JP van Oosten


Jan 19, 2023

Everyone is still talking about ChatGPT! I can’t open Twitter or LinkedIn without directly seeing someone talking about it. “🚀 You’ve been using ChatGPT wrong, here are 10 prompts you’ll also never need. Number 3️⃣ will amaze you!” and stuff like that.

In a previous post, I talked about my reservations regarding ChatGPT, which are mainly due to the closed nature of the OpenAI models. But there’s good news: not all heroes wear capes! Andrej Karpathy (not sure if he wears a cape or not…) has been working on nanoGPT, an open source implementation of the technology behind GPT.

You can train nanoGPT on an expensive machine (either rented or if you happen to own a very expensive computer), or train it on a smaller dataset on your laptop. Of course, if you train your own model on a laptop, it will not be as powerful as the things OpenAI puts out, but it might do the job just fine for some use-cases. The biggest thing though, is that anyone can use this as a basis for their own GPT implementation, trained on their own dataset, with different languages, etc. It becomes reachable for small & medium sized businesses. So, what are you going to build?

If you want to learn how to build your own GPT-like model, Andrej Karpathy has even published a youtube series on neural networks. Happy hacking!

(Also posted on my LinkedIn feed)