ChatGPT will be more likely to deviate from your desired direction as you try to generate more words in one shot. So even with the ability to generate that much, the number of words you find useful may be much smaller.
The required amount of tokens is determined by both the inputs and the outputs, which likely includes behind-the-scenes inputs that we don’t see.
But the simple max-tokens calculation doesn’t consider that OpenAI is likely instructing ChatGPT to limit its output in addition to the token limit. Furthermore, any behind-the-scenes instructions will count towards the token count, too!
Altogether, this means you want to increase your input and decrease your expected output to generate the best text.
Style Limitations
Quality, as in writing style, is an important aspect to consider when deciding how to generate content from an AI. ChatGPT is designed to be “safe.” One way it does that is by producing consistently bland text. This “bland” writing style is great for answering questions in a way anyone can understand. But it’s not so great for writing a hit piece of content.
Remember that robotic-sounding text-to-speech voice? Millennials probably remember using it to say bad words in their middle school library. Well, that’s the voice equivalent of ChatGPT’s writing style.
Fortunately, not all GPT-3 models have the same writing style as ChatGPT. The trade-off is that they are harder to work with in other ways. While ChatGPT requires the least from its users, it also provides the least flexible output.
The more flexible the output, like writing that reflects your unique voice, the more “prompt engineering” is required. The more capable models also lack the easy-to-use conversational interface, but the flexibility means they can be integrated into your custom workflow. If you’re in advanced GPT-3 utilization,
I can help
.
How to generate a 2,000-word article with ChatGPT
After all that, if you still want to generate 2,000 words or more using ChatGPT, here’s how I would do it, step-by-step:
Start with an expectation of breaking the content into at least 4-5 sections that make up the final 2,000 words. ChatGPT is quite capable of writing 400-500 words at a time. Choosing even smaller sections, and carefully curating which make it into your draft, will make for a much better final result. To do this, begin with an outline. The outline can be created with or without ChatGPT.
It’s important to maintain this outline while you instruct ChatGPT to generate each section. This will give ChatGPT the necessary context of the entire article while writing each section without encountering the max-tokens limit.
I recommend using a numbered outline to make things easier. Then, for each section, you can specify which section is the current focus by number.
The ChatGPT interface provides a helpful “try again” button. You should use this to generate multiple outputs for each section. Then, use your intuition to select the best for your article.
Throughout the process, you should maintain a document for your draft. As you complete the sections, copy & paste the best outputs into your draft document.
In the end, your document should contain all the sections generated by ChatGPT to make up your final 2,000-word article.
