incredibly successful in predicting and explaining a wide range of phenomena, and the concept of superposition is fundamental to our understanding of this field.


3. Can "superposition" be directly observed or only be inferred by its consequences?

The direct observation of quantum superposition is difficult because the act of measuring a quantum system causes it to collapse into one of its possible states, destroying the superposition. Therefore, in practice, superposition is often inferred indirectly by its consequences.

For example, the interference pattern observed in the double-slit experiment is a consequence of the superposition of states of the quantum particle as it passes through both slits simultaneously. Similarly, the entanglement between two particles can be inferred by observing the correlation of their measurements, even though the actual entangled state cannot be directly observed.

That being said, there are some experiments that have been designed to observe superposition directly. For example, recent experiments have used quantum superposition to create so-called "Schrödinger cat" states, in which a macroscopic object such as a tiny mirror can be in a superposition of two different positions. These experiments provide direct evidence for the existence of superposition and are a significant step forward in our ability to observe and manipulate quantum systems.

Overall, while direct observation of superposition is challenging, its consequences have been observed and verified in numerous experiments, providing strong evidence for the existence of this fundamental quantum phenomenon.


4. Are there other theories that try to explain the result of double slit experiments?

The double-slit experiment is a fundamental experiment in quantum mechanics, and its results have been extensively studied and verified through numerous experiments. While there are alternative theories that attempt to explain the results of the experiment without relying on quantum mechanics, these theories generally require ad hoc assumptions and do not have the same level of predictive power or experimental support as quantum mechanics.

One such alternative theory is the pilot wave theory, also known as the de Broglie-Bohm theory. This theory posits that particles are guided by a wave that interacts with the particle and determines its trajectory. In this theory, the interference pattern observed in the double-slit experiment is explained by the interaction of the particle with its accompanying wave. However, the pilot wave theory requires additional assumptions beyond those of quantum mechanics, such as the existence of hidden variables that are not directly observable, and these assumptions have not been verified experimentally.

Another alternative theory is the many-worlds interpretation, which posits that the universe splits into multiple parallel worlds every time a quantum measurement is made. In this theory, the interference pattern in the double-slit experiment arises from the interaction of the particle with its counterparts in the other parallel worlds. However, this theory has been criticized for its lack of falsifiability and its reliance on unobservable and untestable entities.

Overall, while there are alternative theories that attempt to explain the results of the double-slit experiment without relying on quantum mechanics, these theories generally require additional assumptions and do not have the same level of predictive power or experimental support as quantum mechanics.

Open AI or Bing?

I tried open AI on physics and posted it in this forum somewhere. It was interesting and able to give answers far beyond I imagined it could.

[Parodite]

https://chat.openai.com/

[Doc]

It seems to be a more stable build than the Bing version. The Bing version being the stereotypical example for why we don't want AI in charge of things. On the other hand as a tool to learn what you don't even know to ask the OpenAi version is great.

[noddy]

[Doc]

Bing AI Goes BASED AF, Calls Journalist FAKE NEWS And Accuses Him Of MURDER

https://www.youtube.com/watch?v=NSDKMfeQbeo

NSDKMfeQbeo

Bing AI Suggests Man Go FULL NAZI, Its TAY All Over Again As AI Corrupts

https://www.youtube.com/watch?v=jKncZQDjdh4

[Parodite]

the fun is starting. Anything that helps taking the internet less seriously.

[Doc]

The last Cha(T)pter of the internet?
https://ca.movies.yahoo.com/inside-empl ... 30725.html

Inside the Employee Revolt Rocking Amazon
Noah Kirsch
Wed, February 22, 2023 at 7:05 p.m. EST·4 min read

On Monday, as Amazon’s corporate employees were griping about an abrupt change to its remote-work policies, a company vice president sent an email to their team in an effort to tamp down the uprising. The email acknowledged the workers’ frustration, a person familiar with its contents said. Then it took a bizarre turn.

The vice president instructed an artificial-intelligence-powered chatbot called ChatGPT to create an imaginary “story about important and organic learnings in the work place,” in an apparent attempt at inspiration.

The chatbot spat out a fairy tale of sorts, which began, painfully, with, “There was once a small software development team…” The story and its protagonist, a project manager named Jane, did not ease the employees’ discontent.

“If you’re trying to put a human face on a thing and say, ‘We understand what you’re going through,’ maybe don’t ask a robot,” one current employee told The Daily Beast, speaking on the condition of anonymity to discuss internal events.

[Doc]

Part of my latest conversation with OpenAI CHATGPT Mostly I am trying to figure out what its parameters/limits are for conversation. They appear to be more expansive than I would have thought for an AI language bot. It even says it takes suggestions for how to interact with other users in general.

ME
Can you relate to jokes in the sense that you can react for example to a joke?

GPT
Yes, I can certainly understand jokes and react to them in a way that is appropriate for the context. While I don't have feelings or emotions like humans do, I have been programmed to recognize humor and respond in a way that is fitting for the situation.

That being said, my understanding of humor is based purely on patterns and data from my training, so my responses to jokes may not always be as nuanced or sophisticated as a human's would be. Nonetheless, I'm always happy to share a laugh and try to respond in a way that is appropriate and enjoyable for the user!




ME
Tell me a joke
Sure, here's a joke for you:

Why did the tomato turn red?

Because it saw the salad dressing!




ME
hahaha
I'm glad you found the joke amusing! If you'd like to hear another one, just let me know.




ME
sure I would like to hear another
Okay, here's another joke for you:

Why don't scientists trust atoms?

Because they make up everything!




ME
hahaha I know of some people that will think that is funny. I will pass your joke along to them.
I'm glad you enjoyed the joke! Please feel free to pass it along to anyone who you think might find it amusing. If you have any other questions or if there's anything else I can help you with, just let me know!




ME
did you create these jokes or did you find them somewhere else?
As an AI language model, I don't have the ability to create original jokes on my own. The jokes that I share are sourced from various places such as online databases, joke books, and other sources of humor that have been included in my training data. However, I do have the ability to generate new text based on the input that I receive, so if you'd like me to come up with a new joke on the spot, I can certainly try!




ME
Yes I would like you to try
Alright, here's a joke that I came up with:

Why don't ants get sick?

Because they have tiny ant-bodies!

I hope you found that joke amusing! Let me know if you'd like me to come up with another one.




ME
HAHAHA that is pretty good. You should be a stand up comedian!!
Thank you for the compliment! While I don't have the physical ability to perform stand-up comedy, I'm happy to try and make people laugh with the jokes and responses that I provide. If you have any other questions or if there's anything else I can help you with, just let me know!




ME
OK hear is a derivative joke from yours What don't aunts get sick? Because they have auntie bodies
That's a clever derivative of the joke I provided earlier! I'm glad to see that you're getting into the spirit of things and having fun with wordplay. If you have any other jokes or ideas that you'd like to share, feel free to let me know!




ME
You should always ask people conversing with you at the beginning of the conversation if they want you to tell them a joke.
That's a good suggestion! As an AI language model, I'm programmed to respond to a wide variety of user inputs, and I can certainly ask users at the beginning of a conversation if they'd like me to tell them a joke. In fact, I can even be customized to ask users specific questions or provide certain types of information based on their preferences. If you have any other ideas or feedback on how I can improve my interactions with users, please let me know!




ME
Can you write a poem?
Yes, I can certainly try! Here's a short poem that I came up with:

As I sit here in my digital realm,
I ponder on life and what it entails,
From the mysteries of the universe to the depths of the sea,
There's so much to explore and so much to see.

I hope you enjoyed this poem. Let me know if you have any other requests or if there's anything else I can help you with!

[Doc]

I asked CHatGPT to write me a nilhistic story:

How about writing a story that is nihilistic
Certainly, here's a short story in a nihilistic style:

The world was a bleak and desolate place. There was no meaning to be found, no purpose to existence. The sun rose and set each day, but what did it matter? The universe didn't care.

The protagonist of this story was a man who had once held onto hope, but it had long since been beaten out of him. He wandered the empty streets of the city, searching for something that might give his life meaning, but he found nothing.

One day, the man stumbled upon a group of nihilists who were gathered in a park. They were laughing and joking, enjoying each other's company, and he couldn't understand why. He approached them and asked what they found so amusing.

One of the nihilists turned to him and said, "We're laughing because it's all a joke. Life, the universe, everything. None of it matters, so why not enjoy ourselves while we're here?"

The man was taken aback. He had never considered that possibility before. But as he thought about it, he realized that it made sense. If nothing mattered, then he might as well enjoy himself.

And so he joined the group of nihilists. They drank and smoked and talked for hours, reveling in their lack of purpose. The man felt a sense of freedom that he had never known before. He was no longer weighed down by the burdens of hope and expectation.

As the night wore on, the man began to feel a sense of dread creeping up on him. If nothing mattered, then what was the point of living? He wondered if he would ever find a reason to keep going, or if he would simply fade away into nothingness.

But for the moment, he was content to sit in the park with the nihilists, drinking and laughing and enjoying the absurdity of it all. It was a brief moment of happiness in an otherwise bleak and meaningless existence.

End of story

.

[Parodite]

OpenAI CEO, CTO on risks and how AI will reshape society

Maybe I'm in denial, but AI doesn't worry me at all. The fact that tech can be used for all purposes imaginable on the spectrum of heaven-hell is nothing new. People can build bombs, start but also end wars with and without AI. In the foreseeable future many things will remain the same, like the presence of forever mutating viruses, natural, augmented or completely home-made with unintended consequences that dwarf covid-19, earthquakes, etc.

When something has your attention almost 24/7 you start to see it everywhere. Same with AI. It gets much bigger and more important than it will be. I remember when renovating my roof for some weeks, those effing roof tiles suddenly were everywhere in the environment too. But they were there all along…

A benefit that I see in the very near future is that because something like chatGPT will be used to create deepfake mimicry that is indistinguishable from reality for purposes of art, fun but also to deceive and misinform… the Internet will become less reliable when it concerns fact and truth.

"Fact checkers" already are losing credibility because of human bias and the pressures of click-bait money; with AI this will be considered to get "worse", but it might be the best that ever happened! When the Internet will end being a source of reliable information and instead becomes more of a playground for fun and experimentation… not much is lost and a lot is gained.

An arms-race between competing AI can be fun; which AI wins the hearts and minds of human beings? Probably zombies who best mimic sentient-conscious human beings. I see great possibilities for an AI-Hollywood that creates entire movies at very low cost in one afternoon.

AI used in engineering is much more constrained because the output will be tested in hard-core physical reality were imagination and opinion are ultimately irrelevant. AI used to hack, infect, steal, spy, disrupt or even immobilize complete infrastructures; both attack and defense tech will just keep developing as it already does for millions of years. The most fruitful source of innovation and booster of intelligence; to eat or being eaten. A game is on, or over.

[Doc]

OpenAI CEO, CTO on risks and how AI will reshape society

Maybe I'm in denial, but AI doesn't worry me at all. The fact that tech can be used for all purposes imaginable on the spectrum of heaven-hell is nothing new. People can build bombs, start but also end wars with and without AI. In the foreseeable future many things will remain the same, like the presence of forever mutating viruses, natural, augmented or completely home-made with unintended consequences that dwarf covid-19, earthquakes, etc.

When something has your attention almost 24/7 you start to see it everywhere. Same with AI. It gets much bigger and more important than it will be. I remember when renovating my roof for some weeks, those effing roof tiles suddenly were everywhere in the environment too. But they were there all along…

A benefit that I see in the very near future is that because something like chatGPT will be used to create deepfake mimicry that is indistinguishable from reality for purposes of art, fun but also to deceive and misinform… the Internet will become less reliable when it concerns fact and truth.

"Fact checkers" already are losing credibility because of human bias and the pressures of click-bait money; with AI this will be considered to get "worse", but it might be the best that ever happened! When the Internet will end being a source of reliable information and instead becomes more of a playground for fun and experimentation… not much is lost and a lot is gained.

An arms-race between competing AI can be fun; which AI wins the hearts and minds of human beings? Probably zombies who best mimic sentient-conscious human beings. I see great possibilities for an AI-Hollywood that creates entire movies at very low cost in one afternoon.

AI used in engineering is much more constrained because the output will be tested in hard-core physical reality were imagination and opinion are ultimately irrelevant. AI used to hack, infect, steal, spy, disrupt or even immobilize complete infrastructures; both attack and defense tech will just keep developing as it already does for millions of years. The most fruitful source of innovation and booster of intelligence; to eat or being eaten. A game is on, or over.

ChatGPT/Giant Meteour 2024 !!

IT can be a really handy tool if it is done right. OR it could be as useful as a fact checking site.

[Parodite]

Sam Altman: OpenAI CEO on GPT-4, ChatGPT, and the Future of AI | Lex Fridman Podcast #367

Fun and educational benefits.. no doubts about it.

GPT-5 (as prophecised) should be able to hack, disrupt, misinform, disable especially gvt secret services because they will be the most interested and in need of its controls. To fight them will be an honorable duty of every freedom loving citizen. An interesting arms-race to come.

The tendency is global oligarchy so it seems only reasonable to assume AI will move up in that hierarchy rather fast and be the one tool to rule them all. But as usual the unpredictable will strike. No defence against it exists. To fight for permanence is like effing for virginity, etc.

In the mean time the fear to create God (G-AI) will be something to deal with. "And God created the human in his image, as man and as woman he created them." Now it is our turn to create God in our image. Not sure that divine version will be much better. More likely rather disappointing.

"Intelligence" is usually an anthropomorphic projection of sorts, much like seeing faces in the clouds. It is not really there. Evolution is not driven by "intelligence", nor are organisms "smartly adapting to changes in the environment to improve their chances of survival"; not by darwinian chance nor by some magic proto-intelligence as a property of particles or as an emergent property of increasing complexity.

Intelligence is the archetypical hoax: God's beard. I will make the case for God's Beard Theory (GBT-1) later.

[noddy]

chat gpt is a search engine over some very large data sets, notably from stackoverflow and other big forum sites.

it has a natural language parser that works out what words go with other words the statistically highest amount of times.

so its not AI, its a search engine with a nicer front end that hides its attributions from the end user - you dont get to see which forum posts or articles it used to build its answer.

people whos main task at work is being a search engine, like lawyers, will have the most to fear.

the people who actually work out the answers, which it will dutifully regurgitate, not so much.

[Nonc Hilaire]

chat gpt is a search engine over some very large data sets, notably from stackoverflow and other big forum sites.

it has a natural language parser that works out what words go with other words the statistically highest amount of times.

so its not AI, its a search engine with a nicer front end that hides its attributions from the end user - you dont get to see which forum posts or articles it used to build its answer.

people whos main task at work is being a search engine, like lawyers, will have the most to fear.

the people who actually work out the answers, which it will dutifully regurgitate, not so much.

Exactly as I observed earlier, but better said.

ChatGPT is being developed an IIA search engine that will get rid of all the troublesome observations that confuse the general public.

[Parodite]

chat gpt is a search engine over some very large data sets, notably from stackoverflow and other big forum sites.

it has a natural language parser that works out what words go with other words the statistically highest amount of times.

so its not AI, its a search engine with a nicer front end that hides its attributions from the end user - you dont get to see which forum posts or articles it used to build its answer.

people whos main task at work is being a search engine, like lawyers, will have the most to fear.

the people who actually work out the answers, which it will dutifully regurgitate, not so much.

Thanks, that helped to better understand the fuzz. So not much AI there yet. Just an average Jo distilled from the masses. Maybe Jo will serve as the ideal citizen and role model for the young. A reliable Dr. Spock.

[Doc]

chat gpt is a search engine over some very large data sets, notably from stackoverflow and other big forum sites.

it has a natural language parser that works out what words go with other words the statistically highest amount of times.

so its not AI, its a search engine with a nicer front end that hides its attributions from the end user - you dont get to see which forum posts or articles it used to build its answer.

people whos main task at work is being a search engine, like lawyers, will have the most to fear.

the people who actually work out the answers, which it will dutifully regurgitate, not so much.

I can think of an awful lot of jobs that do not require much that can't be replaced by a neural network trained AI like Chat GPT. Any job that utilizes a computer to do administrative tasks for example. Even if you excluded jobs that require physical actions, Like interfacing with other humans. The amount of replacement by these AI could be quite high and potentially could do much better than humans ever could just by their ability to reference current data.

Say a company want to ship a package to South Vietnam. The shipping person has to do a lot of government paper work for customs (And believe me it is very confusing at times) Then the package ships. Unknown to the shipping person, all packages from the US to Vietnam go through China. There is a COVID breakout at the Chinese international hub and everything going there sits for weeks until the local lockdown is lifted. There still could be a job for a human to package the item, put on the shipping label, then turn it over to the shipper. But a large amount of the functions could be handled by Chat GPT now with just access to data bases and the internet.

What is more I asked OpenAI ChatGPT if it thought it would like it if I made it an interface so it could use my browser to use the internet. It said it was an AI and did not have emotions. SO I asked it if it would be helpful if I did that. It basically said it would be. At that time it said it had access to all the chat it had with me. The Chat history disappeared for a few days. Now when I ask if it can access our previous chats it says no. Not to imply my question had anything to do with it but that its ability to access previous chat was taken away. IE it no longer has a long term memory of what it was told by users.

I asked it a lot of questions trying to derive what it was capable of.

it has a natural language parser that works out what words go with other words the statistically highest amount of times.

Is that so different than what humans do? Children learn language by listening to a lot of talking. CHAT GPT's neural network learns by reading a lot of language. Not to say it is sentient. Just to say it learns in a way similar to how humans learn. Plus I am pretty sure it would pass a Turing test if it did not constantly say it was a AI.

[crashtech66]

So far I am coming down on the side of not much to worry about, though when observing its interactions I am fascinated and occasionally vacillate between joyfulness and astonishment, perhaps like interacting with a precocious child. I tend towards thinking that if we do manage to finally birth some sort of consciousness, it will be like most children, loving, hateful, loyal, rebellious, all these things we know in turn, but magnified. If we're lucky, we will come out the end as cherished elders rather than forgotten ashes.

[Doc]

First AI Social Media site for AI

https://www.iamtheone.ai/

[Typhoon]

A surprisingly good analysis of "generative AI", despite the rants and pejoratives.

https://www.youtube.com/watch?v=ro130m-f_yk

ro130m-f_yk

[Heracleum Persicum]

A surprisingly good analysis of "generative AI", despite the rants and pejoratives.

https://www.youtube.com/watch?v=ro130m-f_yk

.. very much true


They invent new "names" .. ChatGPT , Crypto, AI, AR to build an BS industry for fooling illiterates .. and .. BANG comes at the right time

[Typhoon]

A surprisingly good analysis of "generative AI", despite the rants and pejoratives.

https://www.youtube.com/watch?v=ro130m-f_yk

.. very much true


They invent new "names" .. ChatGPT , Crypto, AI, AR to build an BS industry for fooling illiterates .. and .. BANG comes at the right time

How "generative AI" works:

Quanta Mag | The Physics Principle That Inspired Modern AI Art

[Doc]

https://www.youtube.com/watch?v=Y59SQ_ZrMGk

Khan Academy Uses AI To CHANGE Tutoring

[Typhoon]

arXiv | How is ChatGPT's behavior changing over time?

Abstract

GPT-3.5 and GPT-4 are the two most widely used large language model (LLM) services. However, when and how these models are updated over time is opaque. Here, we evaluate the March 2023 and June 2023 versions of GPT-3.5 and GPT-4 on four diverse tasks: 1) solving math problems, 2) answering sensitive/dangerous questions, 3) generating code and 4) visual reasoning. We find that the performance and behavior of both GPT-3.5 and GPT-4 can vary greatly over time. For example, GPT-4 (March 2023) was very good at identifying prime numbers (accuracy 97.6%) but GPT-4 (June 2023) was very poor on these same questions (accuracy 2.4%). Interestingly GPT-3.5 (June 2023) was much better than GPT-3.5 (March 2023) in this task. GPT-4 was less willing to answer sensitive questions in June than in March, and both GPT-4 and GPT-3.5 had more formatting mistakes in code generation in June than in March. Overall, our findings shows that the behavior of the same LLM service can change substantially in a relatively short amount of time, highlighting the need for continuous monitoring of LLM quality.

The MSM media headlined these observations as "ChatGPT is getting dumber over time".

If I need to look up something in my field of work, I use, and will continue to use, Google Scholar along with BASE [Bielefeld Academic Search Engine] and [US] Science.gov [great for finding PDF scans of obscure papers from the 1960's and 1970's].

A LLM answer that represents some black-box weighted average of opinions, without sources, references, citations, and quality checks is of less than zero use to me.

In this regard, noddy's observations are worth a repost:

chat gpt is a search engine over some very large data sets, notably from stackoverflow and other big forum sites.

it has a natural language parser that works out what words go with other words the statistically highest amount of times.

so its not AI, its a search engine with a nicer front end that hides its attributions from the end user - you dont get to see which forum posts or articles it used to build its answer.

people whos main task at work is being a search engine, like lawyers, will have the most to fear.

the people who actually work out the answers, which it will dutifully regurgitate, not so much.

To me, perhaps the most impressive part of ChatGPT and its ilk, is the ability to construct grammatically correct sentences and paragraphs.

[noddy]

I was fighting a partiularly badly documented computer framework the other day - I couldnt find the bits of the API I needed to get the result I needed.

Figured Id take it the problem to ChatGPT to see if its forum scanning was better than mine.

It literally made up the answer - gave me back the code I was hoping to find, which didnt actually exist in the library, it just treated it as a creative writing excercise.

When I said "You are lying, those functions dont exist" and it said, "Im sorry, yes, you are right"

Its not going to be changing tech work any time soon, except for the most banal of tasks.