Apparently, spreadsheets are all you need to understand how a LLM works

We all use LLM right now, whether is ChatGPT, Gemini, or the newest Grok-1 released as open source. But do you know how they work to provide you with an answer?

Ishan Anand, a software developer who also admits he’s passionate about spreadsheets, conducted an experiment to explain the mechanism behind LLMs that everyone can understand.

He embedded a small version of GPT-2 into an Excel spreadsheet and tracked its algorithm on how it answers a simple question.

How does ChatGPT 2.0 work in Excel?

First, you must understand that GPT -2 was released in 2019, and it has not been possible to converse. The first conversational AI bot came in 2022 with GPT-3.

While it was all over in the media headlines, GPT-2 could only use 1,5 billion parameters while GPT-3 extended that figure to 175 billion parameters.

Open AI didn’t release it as a conversational bot simply because it had limited input tokens. Ishan Anand’s GPT-2 in Excel can only handle 10 tokens of input which is a tiny amount for any possible query.

However, in his experiment, he tested the LLM to complete the sentence Mike is quick. He moves… If you watch the video above, you will see exactly how GPT-2 used the spreadsheet to come with the deduction that the missing word was quickly.

The point of the developer was to make us believe that if we understand Excel spreadsheets, we can also understand how a LLM works. If you want to experiment yourself, you can grab the spreadsheet from the GitHub page. You can also read more about it on Tom’s hardware.

Did you understand the explanation in the video? Let’s discuss in the comments section below.