Battle of the Chatbots: ChatGPT vs Bard

If you’re reading this you are probably already familiar with OpenAI’s ChatGPT and Google’s Bard, but for those unfamiliar, here is a preview. ChatGPT, accelerated by Microsoft’s hype engine and investments has taken over the internet and has been touted as the Google killer. Google responded with Bard.

What is ChatGPT?

ChatGPT is a Large Language Model that uses the magic of machine learning probabilities and weights to generate text using A LARGE amount of data hence the name Large Language Model. There are many of these large language models, but what makes this one unique is that it has another layer in between that makes it respond to conversations and somehow, magically, it understands what you ask of it and is capable of complex interactions. It’s chat model is called chatgpt-turbo and is currently available as an API. And the latest version ChatGPT-4 is capable of multi-modal input processing. Meaning it can understand images and convert that into text. As in you can draw a website on a piece of paper, input the image, and it can both explain the website AND generate a functioning website. It won’t look great, but will function. I have yet to try this out myself, but the applications are limitless. A lot of easy to see applications include accessibility. Like what’s in the refrigerator? Or in the closet? Have it be a nutritionist or stylist? Will it be able to be a structural engineer (take a picture of a bridge and it’ll tell you its weaknesses)?

GPT-4 also recently released plugin integration functionality so it can now take action and interact with the world. The astonishing part is, GPT-4 can write the integration FOR you. With great power comes great responsibility. DO NOT let ChatGPT initiate conversations with itself and connect to the internet. I hope they built in a fail-safe to prevent this.

Read about how all of it works here: https://openai.com/blog/chatgpt

Here is ChatGPT’s own response to what it is:

What is Bard?

Bard is google’s response to ChatGPT after Microsoft integrated it into Bing. It is a Large Language Model called LaMDa but it appears to have been trained a bit differently than gpt-turbo. It integrates google’s massive search index into it’s models. I also know that it’s contextual memory is a bit more restricted. ChatGPT-3 is limited to 4k tokens. ChatGPT-4 has an 8k and 32k range and Bard is limited at 2k tokens.

Here is Bard’s opinion on the differences between chatgpt and bard.

Based on these results it seems like Google’s AI is geared more towards dialog and speech and less towards text-generation. The limited context and knowledge of referential data sometimes becomes apparent, but the interface is much nicer and cleaner. Note that I was not able to recreate some of the issues that others have been pointing out that Bard has with contextual knowledge from conversation to conversation.

Comparison

They both can do basic calculus (derivative of 2x):

They both can write haikus (about large language models):

They can both create new forms of text (I asked them to create a new language that takes eminem’s style of rapping and integrates it into prose and write me a paragraph on string theory with this new language):

They both work fairly well, ChatGPT appears to be slightly ahead of the game and has made their API very accessible. I’m sure there are some more nuanced language specific observations that experts can identify. Chat GPT’s personality can be tuned rather easily compared to Bard. They seemingly know everything about everything ranging from medicine, and law, to very niche topics like quantum field theory.

Both Google and Microsoft are working on integrations with their enterprise suites and Microsoft has already gone live with some of these things. Want a PowerPoint? Chat will do it for you. Want a spreadsheet? Chat will do it for you. Want a template for a document? Chat will do it for you. Want to code? Chat will do it for you? See my previous post about Copilot.

I don’t have access to these enterprise products so I can’t compare them, but it will revolutionize how people work and dare I say, not work? We aren’t there yet, but getting close. Both can make things up just like humans sometimes do. That’s what makes both them and humans creative. Except these bots have instant access to everything created by humanity. Imagine these bots learning how to interact with the physical world and connected to robotic limbs that are capable of sensory proprioception and touch. They could perform surgeries flawlessly. Add in self-improving feedback loops and you’ve got something that gets close to human.

Sidenote:

ChatGPT is doing for the text world that Tesla did for the vision world. GPT-4 appears to be a step towards that direction. Combine the tech from both of them and you’ve got something that has contextual knowledge of the world and can interact with it physically and through text/voice modalities. And that will lead to Artificial General Intelligence or AGI. Put this into a Tesla Optimus or Boston Dynamics body and you’ve got stuff that movies are made of.

Note: I am not affiliated with and do not represent ChatGPT, OpenAI, Microsoft, Bing, or Google, or Tesla, or Boston Dynamics. These opinions are mine and mine only.