Gemini vs ChatGPT: Clash of the AI Titans

When ChatGPT launched at the tail-end of November 2022, it quickly gained recognition for its exceptional capabilities, shaking the very foundations of Google’s dominance as the go-to source for information on the web. Google’s AI team was then compelled to fast-track the development of their own AI language model, which debuted as Google BARD in March 2023. Less than a year later, on February 8th, 2024, BARD was rebranded to Gemini.

Google’s rationale was that it wanted to shift away from the BARD “persona” to the Gemini “platform.”

What powers the two platforms?

The key difference between ChatGPT and Google Gemini stems from the Large Language Models (LLM) they employ. OpenAI’s ChatGPT uses the Generative Pre-trained Transformer 4 (GPT-4), while Google Gemini relies on its bespoke Language Model for Dialogue Applications (LaMDA). Both have had notable problems, but AI is continuously learning and improving.

At their core, the two models share the revolutionary transformer architecture, a neural network specifically designed for processing sequential input.

The GPT-3 and GPT-4 predecessors were trained on a vast array of text sourced from the internet. We’re talking about websites, books, articles, documents, etc. On the other side, Google LaMDA’s training was enriched by Infiniset, a dataset primarily focused on dialogues and conversations.

Infiniset has been feeding on Common Crawl, Wikipedia, published documents, and a rich assortment of web-based conversations. In addition, Google Gemini’s AI model has also been engineered to scour the web in real time, generating relevant and up-to-date answers to users’ questions.

ChatGPT too relies on training data compiled from Common Crawl, Wikipedia, books, articles, documents, and open internet content. However, the free version (GPT 3.5) can only access information up until June 2021. If you’re willing to upgrade to the paid version (GPT 4), then the training data cut off is pushed to January 2023 – plus it can search the web.

It’s worth noting that both AI models are still works in progress. As their capabilities advance, the dynamics of the Gemini vs ChatGPT rivalry will continue to evolve.

How each platform respond to prompts

In this section, we will compare the responses provided by both Gemini and ChatGPT on three different tasks to determine which AI tool performs better in terms of accuracy, aptitude, information synthesis, and coding.

1. Accuracy and aptitude
2. Information synthesis
3. Coding

1. Accuracy and aptitude

Test question: How much money do I need to run a WordPress ecommerce website until it breaks even?

Upon closely analyzing the responses provided by both ChatGPT 4 and Gemini, we can observe that they cover similar aspects when discussing the costs of running a WordPress ecommerce website. Both AI models mention domain and hosting, themes, plugins, ecommerce platforms, marketing, payment processing, inventory and shipping, and other potential expenses.

However, there are some subtle differences in the way each AI presents the information. ChatGPT 4 delves deeper into the topic, offering more specific details regarding the costs associated with each element.

For example, it includes WooCommerce in the assessment, and then goes on to provide an estimate of payment gateway charges (around 2.9% + 30 cents per transaction). ChatGPT 4 also emphasizes the importance of estimating fixed and variable costs alongside revenue forecasts to calculate the breakeven point.

On the other hand, Gemini’s response is more concise and focuses on a general overview of the costs involved. While it does cover all the major aspects, it does not provide specific details or figures related to payment processing or ecommerce platforms.

Overall, I would tip the hat in favor of GPT on this one, based on its slightly more comprehensive response.

🏆 Winner: ChatGPT

2. Information synthesis

Test task: Explain API to a non-coder.

On prompting each of the AI tools to explain API to a non-coder, both provided solid answers.

Interestingly enough, they also both relied on the same exact analogy of a waiter taking orders at a restaurant.

While GPT goes into more detail like it did in the previous question, in this case, that approach proves to be less effective at grasping the overall concept of how API works.

Gemini’s version offers a much more concise response with parentheses next to each element of the analogy. This makes it easy to follow along and understand the process. With GPT’s explanation, you might find yourself having to re-read it to fully get it.

The additional key points that Gemini provides gives readers further insight on the purpose of APIs and their practical applications across various industries.

🏆 Winner: Gemini

3. Coding

Test task: Generate HTML and CSS code for a chess gaming app.

On prompting them to code, Gemini’s answer offers a solution that’s more refined for visual presentation and accessibility on different devices. This is due to the included responsive full-page styling and use of flexbox – which the GPT version lacks. It also provides a clearer method for adding and styling chess pieces through distinct classes and more control over individual square styles.

Here’s the GPT HTML:

GPT goes on to provide the CSS too, and underneath that, it even gets started on the JavaScript that you would need to actually make the chessboard workable, along with additional tips for creating a fully functional game:

Although the GPT code is more minimalistic and would require additional enhancements to match the level of styling and layout control offered in Gemini’s code output, it does outshine Gemini on this one specific thing (the JavaScript snippet).

Here’s Gemini’s response:

After Gemini generates the CSS, it also gives similar tips to GPT about steps you’d need to take to get a fully functioning game up and running:

The nice thing about Gemini’s tips is that while both of them mention the need for game logic, only Gemini provides an actual link to a Github resource that could potentially be used for this purpose. Overall, even though GPT did provide that JavaScript snippet, the HTML and CSS weren’t as good as Gemini’s and neither were the additional tips. This round goes to Gemini.

🏆 Winner: Gemini

What are the differences between Gemini and ChatGPT?

Approach to answers 🙋‍♂️

LaMDA’s training emphasizes grasping the user’s question intent and contextual nuances. To accomplish this, Google researchers organized high-level concepts into hierarchical clusters, guiding the model’s response selections.

The AI tool aims to produce genuine responses that closely resemble human speech. It even tries to transition between subjects, much like a human conversation. From our limited tests, it didn’t always feel that way, but it also wasn’t a bad thing. For example, with the API analogy, the fact that Gemini placed each API process element in parentheses next to the analogy element wouldn’t be how a human would naturally talk, but it provided for a better answer to the question in that context.

Overall, Gemini’s text-based answers seemed more concise and succinct compared to GPT, which was good in one instance, but not as good in another. In the case of coding, Gemini went more in depth than GPT on the specific request, while both expanded on it in their own ways – GPT with the JavaScript snippet and Gemini with the external resources.

Training data 💽

Whereas much of Gemini’s training data remains relatively unknown, LaMDA’s research paper offers some critical insights ^[1]. It reveals that 12.5% of the training data comes from Common Crawl, akin to GPT-n models, and another 12.5% is sourced from Wikipedia.

It turns out, however, that ChatGPT boasts more model parameters than Gemini, with 175 billion compared to 137 billion. These parameters are basically adjustable elements that help the AI model adapt to the data during training. And while more parameters mean better synthesis of complex language relationships, they also risk overfitting.

Overall, on the question of ChatGPT vs Gemini training data, we could say that even though Gemini may exhibit greater resilience to novel language use cases, ChatGPT is trained to be more versatile.

Internet access 📡

Early models of ChatGPT lacked the ability to access the internet. However, nowadays, if you sign up for the paid plan, you can use the GPT 4 model, which provides internet access (among other things). The free version of ChatGPT still does not allow for real-time web connectivity.

On the other side of the aisle, Gemini has always been able to search the web – even when it was still called BARD. This feature has made it highly proficient in delivering current information with minimal errors.

What’s more, Google Gemini occasionally cites its URL sources for user reference.

It’s worth noting, though, that Gemini’s real-time web access doesn’t guarantee 100% accuracy. The AI has embarrassingly had issues with providing incorrect responses, even during its initial release.

User interface 🖱️

Both tools are quite straightforward, as they primarily interact with users through a search bar. There are, however, some minor differences in functionality.

Gemini supports voice prompts via your device’s microphone, offering a hands-free experience. It also lets you upload images for analysis. Once you get your results, you can click the volume button to have them be read to you.

Beyond that, there are also various ways to share your result and there’s even a “Google it” button for seamless in-line research. If you use it, Gemini will tap into Google to find source content for its response, and will highlight select statements to share those additional resources with you:

The AI interface additionally features a “drafts” option that presents multiple response variations per prompt, allowing users to select the most fitting reply. The user-friendly activity display here even shows past prompts, with options to view details or delete entries. Users can otherwise disable Gemini’s activity storage.

ChatGPT’s interface is honestly very similar. While it lacks a voice prompt feature, it has a one up on Gemini in the sense that if you ask it about something that’s current, where it’s forced to use the internet, it will automatically provide you with the sources it got its information from. You don’t need to tap any additional search engine button:

ChatGPT – at least the paid version – also lets you generate images, which you can’t do in Gemini:

Overall, the premium version of ChatGPT probably slightly edges out Gemini in the interface department, but the free version, with its lack of internet access, does not.

Which is better between Gemini and ChatGPT? 🏅

After comparing the performance of ChatGPT against Gemini, it’s hard to pick a clear winner. Compounding the difficulty is the fact that ChatGPT has a free version and a paid version.

In the end, if I was forced to only choose one, then GPT Plus it is, but if we were going by free versions only, then Gemini would edge out GPT. Keep in mind though that both models are still developing, and their capabilities will continue to change over time.

Their growth will certainly make the ChatGPT vs Gemini rivalry even more interesting, and we’ll be here to provide you with all the updates. Keep watching this space.

What are your thoughts on this showdown between the two AI chatbots? Do you prefer one over the other? Let us know in the comments below!