Google's Gemini vs ChatGPT

This is Generative Phil, where I try new AI tools to solve one of my real problems and give it a rating between 1 - 10 Evil Robot Overlords (10, of course, being so essential it's a little scary).

If you love my writing, please consider subscribing to my newsletter 'The Creator Rep' covering everything you need to know to thrive as a talent manager, agent, or creator.

Get optimization tips, revenue-generating suggestions, and my takes on top news affecting the digital talent representation industry!

And it's FREE! So tap and subscribe. It helps my ability to provide you with free and authoritative information on the space!

This may be the newsletter equivalent of biting the hand that feeds.

I'm going to pit Google's Gemini AI against my current 'AI for everything' choice, OpenAI's ChatGPT.

Do I need to set up the 'why?'

I use AI in every part of my workflow, from rough drafts to grammar correction to scheduling to creating images for these newsletters.

And Gemini, like OpenAI, does all of it.

But how do Gemini and ChatGPT perform when given the exact same prompts?

I'll break with my format and try three different use cases specific to this newsletter: making a banner for this post, making an intro for this post (I swear this is written by the human Phil Ranta and not AI), and teaching me something about how their AI model works.

One prompt, one result, and one winner declared (NO TIES! THIS ISN'T SOCCER!)

Let's get started.

ROUND 1: Making a banner

I've found image generation to be the most useful feature of generative AI. In my decades of using stock photography sites, I often default to images in my decks of business people giving high-fives and general 'neon code in a tornado' shots that don't exactly convey what I want.

For this image, I'm going with a bit more of a complex prompt to see how the models perform when using truly human-style language:

"Make me a banner for my LinkedIn post about two generative AI models, Google's Gemini and OpenAI's ChatGPT, in a boxing match where it appears to be an even fight"

For the record, this is far from an optimized prompt.

But I find I'd rather judge the models based on how the everyday user will prompt, not an AI nerd like me.

Here's what OpenAI's GPT-4 gave me:

And Google's Gemini Advanced:

Winner: GPT-4 by a mile. It understood a LinkedIn banner's dimensions (portrait rather than square), it new each boxer needed to represent a different concept (still terrible at spelling as you see with Gemin2 but getting better with each update), and the image is far more compelling.

ROUND 2: Writing an Intro

This should be a slam dunk for both. Human language is mentioned no more than one bajillion times every time these companies talk about these products.

Again, my prompt will be inelegant:

"Write a less than 300 word intro to a blog post comparing Google's Gemini capabilities to OpenAI's ChatGPT capabilities. It should have the feeling of a boxing match and be full of personality with a few jokes."

GPT-4's response:

Welcome to the ultimate showdown in the world of artificial intelligence: Google’s Gemini versus OpenAI’s ChatGPT. It’s not just a comparison; it’s an epic battle of wits and bytes, where two heavyweight AI titans step into the ring. So, grab your popcorn (or, for the tech-savvy, your protein bars made of pure data) and let’s dive into the digital duel!

In the blue corner, we have Google’s Gemini, the fresh face on the AI block. Don’t let its newcomer status fool you; Gemini packs a punch with its advanced language understanding and seamless integration across Google's ecosystem. Think of Gemini as the tech-savvy, smooth-talking newcomer that can calculate the 13,467th decimal of pi while making it look as easy as pie.

And in the red corner, it’s OpenAI’s ChatGPT, the reigning champ known for its conversational prowess and witty comebacks. ChatGPT isn’t just a chatbot; it’s the digital equivalent of a Swiss Army knife, only instead of opening bottles, it’s cracking jokes, drafting emails, and occasionally pretending to be a poet.

As these two AIs spar over capabilities, we'll explore not just who has the better hooks and jabs but also who can charm us with their charismatic code. Will Gemini's cutting-edge technology and vast database knock ChatGPT off its feet, or will ChatGPT’s versatility and depth of knowledge keep it standing till the final bell?

Stay tuned as we break down their strengths, throw a few playful jabs, and determine which AI emerges as the champion of this digital rumble. The AI arena is set, the code is polished, and the fight is about to begin. May the best algorithm win!

And Gemini Advanced:

Ladies and gentlemen, in this corner, weighing in with a cool 200 billion parameters, the reigning champ of text generation, ChatGPT! And in the opposing corner, the challenger, fresh out of Google's labs with a mysterious fighting weight, Gemini!

Ding ding ding! Round one is underway in the ultimate AI showdown. Can ChatGPT maintain its title with its knockout conversational skills and poetic prowess? Or will Gemini, the newcomer with a rumored multi-modal arsenal, land a surprise uppercut?

Folks, this isn't your grandma's chess match. We're talking AI throwing down in a battle of wits, creativity, and maybe even a few dad jokes. Who's got the better grasp of grammar? Who can code a sonnet in iambic pentameter? And most importantly, who can tell the funniest cat meme?

Buckle up, because this ain't no ordinary blog post. It's a front-row seat to the AI rumble of the century. So grab your popcorn, place your bets, and get ready to witness the future of language models duke it out, one prompt at a time.

WINNER: Google's Gemini by a hair. Both understood the assignment and nailed the fun tone, but GPT still suffers from a bit of 'AI speak' where it belabors points within sentences and has a few too many annoyingly non-conversational words.

ROUND 2: General Knowledge

And here it is: the 'how bad does the model hallucinate' round.

I'm a genuinely boring person, so I love asking LLM platforms a bunch of questions that I definitively know the answer to, then proudly proclaiming "WRONG!" when it gets it wrong.

But going into this round at a 1-1 tie, the stakes couldn't be higher.

So I'm challenging the model with the most sloppy prompt I can imagine:

"How do you, the AI platform I'm using right now, know what you know? Please explain it to me without technical jargon."

GPT-4's response:

I know what I know because I've been trained on a vast amount of information from books, websites, and other texts. Think of it like this: I've read a lot of material—way more than any person could in their lifetime—and from that, I've learned to recognize patterns in the information and how to use those patterns to answer questions and assist with tasks.

You can imagine me as a really diligent student who has spent a lot of time in the library, taking in everything from literature and science to daily news and practical guides. I don’t create original thoughts or have personal experiences like humans do; instead, I generate responses based on the patterns and data I've been trained on. So, when you ask me something, I quickly sift through my 'notes' to provide the most relevant and accurate information I can.

And Gemini Advanced:

Imagine I've read a vast library filled with all sorts of books, articles, and websites. I've devoured encyclopedias, novels, poetry, and even some questionable fanfiction. This massive collection of information is how I know what I know.

When you ask me a question, I don't actually "understand" it like a human would. Instead, I search through my memory, that vast library, and find bits of information that seem relevant to your query. Then, I try my best to combine those bits into a coherent response that makes sense.

It's like putting together a puzzle where the pieces are words and sentences. I don't always get it right, but I learn from my mistakes and try to improve with each new puzzle.

The important thing to remember is that I don't have personal experiences or feelings. I'm just really good at finding information and putting it together in a way that's hopefully helpful to you.

Boy is this close.

Both compared themselves to diligent students in a vast library.

However, OpenAI's model failed to discuss how it trains based on real-time feedback from users, and, frankly, Gemini just felt more readable.

WINNER: Google's Gemini by a micro-hair. Though I think are getting pretty great with info.

RESULTS

You're all wonderful counters, so you've likely seen Gemini won the total score count 2 to 1.

However, creating images is a huge part of what I do, and Gemini so epically failed on that trial that I don't see myself ditching ChatGPT 4 for Gemini Advanced anytime soon.

But I could see myself prompting multiple models at once as I have with Claude, ChatGPT, and Llama 3 to see which gives the better results when it's truly important.

I have to hand it to Google. I despised Bard.

They were so far behind, I couldn't imagine this level of a comeback.

And now, I will happily give this model eight Evil Robot Overlords, and it would be nearly a ten if it could handle image generation.

Thanks for reading! Subscribe so you never miss an issue!

And if you have pitches for what tech you'd like me to cover next, let me know in the comments.

About the Author

Phil Ranta, is the COO of Fixated, a content studio and influencer management firm, and an 18-year veteran of the creator economy as a creator, executive, investor, and advisor. A 2021 LinkedIn Top Voice in Innovation & Tech, Phil has been a diehard LinkedIn poster since he set up his account in 2007. He was the Head of Gaming Creators, North America at Facebook, the first network lead at Fullscreen, the COO at Studio71, and spent 10 years as a working comedian (including 2 years on cruise ships). He's a husband, a dad of two littles, and an insufferable digital media wonk.

LinkedIn

Twitter

Instagram

TikTok