| url | https://minimaxir.com/2020/07/gpt3-expectations/ |
|---|
A skeptical engineer’s review of GPT-3, written days after the API launched in 2020. Important now mainly as a time capsule: what AI looked like before the chat UI changed everything.
The cited claim
The GPT-3 demos on social media often hide the prompt, allowing for some mystique. However, because everyone has the same model and you can’t build your own GPT-3 model, the prompt is doing more work than people think.
Woolf was right. GPT-3’s “magic” was 80% prompt engineering and 20% model. The capability was already there in 2020 — the packaging was the bottleneck. ChatGPT’s contribution wasn’t a smarter model. It was an interface that made the prompt invisible.
Why it matters here
This piece anchors the “the capability was there for 2 years before PMF” claim in chatgpt-pmf. The lesson connects to atomic-concepts: GPT-3’s atomic unit was the prompt (engineering literacy required); ChatGPT’s was the conversation (English literacy required). Same model. Different market.
See product-market-fit, atomic-concepts, chatgpt-pmf.