Building GPT3 side-project in 2021 — Jul 20, 2023

Some insights from building copy.ai clone with OpenAI APIs in 2021:

OpenAI:

• OpenAI is inherently random. Without understanding the underlying model inside-out, predicting its output is not easy. It's like trying to know a lion solely by its claw. Instead, beautify, expand, or provide a TLDR of what users already have.
• Writing requires effort: paraphrasing, maintaining standards, fluency, formality, simplicity, creativity, expansion, and conciseness all demand effort. Provide value that reduces this friction.
• Find an ICP, conduct user interviews, build, repeat. How? Do they have the right budget for the problem? What's the frequency (daily/monthly)? What's the severity of pain (ROI/savings)? For example, D2Cs have good business people, strong marketers, and poor writers. Refer The Mom Test.
• A very simple inbound tip is to have a brag-worthy feature, for example, “Truecaller — I'll show names for any phone number you provide.” Display it on the homepage with a live demo.
• Prediction: Sales, Marketing, GTM, will be OpenAI's key ICP, e.g., ideasby.ai, to generate content for Clubhouse, tweets, YT. Poor draft > no draft. Writing isn't painful, SEO-ready writing is.
• Vertical growth requires complete control over what you sell. Building with OpenAI's API will increase competition due to low barrier entry and kill capital because you can't fully control what you sell.

Building:

• "I'll quit once I raise" isn't working. Building part-time can't compete with market leaders. Building without funding is anti-capitalistic thinking and consumes time, money, and energy. If you're bootstrapping, you should have won the ovarian lottery, been a Harvard dropout, or been an industry expert who worked in tech for years.

Personal Notes:

• Freelancing and consulting don't match my identity. They lack the compound effect that Monthly Recurring Revenue has.