Every big tech company is coming out with their own Large Language Model (LLM), AI is becoming a commodity like oil:
OpenAI released its ChatGPT API at incredibly low prices. Reference the image below by a16z and see that many AI writing apps being released are a wrapper around the OpenAI API.
So why write when you have an army of assistants ready to write for you? I’ll leave that up to you.
Amazon has over 200 e-books with ChatGPT listed as a co-author as of last month.
Reid Hoffman also wrote a book with GPT4 in under 24 hours:
I started my daily writing habit this year with the advent of this blog. The first couple blogs had the help of AI to write it. I asked ChatGPT to write on a certain topic and I edited it for clarity. I decided to stop using AI to help me write because I didn’t know how to prompt it to write like me or ChatGPT wasn’t up-to-date enough with the latest information on the internet. I am glad that I stuck to writing daily because it’s helped me write better emails and texts. It’s also therapeutic. While I write this blog to provide value it’s helping me more in the long run. I appreciate you reading!
Anyway, back to the original topic.
Fine-Tuning for Brand Voice
You can fine-tune ChatGPT to write like you based on hundreds of examples of training data. You can train it to write:
Tweets
LinkedIn Posts
Blogs
Landing page copy
Website copy
YouTube video scripts
UGC video scripts
You can do all that right now via OpenAI’s API with what’s called fine-tuning.
You can think of fine-tuning like learning a new skill. Imagine you already know how to play soccer, but you want to learn how to play basketball. You don’t have to start from scratch and learn everything again.
You can use some of the skills you already have, such as running, jumping and kicking. But you also need to practice some new skills, such as dribbling, passing and shooting.
Fine-tuning is like practicing these new skills until you get better at them.
If you work in a company, would training AI to write copy in your voice be valuable?
Here are some AI tools that do that without you having to learn how to code:
Embedding for AI Chatbots
How about training AI on your first-party data and creating a chatbot to access that information?
You can do that now with embeddings.
You can think of embeddings like a map. Imagine you have a big map of the world with many countries and cities. But this map is too big and complicated to fit in your backpack or on your phone. So you make a smaller map that only shows the most important places and how far they are from each other.
This smaller map is like an embedding: it simplifies the original information but keeps some of the relationships.
Example chatbots
You can create your own AI-powered chatbot with these tools:
What does it all mean for growth marketing?
It’s important to realize that you can do this now!
Create first-party chatbots for knowledge management and ask it:
what’s the best creative concept last quarter?
what’s our best performing headline?
what’s a good trial to subscribe rate?
what’s a good install to trial rate?
what’s the average CPI in the US?
who’s our highest LTV demographic?
Fine-tuning for quick, tailored text solutions:
ingest all of your past winning scripts and generate 3 similar scripts for testing
ingest your competitor’s scripts from Facebook and TikTok Ad Library and generate 3 similar scripts for testing
It’s already happening in the legal industry with PwC. Growth hackers should already be on the ball with LLMs. Are you doing anything with LLMs?
Let me know if I am missing anything obvious in the comments or if you need help implementing.
feels like were just scratching the surface. what are the first few steps i'd need to take to fine-tune?
-create openai API account?
100% I’d gobble that up.