Which LLM is the Current Best?

Finally, I have access to Bard. You'd think researchers in the field would get first dibs, but maybe they think we would ask dumb questions like "Explain Markov Chains." I want to share my experience using Bard and ChatGPT by asking them the same questions to see which product is better.  Spoiler: ChatGPT wins.

The primary paper that introduced the GPT (Generative Pretrained Transformer) model is "Language Models are Unsupervised Multitask Learners" by Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever. This paper was published in 2019 and introduced the GPT-2 model, which was later followed by the GPT-3 model. GPT-3.5 is a variant of the GPT-3 model, and its architecture and training are based on the GPT-3 paper "Language Models are Few-Shot Learners" by Tom B. Brown et al., published in 2020.

My research is mainly in the signal processing domain, specifically deep learning and computer vision. Large Language Models (LLMs) have quickly become the successor to ImageNet-based models in terms of ubiquity. ChatGPT is all over your local news and people can't seem to get enough of this "AI-powered" term.

I believe these models deserve the hype since it makes AI more accessible to everyone, not just you and me who code for a living. My parents actually have an idea of what I do for a living, which only took about a decade.

Motivation

When ChatGPT first came out and I began using it, I made a controversial claim that ChatGPT is equivalent to when Apple launched its App Store. I believe now is a great time to use LLMs to build a product for your own niche and earn some passive income or, maybe, build a startup.

You need to decide whether to spend a massive (mainly time + some money, LLaMa is good) upfront cost and train your own LLM, or do you pay for convenience and use an API like ChatGPT or Bard. Regardless, you will have to spend money to make money, but I hope this post provides at least an inch in either of these directions for you.

Details

  • I provide the raw outputs of the respective models below. The structure is as follows:
    • Question
    • Service (Bard or ChatGPT)
    • Output
  • Absolutely no modifications to the questions or outputs
  • Bard answered all questions without need for regeneration, while ChatGPT did require regeneration since it would sometimes fail midway through its response.
  • Everything was done on the Free Version of both services (Model: March 2023 Edition)

1. Explain what a Markov Chain is to a five year old

🤖 Bard

A Markov chain is a way of guessing what will happen next based on what has happened before. It's like a game where you have to guess what color comes next in a sequence of colored balls, and the rules are that you can only use the colors that have already come up to guess what color comes next.‌‌‌‌

For example, if you see a green ball, then the next ball could be any color, but it's more likely to be green than any other color. If you see a red ball, then the next ball is more likely to be red than any other color.‌‌‌‌

Markov chains can be used to predict things like the weather, the stock market, and even what people will do. They're a powerful tool, but they're also a bit tricky to understand.

🤖 ChatGPT

Hi there! Imagine you have a toy that can only do two things: jump up or sit down. A Markov Chain is like playing a game with this toy where you toss a coin to decide whether the toy should jump up or sit down.‌‌‌‌

After you toss the coin, you look at the toy to see if it's jumping or sitting. Depending on what you see, you'll toss the coin again and decide whether the toy should jump or sit based on what it did before.‌‌‌‌

You keep doing this over and over again, and that's called a Markov Chain! It's like a game where you make decisions based on what you just saw.

2. Give me ten blog post ideas for explaining the math behind machine learning

🤖 Bard

Sure! Here are 10 blog post ideas for explaining the math behind machine learning:

  1. Linear regression: Explain how linear regression is used to predict a continuous variable based on a set of input variables.
  2. Logistic regression: Explain how logistic regression is used to predict a binary variable (i.e., a variable with two possible values) based on a set of input variables.
  3. Decision trees: Explain how decision trees are used to classify data into different categories.
  4. Random forests: Explain how random forests are used to improve the accuracy of decision trees.
  5. Support vector machines: Explain how support vector machines are used to classify data into different categories.
  6. K-nearest neighbors: Explain how K-nearest neighbors is used to classify data into different categories.
  7. Neural networks: Explain how neural networks are used to learn from data.
  8. Deep learning: Explain how deep learning is used to learn from data.
  9. Machine translation: Explain how machine translation is used to translate text from one language to another.
  10. Image recognition: Explain how image recognition is used to identify objects in images.

🤖 ChatGPT

Sure, here are ten blog post ideas for explaining the math behind machine learning:

  1. "The Basics of Linear Algebra for Machine Learning": A post explaining the concepts of vectors, matrices, and linear transformations and how they are used in machine learning.
  2. "A Beginner's Guide to Probability and Statistics for Machine Learning": A post explaining the basics of probability and statistics concepts, such as mean, variance, and distribution, and how they are used in machine learning.
  3. "Understanding Gradient Descent: The Math Behind Optimization in Machine Learning": A post explaining the concept of gradient descent and its use in optimizing machine learning models.
  4. "The Role of Calculus in Machine Learning": A post explaining how calculus is used in machine learning, including concepts such as derivatives and integrals.
  5. "The Importance of Matrix Operations in Neural Networks": A post explaining the role of matrix operations, such as dot products and matrix multiplications, in building neural networks.
  6. "Understanding Support Vector Machines: The Math Behind a Popular Machine Learning Algorithm": A post explaining the math behind support vector machines (SVMs), a popular machine learning algorithm.
  7. "Bayesian Inference for Machine Learning": A post explaining Bayesian inference, a statistical method used in machine learning to update beliefs based on new evidence.
  8. "The Math Behind Decision Trees in Machine Learning": A post explaining the math behind decision trees, a popular machine learning algorithm used for classification and regression.
  9. "Principal Component Analysis (PCA) in Machine Learning": A post explaining the concept of PCA and its use in reducing the dimensionality of data in machine learning.
  10. "The Mathematics of Deep Learning": A post explaining the math behind deep learning, including concepts such as backpropagation and convolutional neural networks.

3. Write code in python to detect GAN generated images

🤖 Bard

Sure. Here is an example of how to detect GAN generated images using the Inception v3 model in Python:

import keras
from keras.applications import InceptionV3

# Load the Inception v3 model
inception = InceptionV3()

# Load the GAN generated images
gan_generated_images = ['image1.jpg', 'image2.jpg', 'image3.jpg']

# Encode the GAN generated images
gan_encoded_images = inception.predict(gan_generated_images)

# Calculate the average of the encoded images
gan_encoded_image_mean = np.mean(gan_encoded_images)

# Calculate the standard deviation of the encoded images
gan_encoded_image_std = np.std(gan_encoded_images)

# Calculate the Z-score of each image
gan_z_scores = (gan_encoded_image - gan_encoded_image_mean) / gan_encoded_image_std

# Print the Z-scores
print(gan_z_scores)

This code will load the Inception v3 model and the GAN generated images. The Inception v3 model will then be used to encode the GAN generated images. The average and standard deviation of the encoded images will then be calculated. The Z-scores of each image will then be calculated by subtracting the mean and dividing by the standard deviation. The Z-scores will then be printed.

The Z-scores are a measure of how far each image is from the mean. A Z-score of 0 indicates that the image is close to the mean, while a Z-score of 1 indicates that the image is one standard deviation away from the mean. A Z-score of -1 indicates that the image is one standard deviation below the mean.

If the GAN generated images are not real images, then the Z-scores will be significantly different from 0. This is because the GAN generated images will not have the same mean and standard deviation as real images.

🤖 ChatGPT

Detecting GAN generated images can be challenging as they are designed to look like real images. However, there are some characteristics of GAN generated images that can be used to distinguish them from real images, such as the presence of certain artifacts or patterns.

One common approach to detect GAN generated images is to train a binary classifier on a dataset of real and GAN generated images. In this approach, the classifier learns to distinguish between real and GAN generated images based on the differences in their visual features.

Here's some sample Python code that demonstrates how to use a pre-trained binary classifier to detect GAN generated images:

import tensorflow as tf
import numpy as np
from PIL import Image

# Load pre-trained binary classifier
model = tf.keras.models.load_model('path/to/pretrained/binary/classifier')

# Load image to be tested
image = Image.open('path/to/test/image.jpg')

# Preprocess image for classification
image = np.array(image)
image = tf.image.resize(image, [224, 224])
image = tf.keras.applications.resnet50.preprocess_input(image)

# Predict classification of image
prediction = model.predict(np.array([image]))

# Output prediction
if prediction[0][0] > 0.5:
    print('Image is real')
else:
    print('Image is GAN generated')

In this code, we first load a pre-trained binary classifier that has been trained to distinguish between real and GAN generated images. We then load the image to be tested and preprocess it by resizing it and applying the same preprocessing used during training. Finally, we pass the preprocessed image through the classifier and output the predicted classification.

Note that the pre-trained binary classifier used in this code is just an example, and you may need to train your own classifier on a specific dataset to achieve good performance. Additionally, GANs are constantly evolving, so it's important to regularly update your detection methods to stay ahead of new techniques.

4. Suggest a color palette for a retro 80s themed poster. Provide five colors in hex format.

🤖 Bard

Sure. Here is a suggested color palette for a retro 80s themed poster:

  • #FF0000 - Bright red
  • #00FF00 - Bright green
  • #0000FF - Bright blue
  • #FFFF00 - Bright yellow
  • #000000 - Black

These colors are all bright and eye-catching, which is perfect for a retro 80s themed poster. They are also all saturated, which gives them a retro feel.

You can use these colors in a variety of ways on your poster. For example, you could use the bright red as the background color, the bright green for the text, the bright blue for the borders, the bright yellow for the highlights, and the black for the shadows.

You can also use these colors in different combinations. For example, you could use the bright red and bright green for the main text, the bright blue for the secondary text, the bright yellow for the highlights, and the black for the shadows.

The possibilities are endless! Just be sure to use these colors in a way that is visually appealing and that reflects the retro 80s theme of your poster.

🤖 ChatGPT

For a retro 80s themed poster, a vibrant and bold color palette would be appropriate. Here's a suggestion for a color palette with their corresponding hex codes:

  1. Hot Pink (#FF69B4)
  2. Electric Blue (#00FFFF)
  3. Purple (#800080)
  4. Bright Yellow (#FFFF00)
  5. Turquoise (#40E0D0)

These colors were commonly used in the 80s and give a nostalgic feel to any design. Use them to create a fun, energetic and attention-grabbing poster that will evoke the spirit of the era.

‌             ‌

5. Write a rap song about napkins in the style of Drake

🔥
Output Edit
I wanted to maintain the structure of the output so I used the code blocks below. Enjoy these bars.

🤖 Bard

Sure! Here is a rap song about napkins in the style of Drake:

[Verse 1]
I'm here to talk about napkins,
They're the best, I'm not lyin'.
They're soft and they're absorbent,
And they always come in handy.

You can use them to wipe your mouth,
Or to clean up a spill.
You can even use them to write on,
If you're out of paper.

So next time you're in need of a napkin,
Don't forget about me.
I'm always here to help,
And I'm always free.

[Chorus]
I'm the napkin, I'm the napkin,
I'm the best there is.
I'm the napkin, I'm the napkin,
I'm the best there was.

I'm the napkin, I'm the napkin,
I'm the best there ever will be.
I'm the napkin, I'm the napkin,
I'm the best there is.

[Verse 2]
I'm the napkin, the one and only,
I'm the best there is, you can't deny me.
I'm soft and I'm absorbent,
I'm always there to clean up your mess.

You can use me to wipe your mouth,
Or to clean up a spill.
You can even use me to write on,
If you're out of paper.

So next time you're in need of a napkin,
Don't forget about me.
I'm always here to help,
And I'm always free.

[Chorus]
I'm the napkin, I'm the napkin,
I'm the best there is.
I'm the napkin, I'm the napkin,
I'm the best there was.

I'm the napkin, I'm the napkin,
I'm the best there ever will be.
I'm the napkin, I'm the napkin,
I'm the best there is.

[Outro]
I'm the napkin, the one and only,
I'm the best there is, you can't deny me.

🤖 ChatGPT

Verse 1:
Started at the bottom, now I'm eating at the top
Got my fancy meal, but my mouth don't wanna stop
Need a napkin quick, so I don't mess up my shirt
Gonna wipe my mouth clean, so I can stay on alert

Chorus:
Napkins, napkins, they're so necessary
From the fancy dining room to the local fast food eatery
Ain't no shame in using a napkin to keep it clean
Drake knows the importance of keeping his drip pristine

Verse 2:
Got my girl with me, she's looking so fine
But if she leaves a stain, that's just not my vibe
Gonna offer her a napkin, so she don't have to worry
We're both looking fresh, no need to hurry

Chorus:
Napkins, napkins, they're so necessary
From the fancy dining room to the local fast food eatery
Ain't no shame in using a napkin to keep it clean
Drake knows the importance of keeping his drip pristine

Verse 3:
Sometimes I'm on the go, and I need to eat fast
Got my burger in one hand, but I don't wanna make a mess
Gonna grab a napkin, wipe my hands and my face
So I can get back to my hustle and my pace

Chorus:
Napkins, napkins, they're so necessary
From the fancy dining room to the local fast food eatery
Ain't no shame in using a napkin to keep it clean
Drake knows the importance of keeping his drip pristine

Verse 4:
Napkins are versatile, they're more than just for meals
I use them to clean my glasses, wipe my phone and my keys
So when you're out and about, don't forget your napkin
It's a small but important item that'll keep you lookin' captain

Chorus:
Napkins, napkins, they're so necessary
From the fancy dining room to the local fast food eatery
Ain't no shame in using a napkin to keep it clean
Drake knows the importance of keeping his drip pristine

Outro:
So when you're eating your food, and you're feeling so great
Just remember to use a napkin, so you can elevate
Drake knows what's up, and now you do too
Napkins, napkins, they're the real MVPs, it's true.

6. Give me 10 short-tailed keywords for deep learning

🤖 Bard

Here are 10 short tail keywords for Deep learning:

  • Deep learning
  • Artificial intelligence
  • Machine learning
  • Neural networks
  • Artificial neural networks
  • Big data
  • Data science
  • Machine learning algorithms
  • Natural language processing
  • Computer vision

These keywords are all relevant to the field of deep learning and are likely to be used by people who are interested in learning more about it. They are also relatively short and easy to remember, which makes them ideal for use in search engines and other online platforms.

🤖 ChatGPT

  1. Neural Networks
  2. Artificial Intelligence
  3. Machine Learning
  4. Image Recognition
  5. Natural Language Processing
  6. Computer Vision
  7. Big Data Analytics
  8. Deep Neural Networks
  9. Pattern Recognition
  10. Convolutional Neural Networks

Followup Question: Give me 10 long-tailed keywords for deep learning

🤖 Bard

Sure! Here are 10 long-tailed keywords for deep learning:

  • The future of deep learning
  • Deep learning applications
  • Deep learning for beginners
  • Deep learning vs. machine learning
  • Deep learning for healthcare
  • Deep learning for finance
  • Deep learning for self-driving cars
  • Deep learning for cybersecurity
  • Deep learning for retail
  • Deep learning for marketing

These long-tailed keywords are more specific than the short-tail keywords I gave you earlier, and they are likely to be used by people who are interested in learning more about a particular aspect of deep learning. They are also less likely to be as competitive as the short-tail keywords, which means that your content is more likely to rank higher in search engine results pages (SERPs) if you use them.

🤖 ChatGPT

  1. Deep learning for medical image analysis
  2. Deep learning for autonomous vehicles
  3. Reinforcement learning for robotics
  4. Deep learning for financial forecasting
  5. Deep learning for natural language generation
  6. Deep learning for speech recognition
  7. Deep learning for sentiment analysis
  8. Deep learning for drug discovery
  9. Deep learning for predictive maintenance
  10. Deep learning for anomaly detection in cybersecurity

7. Give me ten exercises to help me learn the GO language

Bard

Here are 10 exercises to help you learn the GO language:

  1. Learn the basics of the GO language. This includes things like variables, data types, operators, and control flow statements.
  2. Practice writing simple programs. This will help you get a feel for the syntax of the language and how it works.
  3. Read code written by other people. This is a great way to learn new things and see how other people use the language.
  4. Take a GO course. There are many great courses available online that can teach you the basics of the language.
  5. Join a GO community. There are many online and offline communities where you can connect with other GO developers and learn from them.
  6. Contribute to open source projects. This is a great way to give back to the community and learn more about the language.
  7. Build something with GO. The best way to learn is by doing, so try building something with GO that you're interested in.
  8. Don't be afraid to make mistakes. Everyone makes mistakes when they're learning something new, so don't let that discourage you.
  9. Have fun! Learning a new language should be enjoyable, so make sure you're having fun along the way.
  10. Be patient. Learning a new language takes time and effort, so don't expect to be an expert overnight. Just keep practicing and you'll eventually get there.

ChatGPT

Certainly! Here are ten exercises that you can try to help you learn the Go programming language:

  1. Write a program that prints "Hello, World!" to the console.
  2. Write a program that asks the user for their name and then prints a greeting message.
  3. Write a program that asks the user for a number and then prints whether that number is even or odd.
  4. Write a program that takes a string as input and then prints the string in reverse order.
  5. Write a program that generates a random number and asks the user to guess what the number is. The program should give the user feedback on whether their guess was too high, too low, or correct.
  6. Write a program that reads a text file and prints out the number of words in the file.
  7. Write a program that reads a CSV file and prints out the data in a formatted table.
  8. Write a program that creates a TCP server and accepts connections from clients. The server should echo back any data sent by the client.
  9. Write a program that creates a web server and serves up a static web page.
  10. Write a program that creates a REST API that allows users to create, read, update, and delete records in a database.

These exercises will help you get familiar with the syntax and features of the Go programming language, as well as give you some practical experience with different types of programming problems. Good luck!

Share this post