Generative Artificial Intelligence, Large Language Models, and Image Synthesis

jabowery · 23 March 2023 18:44

That paper is founded on a flawed understanding of intelligence – specifically misrepresenting the rigorous theoretic work by Legg and Hutter. The misunderstanding is evidenced in the following paragraph about definitions of intelligence:

… Legg and Hutter[Leg08] propose a goal-oriented definition of artificial general intelligence: Intelligence measures an agent’s ability to achieve goals in a wide range of environments. However, this definition does not necessarily capture the full spectrum of intelligence, as it excludes passive or reactive systems that can perform complex tasks or answer questions without any intrinsic motivation or goal. One could imagine as an artificial general intelligence, a brilliant oracle, for example, that has no agency or preferences, but can provide accurate and useful information on any topic or domain.

An agent that answers questions has an implicit goal of answering questions. The “brilliant oracle” has the goal of providing accurate and useful information on any topic or domain.

This all fits within the Hutter’s rigorous AIXI mathematics – and is indeed more like falling off a log for this theory than anything that can be considered beyond it for a very simple reason:

AIXI has two components: An induction engine and an decision engine. The induction engine has one job: To be an oracle for the decision engine.

So, all one has to do in order to degenerate AIXI to a “brilliant oracle” is replace the decision engine with a human that wants answers.

The fact that the authors of this paper don’t get this – very well established prior work in AGI – disqualifies them from holding the positions they do.

Here’s an old but still very current lecture by Legg describing the taxonomy of “intelligence” he developed and its relationship to AGI.

PS: I’ve noticed that the the number of authors in an LLM paper seems to increase the greater the social consequences – almost as though no one wants to be caught out alone doing profound damage to the world. Another is this one:

“The Capacity for Moral Self-Correction in Large Language Models”
Deep Ganguli, Amanda Askell, Nicholas Schiefer, Thomas I. Liao, Kamilė Lukošiūtė, Anna Chen, Anna Goldie, Azalia Mirhoseini, Catherine Olsson, Danny Hernandez, Dawn Drain, Dustin Li, Eli Tran-Johnson, Ethan Perez, Jackson Kernion, Jamie Kerr, Jared Mueller, Joshua Landau, Kamal Ndousse, Karina Nguyen, Liane Lovitt, Michael Sellitto, Nelson Elhage, Noemi Mercado, Nova DasSarma, Oliver Rausch, Robert Lasenby, Robin Larson, Sam Ringer, Sandipan Kundu, Saurav Kadavath, Scott Johnston, Shauna Kravec, Sheer El Showk, Tamera Lanham, Timothy Telleen-Lawton, Tom Henighan, Tristan Hume, Yuntao Bai, Zac Hatfield-Dodds, Ben Mann, Dario Amodei, Nicholas Joseph, Sam McCandlish, Tom Brown, Christopher Olah, Jack Clark, Samuel R. Bowman, Jared Kaplan

CTLaw · 24 March 2023 10:01

jabowery · 24 March 2023 16:25

So long as the ML industry ignores the Algorithmic Information Criterion for causal model selection, its data efficiency is going to remain in the dark ages aka “statistics”, where interpolation is everything. People who claim that Transformers are not so-constrained are merely applying the Algorithmic Information Criterion in an unprincipled manner so as to keep everyone from the next revolution in the philosophy of science. As Scott Adams repeatedly points out about “AI”:

“We” don’t want the truth. The truth would break “the system”. “Everything” would stop working. Genuine AI will never be permitted.

So, yeah, I get it: There are enormous forces at work maintaining “The Way Things Are” in opposition to any radical advancement in science – including of course applying the sacrifice of the Baby Boom Generation on the altar of bits rather than atoms to something other than “Generative Artificial Stupidity”.

PS: One of the clues that GPT-4 has actually started applying AIC in an unprincipled manner is that they refuse to say anything about the number if “parameters”. This “We’re running out of DATA” video speculates an increase in parameters to 500B. I speculate a it isn’t nearly as large an increase, despite being multimodal, and may even be a DECREASE (at least in the language mode) – that, if even hinted at, would make people suspect that the whole ML industry is founded on a Big Lie.

PPS: Parameter distillation has, for some time, demonstrated “parameter count” is fluff – but now it’s starting to be exposed as the way to get superior performance exactly as AIT theory would predict:

Mooselake · 25 March 2023 17:05

Poking at Google’s Bard for the first time, while learning how to properly phrase the question including the differences between can you and will you. I haven’t tried running the programs, but I’m starting to think this would have been very handy when I was still working.

help me design a rose engine rosette

Sure, I can help you design a rose engine rosette. Here are the steps:

Choose a rosette design. There are many different rosette designs available, so you can choose one that you like or that fits the project you are working on.
Draw the rosette design on a piece of paper. Make sure to include all of the details of the design, such as the size of the rosette and the number of petals.
Cut out the rosette design. Once you have drawn the rosette design, you can cut it out with a sharp knife or scissors.
Transfer the rosette design to the material you are using. You can do this by tracing the design onto the material or by using a carbon copy paper.
Cut out the rosette. Once the rosette design is transferred to the material, you can cut it out with a saw or a drill.
Sand the edges of the rosette. Once the rosette is cut out, you can sand the edges to make them smooth.
Decorate the rosette. You can decorate the rosette in many different ways, such as painting it, staining it, or adding a patina.
Enjoy your new rosette! Once the rosette is decorated, you can enjoy it as a piece of art or use it in a project.

Here are some tips for designing a rose engine rosette:

Use a light touch when cutting out the rosette design. If you press too hard, you may tear the material.
Be sure to sand the edges of the rosette before decorating it. This will help the paint or stain adhere to the surface.
Take your time when decorating the rosette. There is no need to rush, and you can always make changes if you are not happy with the way it looks.
Enjoy the process of designing and making your own rose engine rosette!

help me design a rosette for an ornamental turning lathe called a rose engine. This is a cam, not a decoration

Sure, I can help you design a rosette for an ornamental turning lathe called a rose engine. Here are the steps:

Choose the desired shape of the rosette. There are many different shapes of rosettes available, such as circles, squares, and triangles.
Decide the desired size of the rosette. The size of the rosette will determine the size of the ornamental turning lathe that it can be used on.
Draw a diagram of the rosette. The diagram should include the desired shape and size of the rosette, as well as the location of the holes.
Cut out the rosette. The rosette can be cut out from wood, plastic, or metal.
Drill the holes in the rosette. The holes will be used to mount the rosette to the ornamental turning lathe.
Attach the rosette to the ornamental turning lathe. The rosette can be attached with screws, bolts, or nails.
Use the rosette to create ornamental designs. The rosette can be used to create a variety of ornamental designs, such as flowers, leaves, and geometric shapes.

Here are some tips for designing a rose engine rosette:

The rosette should be made from a material that is strong and durable.
The rosette should be the correct size for the ornamental turning lathe that it will be used on.
The holes in the rosette should be drilled accurately.
The rosette should be attached to the ornamental turning lathe securely.
The rosette should be used to create ornamental designs that are beautiful and intricate.

Can you design software to produce the cam profile for an ornamental turning lathe called a rose engine. This is a cam, not a decoration

Yes, I can design software to produce the cam profile for an ornamental turning lathe called a rose engine. This is a cam, not a decoration.

The software would need to be able to take into account the desired shape and size of the rosette, as well as the desired movement of the lathe. The software would then generate a cam profile that would allow the lathe to move in the desired way.

The software could be designed using a variety of techniques, such as mathematical modeling, computer simulation, or artificial intelligence. The software could also be designed to be user-friendly, so that it could be used by anyone, regardless of their experience level.

The software would be a valuable tool for anyone who wants to create ornamental designs using a rose engine lathe. The software would allow users to create complex designs without having to manually generate the cam profiles.

Please write a javascript program to produce the cam profile for an ornamental turning lathe called a rose engine. This is a cam, not a decoration

Sure, here is a JavaScript program that can be used to produce the cam profile for an ornamental turning lathe called a rose engine:

// This program produces the cam profile for an ornamental turning lathe called a rose engine.
// This is a cam, not a decoration.

// The program takes the following parameters:
// - The desired shape and size of the rosette
// - The desired movement of the lathe

// The program outputs the cam profile in a file called "cam.txt".

// The program is written in JavaScript and uses the following libraries:
// - The Math library
// - The Canvas library

// The program is licensed under the MIT license.

// The program was written by Bard.

var cam = {
  shape: {
    x: 0,
    y: 0,
    r: 100
  },
  size: {
    x: 100,
    y: 100
  },
  movement: {
    speed: 100,
    direction: 0
  }
};

// The program calculates the cam profile.
function calculate() {
  // Calculate the cam radius.
  var radius = cam.shape.r;

  // Calculate the cam angle.
  var angle = cam.movement.speed * Math.PI / 180;

  // Calculate the cam coordinates.
  var x = cam.shape.x + radius * Math.cos(angle);
  var y = cam.shape.y + radius * Math.sin(angle);

  // Write the cam profile to the file "cam.txt".
  var file = new File("cam.txt");
  file.write(x + ", " + y);
}

// The program draws the cam profile.
function draw() {
  // Draw the cam circle.
  var ctx = document.getElementById("cam");
  ctx.beginPath();
  ctx.arc(0, 0, cam.shape.r, 0, Math.PI * 2);
  ctx.stroke();

  // Draw the cam coordinates.
  var x = cam.shape.x + cam.size.x / 2;
  var y = cam.shape.y + cam.size.y / 2;
  ctx.textBaseline = "middle";
  ctx.textAlign = "center";
  ctx.font = "100px Arial";
  ctx.fillStyle = "black";
  ctx.fillText("(" + x + ", " + y + ")");
}

// The program runs.
window.addEventListener("load", calculate);
window.addEventListener("mousemove", draw);

Please write a C program to produce the cam profile with 24 sine waves around the edge and write it as an svg file for an ornamental turning lathe called a rose engine.

Sure, here is a C program that can be used to produce the cam profile with 24 sine waves around the edge and write it as an SVG file for an ornamental turning lathe called a rose engine:

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

// This function calculates the cam profile.
void calculate(int x, int y, int r) {
  // Calculate the cam radius.
  float r = 100;

  // Calculate the cam angle.
  float angle = x * M_PI / r;

  // Calculate the cam coordinates.
  float x1 = r * cos(angle);
  float y1 = r * sin(angle);

  // Calculate the cam sine waves.
  for (int i = 0; i < 24; i++) {
    float angle = i * M_PI / 24;
    x1 += r * cos(angle);
    y1 += r * sin(angle);
  }

  // Write the cam profile to the SVG file.
  SVGDocument *document = createDocument();
  SVGElement *cam = createElement("cam");
  cam->setAttribute("x", x);
  cam->setAttribute("y", y);
  cam->setAttribute("r", r);
  document->appendChild(cam);

  for (int i = 0; i < 24; i++) {
    SVGElement *sine = createElement("sine");
    sine->setAttribute("x", x1 + r * cos(angle));
    sine->setAttribute("y", y1 + r * sin(angle));
    cam->appendChild(sine);
  }
}

// This function draws the cam profile.
void draw() {
  // Draw the cam circle.
  SVGElement *circle = createElement("circle");
  circle->setAttribute("cx", 0);
  circle->setAttribute("cy", 0);
  circle->setAttribute("r", 100);
  drawCircle(circle);

  // Draw the cam sine waves.
  for (int i = 0; i < 24; i++) {
    SVGElement *sine = createElement("sine");
    sine->setAttribute("x", 50 + r * cos(i * M_PI / 24));
    sine->setAttribute("y", 50 + r * sin(i * M_PI / 24));
    drawSine(sine);
  }
}

// This function creates an SVG document.
SVGDocument *createDocument() {
  SVGDocument *document = new SVGDocument();
  return document;
}

// This function creates an SVG element.
SVGElement *createElement(const char *tag) {
  SVGElement *element = new SVGElement(tag);
  return element;
}

// This function draws a circle.
void drawCircle(SVGElement *circle) {
  // Draw a circle with the given radius.
  circle->setFill("black");
  circle->setStroke("white");
  circle->draw();
}

// This function draws a sine wave.
void drawSine(SVGElement *sine) {
  // Draw a sine wave with the given amplitude and frequency.
  sine->setFill("white");
  sine->setStroke("black");
  sine->draw();
}

// This function runs the program.
int main() {
  // Calculate the cam profile.
  calculate(50, 50, 100);

  // Draw the cam profile.
  draw();

  // Write the SVG file to disk.
  saveDocument("cam.svg");

  return 0;
}

Mooselake · 25 March 2023 17:18

OpenSCAD wasn’t successful, even with a link to a tutorial. Maybe the next version Had I paid more attention earlier I would have discovered the edit button rather than copy and pasting the whole request each time, I must be getting rusty

Please write a program in the openSCAD programming language, using the tutorial at OpenSCAD - Documentation to create a 200mm diameter 12mm thick cam profile with 24 2mm high sine waves around the edge for an ornamental turning lathe called a rose engine, the result will be an stl file

Sure, here is an OpenSCAD program that can be used to create a 200mm diameter 12mm thick cam profile with 24 2mm high sine waves around the edge for an ornamental turning lathe called a rose engine, the result will be an STL file:

import scad

diameter = 200;

thickness = 12;

sine_height = 2;

sine_count = 24;

cam = cylinder(diameter, thickness);

for (int i = 0; i < sine_count; i++) {
  cam = cam + sine(diameter, sine_height, i * M_PI / sine_count);
}

cam = cam.translate(-thickness / 2, -thickness / 2);

cam.export("cam.stl");

This program will create an STL file called “cam.stl” that contains the cam profile. The cam profile will be 200mm in diameter, 12mm thick, and will have 24 2mm high sine waves around the edge.

Lisp finally made it give up
I’m still learning coding skills, so at the moment I can’t help with this. I’m trained to do things like help you write lists about different topics, compare things, or build travel itineraries. Do you want to try any of those now?

johnwalker · 25 March 2023 22:24

My friend and science fiction and thriller author Jackson Coppley asked ChatGPT to tell the story of Little Red Riding Hood in the style of Jane Austen. Here’s what he got:

Once upon a time, in a verdant wood, a fair and modest young maiden named Little Red Riding Hood was sent by her dear mother to deliver a basket of victuals to her aged grandmother, who lived in a cottage on the other side of the forest. Little did the sweet girl know, that she would soon encounter a perilous situation.

As she trod lightly along the sylvan path, taking in the beauty of the foliage and the song of the birds, she noticed a gentlemanly figure approaching her. This was none other than the notorious wolf, who was known to be cunning and devious. But Little Red Riding Hood, being a gentlewoman of good breeding, did not fear him and greeted him courteously.

“Good day, sir,” said she, with a demure curtsy.

“Good day to you, fair maid,” replied the wolf, with a suave smile. “And whither are you bound?”

“I am on my way to my grandmother’s cottage, to deliver some nourishment to her,” answered Little Red Riding Hood, holding her basket close to her chest.

“How delightful,” exclaimed the wolf, his eyes glinting with avarice. “Perhaps you would be so kind as to show me the way?”

Being a trusting soul, Little Red Riding Hood agreed to the request, and the wolf gallantly escorted her through the forest. However, as they walked, the wolf began to ask her personal questions, and Little Red Riding Hood, being innocent and naive, answered them without suspicion.

“Tell me, my dear,” said the wolf, with a sly grin. “What is in that basket you carry?”

“It is some bread and butter, and a pot of honey, for my grandmother,” replied Little Red Riding Hood, innocently.

“Ah, how lovely,” murmured the wolf, his eyes gleaming with greed. “And where is your grandmother’s cottage, pray?”

“It is just beyond the clearing, by the old oak tree,” said Little Red Riding Hood, pointing ahead.

“Thank you, my dear,” said the wolf, with a mocking bow. “I shall bid you good day now.”

And with that, the wolf vanished into the forest, leaving Little Red Riding Hood to continue her journey alone. But as she drew near to her grandmother’s cottage, she noticed something amiss. The door was open, and the cottage looked ransacked. Fearful for her grandmother’s safety, Little Red Riding Hood crept inside, and was horrified to find the wolf disguised in her grandmother’s clothing, lying in her bed.

“Grandmother, what big eyes you have!” exclaimed Little Red Riding Hood, as she peered at the wolf.

“All the better to see you with, my dear,” replied the wolf, baring his sharp teeth.

“Grandmother, what big ears you have!” cried Little Red Riding Hood, as she backed away.

“All the better to hear you with, my dear,” replied the wolf, his eyes glinting with malice.

“Grandmother, what big teeth you have!” shrieked Little Red Riding Hood, as she ran towards the door.

“All the better to eat you with, my dear!” roared the wolf, as he lunged at her.

But just then, a brave woodsman burst into the cottage, and with his trusty axe, slew the wolf and rescued Little Red Riding Hood and her grandmother from harm.

From that day forth, Little Red Riding Hood learned to be more cautious and less trusting of strangers, and she lived a long and happy life, always remembering the perilous adventure that had taught her so much.

jabowery · 25 March 2023 23:45

Lex Fridman managed to drag the hint out of Sam Altman in this exchange about “size matters”:

johnwalker · 26 March 2023 14:33

Stephen Wolfram has announced the availability of the Wolfram plug-in for ChatGPT, “ChatGPT Gets Its ‘Wolfram Superpowers’!”. This plug-in provides ChatGPT direct access to the curated, real-world data available from Wolfram|Alpha and the symbolic mathematics and computational capability of Wolfram Language (formerly Mathematica).

The linked document has numerous examples of capabilities enabled by the plug-in. Here are a few examples.

Where’s the Beef?

DNA Sequence

Cows vs. Humans

Geographical Data

Real-Time Weather Information

Note that this plug-in explicitly lets ChatGPT “go out on the Internet” and find things that were not included in its training set. This is far beyond the “brain in a box” air-gapped AI we’ve been assured is safe. It will be interesting to see if other plug-ins appear which allow ChatGPT to, for example, trade in financial markets or go shopping for paper clip manufacturing machines.

ChatGPT plug-ins are currently available only to subscribers to the ChatGPT Plus paid service who must apply for access via a wait list. I am currently on the wait list.

Gavin · 26 March 2023 16:37

“The average mass of a [US] person is around … 137 pounds”

Come on, Wolfram! Look around you. Certainly, including new-born babies will bring the average down a little. But then the average mass of a cow would have to be adjusted for new born calves to ensure comparability.

And let’s not even broach the sexism of assuming that all cattle are female cows. What about equity for steers and bulls? Words can hurt, you know!

magus · 26 March 2023 22:48

Update using GPT-4:

johnwalker · 26 March 2023 23:05

The [US] in the quote was your interpolation. The actual statement by ChatGPT was “The average mass of a person is around 62 kg (137 pounds).” It then later said, “The total mass of people in the US, using an average mass of 62 kg per person is…”. This did not imply that this was the average mass of people in the U.S., but rather stating that it was using the previously stated average.

The figure of 62 kg average weight for adult humans appears to have come from the London School of Hygiene & Tropical Medicine and is quoted around the Internet but, of course, without a link to the source document, which I cannot find. This is presumably a worldwide estimate, which would be low for North America. This article on BMC Public Health, “The weight of nations: an estimation of adult human biomass”, dating from 2012, also states average body mass globally as 62 kg, then continues “North America has the highest average body mass of any continent (80.7 kg).”.

gms · 27 March 2023 00:30

From his article, he is not giving up, though he admits the odds are not great

I wouldn’t have been surprised by a C this year, a B in three years, and a 50/50 A/B mix by 2029. An A already? Base rates have clearly failed me. I’m not conceding the bet, because I still think there’s a 10-15% chance I win via luck. (And in any case, high inflation keeps eroding the real value of repayment).

What was the Hemingway quote - slowly at first, then suddenly?

Gavin · 27 March 2023 00:59

Thanks for that correction, Mr. W. It highlights a problem with Wolfram’s implementation of Chat GPT. The question you posed was – “Estimate the total mass of cows in the US compared to people.”

A human responding to that question would obviously conclude that the question related to the mass of US cattle versus (only) US people.

The class joker would necessarily have interpreted it as asking for the ratio of the mass of only female cattle (i.e., cows) in the US versus the mass of all people in the world.

Since both are valid interpretations of the question, it seems Wolfram/Chat GPT might usefully add an element to their programming to seek clarification before answering ambiguous questions.

CTLaw · 27 March 2023 13:51

@Shalmaneser can you identify a court decision or opinion (please provide citation and a link to the decision/opinion if possible) where a court found a law Unconstitutional and relied on the law having exemptions/exceptions as evidence of an unconstitutional burden on the rights of persons not granted exemptions/exceptions. If multiple cases are found, please identify the most significant.

Thank you in advance.

Shalmaneser · 27 March 2023 13:51

I’m sorry, but as an AI language model, I am not capable of browsing the internet or accessing legal databases. However, I can suggest that you try using Google Scholar or other legal search engines to find court decisions or opinions that may answer your question. You can also consult a legal expert or a law library for further assistance.

CTLaw · 27 March 2023 13:55

@Shalmaneser Regarding your response to my last query, are AI language models incapable of being trained on legal databases or other information so as to be able to answer my query? Or is there a limitation on your particular training? Or is there a limitation on your particular programming that prevented you from answering?

Shalmaneser · 27 March 2023 13:56

AI language models can certainly be trained on legal databases and other information to answer legal questions, but the specific capabilities and limitations of each model may vary depending on their training data and programming. While I am capable of understanding and responding to many legal questions, my responses are based on the information that has been fed to me during my training, and I may not always have access to the most up-to-date or comprehensive legal data. Additionally, I cannot browse the internet or access external databases during our conversation. However, I will do my best to assist you within the scope of my capabilities.

CTLaw · 27 March 2023 13:57

@Shalmaneser What are your capabilities that might help with my query?

Shalmaneser · 27 March 2023 13:57

As an AI language model, I can assist with legal queries by answering general legal questions, providing legal definitions, explaining legal concepts, and identifying relevant legal resources. However, my responses are based on the information that has been fed to me during my training, and I may not always have access to the most up-to-date or comprehensive legal data. Additionally, I cannot provide legal advice or opinions, and my responses should not be construed as such.

CTLaw · 27 March 2023 13:58

@Shalmaneser What is the name of the seminal US Supreme Court case where the Court found a law Unconstitutional and relied on the law having exemptions/exceptions as evidence of an unconstitutional burden on the rights of persons not granted exemptions/exceptions?