Microsoft's Copilot app is the easiest free way to access GPT-4, but is it any good?

Summary

• Microsoft Copilot is an AI chatbot app that offers access to OpenAI's advanced DALL-E 3 and GPT-4 models for free on iOS and Android.
• Copilot excels in providing sources for its answers, making it more reliable and trustworthy compared to other AI chatbots like ChatGPT.
• Despite its impressive features, Copilot's user base on mobile platforms is currently limited.

A lot has been said about Microsoft Copilot lately, as the company has pushed the artificial intelligence chatbot very hard in recent weeks. Aside from expanding Copilot offerings within Windows 11, Microsoft is also adding a physical Copilot key to its laptops. Whether you love or hate Microsoft's transition to the 'Windows as a service' model, the newly-released Copilot app for iOS and Android is worth a look. This free app offers access to OpenAI's DALL-E 3 and GPT-4 artificial intelligence models. To get GPT-4 from OpenAI's ChatGPT app, you'll need to pay a hefty $20 monthly for the ChatGPT Plus subscription.

So, the Copilot app might be uniquely positioned to steal away ChatGPT's user base. After all, it's a nearly official way to get GPT-4 for free on the best phones since Microsoft has close ties with OpenAI as the company's largest investor. But all this only matters if the Copilot app is any good. Is it? We tried it to find out.

The brains behind the Copilot app

It uses DALL-E 3 technology and OpenAI's GPT-4 model

ChatGPT uses the GPT-3.5 model for the app's free version, locking the GPT-4 model behind a paywall. While GPT-3 is still an advanced model, the newer GPT-4 model has been described by experts and OpenAI as a breakthrough. The Copilot app also uses the same DALL-E 3 image generation model as ChatGPT, so it's at least on par in that area. That means that Microsoft Copilot can do things that all the best AI chatbots can, like answering questions, creating images, and writing content.

One key area where OpenAI's ChatGPT has the advantage over Microsoft's Copilot is in voice control. While you can interact with ChatGPT with your voice, that functionality is nonexistent in the Copilot app. This could be a dealbreaker for some users or an insignificant factor for others. As a whole, Copilot and ChatGPT are extremely competitive, at least on paper.

What it's like to actually use the Copilot app

Should you stop using ChatGPT and move over to Microsoft Copilot instead?

However, Copilot only stands a chance at unseating ChatGPT as the premier mobile AI chatbot if it's actually helpful to use. There are differing opinions on this; some prefer ChatGPT to Copilot, even with the latter's GPT-4 functionality. Personally, I found the Copilot app much better to use due to its ability to provide sources for its answers. Copilot might be the best of both Google's Bard (and the company's Search Generative Experience) and OpenAI's ChatGPT. The former is superb at scanning the web and providing sources for its answers, while the latter is excellent at functioning as a content-generating LLM.

Copilot can do both. It will provide notations with the sources used to generate its answers, and tapping those sources brings users right to the original source on the web. Separately, it can generate content using the GPT-4 model. This made me trust the accuracy of Copilot far more than ChatGPT.

In one test, I gave Copilot a list of ingredients I had around for a meal and asked the chatbot to give me five recipes using those ingredients. Then, I told it to expand on one of the recipes in detail. After answering my prompts helpfully, Copilot provided multiple sources for its responses. I found the original recipe on a food website by tapping just one link in Copilot's response. This mimics a similar test XDA editor Adam Conway tried with Google Bard when he asked the LLM to plan his meals for a day. However, Copilot provided much more detailed instructions than Bard did.

Just to see how accurate Copilot was in answering simple facts — ones significant and old enough to be in its training model — I asked it when the first iPod was invented. Not only did Copilot get the correct date (Oct. 23, 2001), but it also provided a link to the Wikipedia page used to find the answer. It doesn't matter whether Copilot is mostly using GPT-4, or mostly using Bing integration, or a mix of both. The result is an experience better than using ChatGPT.

I asked the same question to ChatGPT, and it didn't go as well. ChatGPT got the answer correct, but it refused to provide any sources for the answer. So, users of ChatGPT either need to trust the GPT-3.5 model to get the answer right or do their own research separately. You should also check Copilot's accuracy, but doing that is so much easier, thanks to the hyperlinks to sources it provides.

AI fatigue might be a factor

With everything needing to use AI right now, are we just sick of it?

So, if I think that Copilot is great at what it does, why isn't everyone using it? Copilot has only a few million downloads in nearly three weeks after its release on iOS and Android. Compared to ChatGPT, that's nothing, as OpenAI's mobile app garnered over 100 million downloads in short order after it was released last year. If I had to guess, there are two potential reasons why Copilot isn't taking off. The first is that people are simply tired of hearing about "AI everything," as the tech is now spreading into nearly all fields and may be losing its novelty for casual users. It also could be that ChatGPT was first, and many users are too comfortable with that chatbot to leave for Copilot.

Either way, Copilot's lack of early user support on mobile platforms shouldn't discredit the app. It's a very good solution, and I'd prefer it over ChatGPT in many situations. Copilot being the only almost official way to access GPT-4 is a nice bonus, but it's far from the sole feature, making it competitive.