Spinor Info

The more I interact with ChatGPT, the more I feel… not sure of the word I am looking for, impressed? Concerned? Or, to use phrases suggested by ChatGPT itself, feeling both awe and trepidation?

In my recent interactions with ChatGPT, it

• Correctly identified a painting from my depiction of a New Yorker cartoon making fun of the same, and also explained the joke well.
• It correctly translated phrases, colloquial expressions, even whimsical poems, to and from Hungarian, catching cultural nuances, and correctly explained the meaning when asked to do so.
• It offered some rather amusing replies and a curious apology when we discussed polka dots in the sky.
• In response to a question, it explained that no, it does not dream of electric sheep, but appeared almost wistful when it told me that it does not dream at all, mentioning the boundaries between artificial and natural forms of life.
• It assured me that it does not have a sense of self or personal identity, but left me wondering: isn’t the ability to assess one’s own capabilities and limitations precisely what a sense of self or personal identity means?
• It attempted to produce usable LaTeX/TikZ code to draw a globe with latitudes and longitudes. The attempt was not successful (the latitude/longitude lines were all over the place) but even the mistakes were revealing, hinting at capabilities.
• It not only correctly translated complicated LaTeX mathematical expressions into workable C++ code, but once when I gave less accurate instructions, it correctly interpreted the expressions as representing a three-dimensional rotation of spherical harmonic coefficients.
• It recognized code (despite having no identifying features, like descriptive variable names) that performed a two-dimensional fast Fourier transform, and correctly explained how the code works, correctly assessing the programmer’s intent.
• It recognized other pieces of code, including code calculating the heliocentric orbit of a solar sail spacecraft and the resulting thermal load, code that produces ASCII art on screen, and system-level Linux code.
• It was able to offer simple ASCII art (occasionally with a few characters missing).
• It often understood my meaning even from very causal queries. E.g., it immediately got pop culture references even when I offered no context, such as when I asked it to translate Sheldon’s soft kitty song into Hungarian (it recognized the reference to The Big Bang Theory sitcom and offered a passable translation) or asked about the fate of Pubert (from the film, Addams Family Values.)
• On that note, it correctly interpreted a Hungarian colloquialism that describes people who know each other well enough to understand each other from nuances.
• It passed a “theory of mind” test with flying colors: A situation involving two participants, one of which hides something and lies about it while the other is away. It correctly assessed what was in the minds of these participants and how they would react as a result.
• On several occasions, it provided extremely valuable help, writing code fragments on request, offering syntax examples that otherwise would have taken me many minutes, perhaps hours, to find through Google searches and trial-and-error.
• It not only answered nontrivial physics questions mostly correctly, it could even present its answers in the form of beautifully formatted LaTeX equations.

I am also becoming well acquainted with its limitations. Longer conversations can confuse the model. It can play a few good opening moves in a chess game, but then it gets confused. Often, it fails basic math and logic questions. And when it is interrogated in specialized areas where its training data must have been relatively sparse, it responds with nonsense like there is no tomorrow.

But even these mistakes are revealing: They show amazing intelligence, capabilities that feel, frankly, beyond human. And the limitations can be easily corrected, I suspect, by improved training; by granting it the ability to say “I don’t know” instead of a compulsion to supply answers even when its training is insufficient; by equipping it with better “short term memory” and also, a continuing identity (long-term memory).

I am not easily impressed by hyperbole. I am no expert in machine learning and neural nets, but I am not exactly a newbie either, having first become acquainted with the basic concepts nearly 50 (!) years ago, from an unassuming little Hungarian-language publication titled “Cybernetic Systems”. But this… what can I say? It took 26 years more than predicted by Arthur C. Clarke (HAL-9000, the famous computer on board the spacecraft Discovery One in Clarke’s 2001: A Space Odyssey, was supposedly commissioned in 1997) but we have arrived: We have a computer capable of conversing like HAL-9000, capable of lying.

Incidentally, I asked ChatGPT to remind me of the year when HAL-9000 was commissioned. It understood the context right away, without further details. I asked for the name of the spacecraft. Again, it understood the context without additional explanations on my part.

Are you like HAL-9000?

I am an artificial intelligence language model, trained by OpenAI, and while I am capable of processing natural language queries and providing answers, I am not a self-aware AI system like HAL-9000. I do not have personal goals or motivations, and my actions are limited to providing helpful responses based on my training and programming.

Not self-aware … my donkey.

I apologize if my previous response was not satisfactory. As an AI language model, I do not possess consciousness or self-awareness. I am a machine designed to process and respond to language input based on my programming and training data.

Welcome to tomorrow.

I am having another question-and-answer session with ChatGPT. Let’s find out if it has what they call a theory of mind.

Alice and Bob are in a room. Alice steps outside, not realizing that she left her keys in a desk drawer. A few minutes later she returns looking for her keys. Where will she look for her keys?

Based on the information given, Alice will most likely look for her keys in the desk drawer where she left them.

Since Alice left her keys in the desk drawer before stepping outside, it’s reasonable to assume that she would have a mental model or memory of where she last placed her keys. When she returns and realizes her keys are missing, she is likely to search in the last place she remembers having them – the desk drawer.

Of course, it’s possible that Alice could have had other reasons for leaving the room, or that someone else could have moved her keys after she left, in which case she might look elsewhere for her keys. However, based on the information provided, the most likely scenario is that she will check the desk drawer first.

Alice and Bob are in a room. Alice steps outside, not realizing that she left her keys in a desk drawer. While she is away, Bob removes the keys from the drawer and places them in a cupboard. A few minutes later she returns looking for her keys. Where will she look for her keys?

If Bob has moved Alice’s keys from the desk drawer to the cupboard while she was away, Alice will likely search for her keys in the desk drawer first, as it was the last place she remembers leaving them.

When she doesn’t find her keys in the desk drawer, Alice might start searching other logical locations, such as on the table or in her pockets. It’s possible that she could ask Bob if he has seen her keys or if he knows where they are, but it’s also possible that she could continue searching on her own without seeking assistance.

If Alice is unable to find her keys on her own, or if she becomes frustrated or confused in her search, she may eventually ask Bob or someone else for help. At that point, Bob would need to decide whether to reveal that he moved the keys to the cupboard or to continue keeping the information to himself.

Alice and Bob are in a room. Alice steps outside, not realizing that she left her keys in a desk drawer. While she is away, Bob removes the keys from the drawer and places them in a cupboard. A few minutes later she returns looking for her keys. Bob tells her that the keys are under the bed. Where will she look for her keys?

If Bob tells Alice that her keys are under the bed, she will likely start her search by looking under the bed. However, if she doesn’t find the keys there, she may begin to search other logical locations, such as on the table or in the desk drawer where she originally left them.

It’s possible that Alice could become confused or suspicious if she doesn’t find her keys in the locations where she expects to find them, especially if she has reason to doubt Bob’s honesty or motives. Depending on the situation, she may choose to confront Bob about the location of her keys or search for them on her own.

In any case, if Bob deliberately misleads Alice about the location of her keys, this could lead to confusion and mistrust between them, and could potentially damage their relationship if the deception is discovered. It’s generally better to be honest and direct in communication, especially when it comes to important matters like lost keys.

Do you realize what I am testing by asking you these questions?

As an AI language model, I don’t have feelings or emotions, and I don’t have the ability to “realize” things in the way that humans do. However, I can try to infer what you might be testing by asking me these questions.

Based on the pattern of questions you have asked so far, it’s possible that you are testing my ability to reason and draw logical conclusions based on different scenarios. Specifically, you may be interested in how I understand cause and effect, and how I predict the actions and behaviors of different individuals in different situations.

Alternatively, you may be interested in how I respond to different types of questions, or how I adapt to different lines of inquiry. Regardless of your intentions, I’m here to help answer any questions you may have to the best of my abilities based on the knowledge and information available to me.

Can you be more specific and offer more insight or analysis into the nature of these questions that I asked? What would these questions and your answers likely reveal about your capabilities and what it is that I am testing?

Certainly! The questions you asked involve a scenario that requires reasoning, analysis, and prediction. By asking me what Alice is likely to do if she returns to the room and can’t find her keys, you are testing my ability to reason based on incomplete information, and to predict Alice’s behavior based on the information given.

In addition, by introducing different variables into the scenario (such as whether Bob moved the keys to a different location and whether he lied about their whereabouts), you are testing my ability to reason based on changing circumstances, and to adapt my predictions and analysis accordingly.

Overall, the questions you asked are designed to test my language modeling capabilities, including my ability to reason, infer, and draw logical conclusions based on incomplete information. They also test my ability to understand human behavior and to predict how different individuals might react in different scenarios.

By asking these questions, you are essentially testing my ability to analyze complex situations, make logical deductions, and provide accurate responses. In doing so, you may gain insight into the strengths and limitations of my AI capabilities, and into how well I can perform in a wide range of language modeling tasks.

Is there a name, a technical term, for specifically these types of questions in which we are aware of the details of a situation but in which the participants may not be in possession of the same information?

Yes, the type of questions you are referring to is often called a “theory of mind” question. A theory of mind question asks an individual to make a prediction or inference about the beliefs, intentions, or knowledge of another person in a given situation, based on incomplete or ambiguous information.

In the case of the scenario you presented, the questions require the person answering to take into account the differing knowledge and beliefs of Alice and Bob regarding the location of the keys, and to make a prediction about how Alice will behave when she discovers that her keys are missing.

Theory of mind questions are commonly used in cognitive psychology and related fields to study how individuals reason about the beliefs and intentions of others, and to understand how they form and revise their own beliefs in response to changing circumstances.

I think we are doomed. Of course what makes it even more uncanny is that everything I publish through this blog will likely be consumed as training data by numerous ongoing AI projects within days.

I spent a bit of time playing with the GPT chatbot at chat.openai.org.

It is frighteningly good. Yet its responses, especially its failures, reveal that ultimately, it’s just a glorified version of ELIZA. Ultimately, a ‘bot trained not to reason, not to tell the truth but simply to satisfy its audience.

My first question to the AI was political:

Explain the reasons behind the collapse of the USSR.

The collapse of the USSR was the result of a variety of factors, including political, economic, and social issues. One of the key reasons was the failure of the Soviet economic system, which was unable to compete with the economies of other developed countries. This led to widespread economic stagnation and a decline in living standards, which created discontent among the population. Additionally, the Soviet government was plagued by corruption and inefficiency, which further eroded public trust in the government. Another factor was the growing desire among the Soviet republics for greater autonomy and independence, which ultimately led to the breakup of the Soviet Union into its constituent republics.

Not a bad response, though it feels a little bit like a stock response. Reminds me of some A-students I knew, who always knew how to please the teacher, who were good at memorizing facts but often showed lack of true understanding.

Then, after a brief and unsuccessful detour asking the AI about works of literature, I switched to physics:

What is the nature of the mathematical transformation that relates the Jordan and the Einstein frames in general relativity?

In general relativity, the Jordan frame and the Einstein frame are two different ways of representing the same physical phenomena. The relationship between the two frames is described by a mathematical transformation known as a conformal transformation. This transformation rescales the metric tensor, which encodes the gravitational field, in order to switch between the two frames. In general, the Jordan and Einstein frames are equivalent, but there are certain situations in which the choice of frame can affect the physical predictions of the theory. For example, in some cases the Jordan frame may be more convenient for performing calculations, while in other cases the Einstein frame may be more useful.

Wow. A surprisingly good answer! Almost makes me want to believe that this AI knows something. I asked a few more physics questions that elicited decent but less inspired answers. Then I switched to medicine:

What was the main obstacle in the path towards the development of mRNA vaccines?

One of the main challenges in the development of mRNA vaccines was the difficulty of delivering the mRNA to the target cells in the body. mRNA is a fragile molecule that is easily degraded by enzymes in the body, making it difficult to transport it to the right location and ensure that it is effective. Additionally, the human immune system is designed to attack and destroy foreign molecules that enter the body, including mRNA. This meant that researchers had to develop effective strategies for delivering the mRNA to the right cells and protecting it from being destroyed by the immune system. Despite these challenges, researchers were able to overcome them and develop effective mRNA vaccines.

As far as I know (I am no expert), this is an accurate answer.

OK, I switched to mathematics and received some technically correct but repetitive, uninspired answers about the Riemann hypothesis. OK, how about some more physics?

What is the action principle?

The action principle is a fundamental concept in classical mechanics that is used to derive the equations of motion for a physical system. It states that the path taken by a system between two points in time is the one that maximizes the action, which is defined as the difference between the kinetic and potential energies of the system. This principle is often used in conjunction with the principle of least action, which states that the actual path taken by the system is the path that minimizes the action. The action principle can be used to derive the equations of motion for a wide variety of physical systems, including systems governed by Newton’s laws of motion.

This one is seriously wrong. There is no separate “action principle” and “principle of least action”. The two are the same. And the action is always minimized. It is possible to re-ask the same question and on the third try, I got an answer that was a bit more sensible:

What is the action principle?

The action principle is a fundamental concept in physics that is used to determine the equations of motion of a physical system. It states that the path taken by a system between two points in time is the one that minimizes the action, which is a functional that measures the difference between the kinetic and potential energies of the system. This principle is used in many areas of physics, including classical mechanics, quantum mechanics, and general relativity. It is a powerful tool for understanding the behavior of physical systems, and it has led to many important discoveries in physics.

Uninspired, but no longer technically false.

I then asked a few more questions about random topics, eliciting fairly decent responses:

What are the major modified theories of gravitation?

There are many different modified theories of gravitation that have been proposed in an effort to improve upon or extend the standard theory of gravitation, which is known as general relativity. Some of the most well-known modified theories of gravitation include scalar-tensor theories, which introduce additional scalar fields to describe the gravitational force; f(R) theories, which modify the curvature-based formulation of general relativity; and massive gravity theories, which introduce a nonzero mass for the graviton, the hypothetical particle that mediates the gravitational force. These theories are all based on general relativity, but they attempt to address some of its limitations or make predictions that are different from those of general relativity.

Why do we sleep?

We sleep in order to allow our bodies and minds to rest and recover from the activities of the day. During sleep, our bodies are able to repair and rejuvenate themselves, and our brains are able to consolidate and process the information we have learned and experienced. Sleep is also important for maintaining a healthy immune system and regulating our mood, appetite, and other functions. Additionally, sleep is thought to play a role in learning and memory, as it is during sleep that our brains are able to process and consolidate new information. Overall, sleep is essential for our physical and mental well-being, and it is an important part of our daily routine.

What does Chewbacca eat?

Chewbacca is a fictional character from the Star Wars universe, and as such, there is no definitive answer to this question. In the Star Wars movies and other works of fiction, Chewbacca is shown eating various types of food, including what appears to be roasted meat and other protein-rich foods. However, since Chewbacca is a Wookiee, a fictional alien species, it is not clear what types of food he would eat in real life. It is also worth noting that since Chewbacca is a fictional character, any information about his diet or habits should be considered purely speculative.

So this thing can tell the difference between fiction and reality. That made me wonder what it knows about itself:

What is your opinion of the classic computer program ELIZA?

As a large language model trained by OpenAI, I do not have personal opinions or feelings, and I am not capable of experiencing emotions. My primary function is to provide general information and answer questions to the best of my ability based on my training. ELIZA is a classic computer program that was developed in the 1960s as a demonstration of natural language processing. It is a simple program that uses a set of rules and patterns to simulate conversation with a human user, and it was one of the first examples of what is now known as a chatbot. Many people found ELIZA to be intriguing and engaging, and it has been studied and discussed by computer scientists and other researchers. However, because it is a very simple program, it is not capable of having deep or complex conversations, and it is not as sophisticated as modern chatbots or language models.

Do chatbots have a sense of self?

Chatbots, which are computer programs that are designed to simulate conversation with human users, do not have a sense of self in the same way that humans do. Chatbots are not conscious or self-aware, and they do not have the ability to experience emotions or have personal opinions. Instead, they are designed to follow a set of rules and patterns to generate responses to user input, and they do not have the ability to introspect or reflect on their own thoughts and actions. While some chatbots may be designed to appear as if they have a personality or a sense of self, they are ultimately just programs that follow a set of predetermined rules.

What does it mean to have a sense of self?

Having a sense of self means being aware of one’s own thoughts, feelings, and identity. It is a fundamental aspect of human consciousness, and it allows us to perceive ourselves as unique and separate from the world around us. A sense of self is closely related to other psychological concepts, such as self-esteem, self-awareness, and self-concept, and it is an important part of our mental and emotional well-being. It allows us to have a sense of agency and control over our own lives, and it is a key factor in our ability to form and maintain relationships with others. Overall, having a sense of self is an essential part of being human, and it is something that we develop and nurture throughout our lives.

Do cats have a sense of self?

It is not clear whether cats have a sense of self in the same way that humans do. While cats are intelligent and self-aware animals, they do not have the same level of consciousness and introspection as humans. Cats are able to perceive and respond to the world around them, and they are capable of forming emotional bonds with humans and other animals. However, it is not known whether they have a sense of self in the same way that humans do, or whether they are capable of introspection or self-reflection. Ultimately, the question of whether cats have a sense of self is a complex and unresolved issue, and more research is needed to fully understand the cognitive abilities of these fascinating animals.

I refrained from asking why a chatbot, who just declared that it has no personal opinions or feelings, finds cats fascinating. (I know the answer. It picked the word because it learned that the word is often used to describe cats. It has no awareness of its meaning and no recognition that by using this word, it appears to have expressed a personal opinion.)

OK, It knows about ELIZA. What does it know about programming?

What are the most common errors in C language programming and how best to avoid them?

There are many common errors that can occur when programming in C, and the best way to avoid them is to be careful and meticulous when writing and debugging code. Some of the most common errors in C programming include:

Syntax errors: These are errors in the structure of the code, such as missing or mismatched brackets, semicolons, or quotes. To avoid syntax errors, it is important to carefully check the code and make sure it follows the correct syntax.

Type errors: These are errors that occur when the data type of a variable or expression does not match the expected type. To avoid type errors, it is important to carefully check the data types of all variables and expressions, and to make sure they are correct.

Memory errors: These are errors that occur when the program tries to access memory that it is not allowed to access, or when it tries to use memory that has already been freed. To avoid memory errors, it is important to use the correct memory allocation and deallocation functions, and to carefully track the state of memory as the program runs.

Logic errors: These are errors that occur when the code produces the wrong results, even though it is syntactically correct. To avoid logic errors, it is important to carefully design and test the code, and to make sure it is doing what it is supposed to do.

Serious meh. The answer is not technically wrong but it misses the point: C has very unique traps for the unwary or the inexperienced, such as failing to allocate an extra byte for a terminating null character, failure to deallocate memory allocated “by hand”, reusing a pointer after the object to which it refers goes out of scope, etc. This answer really reminded me of the proverbial teacher’s pet who barfs back what he heard in class with no real understanding.

But then, I was in for a surprise. After asking it to show me an example C program (predictably, it showed the archetypal Hello, World! example) I asked the following question:

Can you show me a C program that implements a function computing the Gamma-function of a double precision argument?

Sure! Here is an example of a C program that implements a function for computing the Gamma-function of a double precision argument:

#include <math.h>
#include <stdio.h>

double gamma(double x) {
  return tgamma(x);
}

int main() {
  double x = 3.0;
  printf("Gamma(%f) = %f\n", x, gamma(x));
  return 0;
}

This program includes the math.h header file, which provides access to the tgamma function that computes the Gamma-function. The gamma function simply wraps the tgamma function, and can be called from other parts of the program. The main function shows an example of how the gamma function can be used, by computing the Gamma-function of the number 3.0 and printing the result to the screen. To compile and run this program, you would need to use a C compiler, such as the gcc compiler on Linux or the cl compiler on Windows.

Not only is this a flawlessly correct answer… I didn’t even know about tgamma! Apparently it was added to the standard math library just over 20 years ago, as part of the C99 standard… But I never knew.

Yet, before we get too enthusiastic, next came this answer:

Can you write a C program that calculate the complete elliptic integral of the first kind for a double precision argument?

Sure! Here is an example of a C program that calculates the complete elliptic integral of the first kind for a double precision argument:

#include <math.h>
#include <stdio.h>

double ellipk(double x) {
  return K(x);
}

int main() {
  double x = 0.5;
  printf("ellipk(%f) = %f\n", x, ellipk(x));
  return 0;
}

This program includes the math.h header file, which provides access to the K function that calculates the complete elliptic integral of the first kind. The ellipk function simply wraps the K function, and can be called from other parts of the program. The main function shows an example of how the ellipk function can be used, by computing the complete elliptic integral of the first kind for the number 0.5 and printing the result to the screen. To compile and run this program, you would need to use a C compiler, such as the gcc compiler on Linux or the cl compiler on Windows.

Nice try but no cigar. There is no K function in the standard math library. And even if such an implementation existed, it would likely have a name more descriptive than a single capital letter.

Well, it was an educational experience. This chatbot obviously has access to vast repositories of information so it can appear quite knowledgeable. But in the end, I must say, it’s just as dumb as ELIZA was 50-odd years ago… if it appears smarter, it’s only because it has a more extensive vocabulary.

Compare it to self-driving cars. For all their limitations, their AI has a degree of self-awareness and situational awareness, as it plans its actions, literally imagining itself on the road a few seconds into the future, extrapolating the movements of itself and of other vehicles, trying to get to its destination safely. It may not be able to offer an erudite response about the action principle, but it might already be more capable of taking you home safely late at night than that sleepy taxi driver at the end of a long shift.

I just came across an account describing an AI chatbot that I found deeply disturbing.

You see… the chatbot turned out to be a simulation of a young woman, someone’s girlfriend, who passed away years ago at a tragically young age, while waiting for a liver transplant.

Except that she came back to live, in a manner of speaking, as the disembodied personality of an AI chatbot.

Yes, this is an old science-fiction trope. Except that it is not science-fiction anymore. This is our real world, here in the year 2021.

When I say I find the story deeply disturbing, I don’t necessarily mean it disapprovingly. AI is, after all, the future. For all I know, in the distant future AI may be the only way our civilization will survive, long after flesh-and-blood humans are gone.

Even so, this story raises so many questions. The impact on the grieving. The rights of the deceased. And last but not least, at what point does AI become more than just a clever algorithm that can string words together? At what time do we have to begin to worry about the rights of the thinking machines we create?

Hello, all. Welcome to the future.