Can an Excel spreadsheet with GPT-2 replace Copilot Pro? No, but it shows us how AI works.

GPT-2 in an Excel spreadsheet
(Image credit: Spreadsheets are all you need on YouTube)

What you need to know

  • A software developer named Ishan Anand put the entire GPT-2 small model inside a 1.25GB Excel spreadsheet.
  • The model was placed inside the spreadsheet to illustrate how AI processes information.
  • Anand has a series of videos walking through how the spreadsheet works and explaining insights learned by placing GPT-2 inside Excel.

Microsoft wants to pack GPT technology into just about every service and app it owns these days, and other tech companies are eager to integrate GPT models into various services. But what happens when you put all of GPT-2 inside an Excel spreadsheet? Rather than a tool for performing tasks or coding, you end up with a teaching tool.

According to Ishan Anand, the software developer behind "spreadsheets are all you need," "if you can understand a spreadsheet, then you can understand AI!" Those aren't just empty marketing words either. Anand placed the entirety of GPT-2 inside an Excel spreadsheet to illustrate how AI functions. Specifically, GPT-2 small has been packed inside of Excel to teach how AI works.

As you would expect from a sheet that holds an entire GPT model, you may run into issues when navigating the spreadsheet. "Unfortunately, it is not unusual for Excel to lock up (but only on a Mac) while using this spreadsheet," said Anand. "It is highly recommended to use the manual calculation mode in Excel and the Windows version of Excel (either on a Windows directory or via Parallels on a Mac)."

Anand has a series of lessons that use the spreadsheet to illustrate how AI works. The videos break things down in a way that makes it much easier to stand how a model receives information, categorizes it, and then acts. You can download the spreadsheet used by Anand through GitHub.

The first lesson is a 10-minute video called Demystifying GPT with Excel. The video illustrates how GPT-2 processes information, providing insight into how models function and process information. The second lesson goes deeper into detail, including explaining the tokenization phase and the Byte Pair Encoding algorithm that's used in models like ChatGPT.

Anand has an extra lesson that goes even deeper to explain how he used Excel throughout his lessons.

GPT-2 is a precursor to more powerful GPT tech, such as GPT-4 Turbo that now powers the free version of Microsoft Copilot. GPT-2 was around before "chat" was added to a GPT model.

Grounding AI

A robot that looks like a Terminator looking over AI

AI doesn't need to be scary if you understand how it works and there are limits and regulations in place. (Image credit: Windows Central | Image Creator by Designer)

While there are genuine security concerns surrounding AI, a lot of fear about technology comes from a lack of information or misinformation. Talking with people about AI reminds me of when I spoke with a person afraid of using a car with an automatic transmission because "the car may just drive itself and get in a crash!" People hear tidbits about tech and then conflate things that sound similar with cursory knowledge but are actually quite different.

Videos like the ones Anand shared help illustrate how AI models work rather than focusing on what they can do. In addition to being interesting and educational, the videos show how far AI has to go before it starts looking like Skynet. Of course, if we ever do reach Skynet levels of AI, we should have an emergency brake in place, thanks to Microsoft President Brad Smith and regulations.

Sean Endicott
News Writer and apps editor

Sean Endicott is a tech journalist at Windows Central, specializing in Windows, Microsoft software, AI, and PCs. He's covered major launches, from Windows 10 and 11 to the rise of AI tools like ChatGPT. Sean's journey began with the Lumia 740, leading to strong ties with app developers. Outside writing, he coaches American football, utilizing Microsoft services to manage his team. He studied broadcast journalism at Nottingham Trent University and is active on X @SeanEndicott_ and Threads @sean_endicott_.