#2: How To Make Products With GPT-3

Own the model, own the UX

OpenAI’s former head of product Fraser Kelton explains why owning the underlying AI model is important for building a good AI-based product. You can prototype with a 3rd-party model. The critical product decision has to do with whether a model passes the vibe-check. After that, you build your own model to perfect the user experience and get the product/market fit.

How ChatGPT-like products are made

Another post by Fraser Kelton describes the product technique behind ChatGPT, which is surprisingly human-heavy. It’s also surprising how few hours are needed to provide a training set for a ChatGPT-like model, something perfectly doable by a startup.

Making a game with ChatGPT (look, no hands, ma!)

A fun 10-minute video on making a simple, yet fully playable and kind of interesting game with only ChatGPT prompts.

How to make a chatbot out of a book with GPT-3

A twitter thread by CEO of every.to makes a chatbot from Marcus Aurelius’ Meditations. Very straightforward with minimal technical knowledge needed. This is definitely a big part of the AI future: making chatbots from static documents.

CTO of OpenAI on how they made GPT-3 and what’s to come

In a 40 minute chat Greg Brockamn doesn’t share a lot of new stuff, but it’s fascinating to see the way OpenAI thinks and operates, in particular, how they got to GPT. They realised that the models they were good only in areas that they trained them in. So they simply decided to train the model on EVERYTHING, by feeding GPT the entirety of the Internet.