Skip to content

Using GPT / LLM / GenAI models

Note, the below text uses "GPT" but applies to all other LLMs and generative AI too.

We encourage our writers to use GPT and other models to help them learn new technology or to help them figure out technical steps, but we do not use LLM-generated text in our published articles. This is for a few reasons:

  • Generally, GPT's style is noticeable and noticeably different from our writers'. If you mix human and GPT writing, it creates a style clash. While this can be overcome sometimes with clever prompting, some parts of it will always remain.
  • For writing, verifying that something is correct is often as time-consuming as writing it yourself. By using GPT-generated text, you push more of the production burden onto editing and QA.
  • Often the stuff that GPT is good at writing is the stuff that doesn't need to be written. The value we provide is in creating stuff that doesn't already exist. GPT is best at rehashing existing knowledge. If GPT is providing very high-quality output for a specific paragraph, it probably means that the paragraph can be cut completely.
  • GPT can be very confident, which can make you doubt yourself. If X is true and you think X is true but GPT claims X is false, you might waste significant time sanity-checking yourself.
  • We want to provide value to readers. Most of our readers use GPT actively themselves, so they might find the article after GPT fails to help them. Therefore if we just give them GPT again we are not doing them a service.
  • Search engines likely detect and downrank GPT-generated text, so even a small amount can 'poison' a whole article.

This is what we find LLMs most useful for:

  • Creating code samples (that you then need to run, verify, and check for style and best practice).
  • Giving you ideas if you're stuck. If you're struggling with a sentence ask it to give you five different versions, and then manually write the best version of all by using pieces that look useful.
  • Teaching you concepts before you explain them to a reader. Often our writers are learning new concepts that they then need to explain. GPT can drastically reduce the time it takes for a writer to really "grok" something, and at that point, they are in a good position to explain it to someone else.
  • Formatting. If you need to reformat your text as markdown or add formatting, GPT can often do this very accurately with only a short description of what you need in the prompt.
  • Reviewing. If you ask GPT to review your text, it might give you helpful suggestions on how to improve it. (But again, don't ask it to rewrite your text and then use that. Ask it for what is wrong, and then fix it yourself.)

We encourage you to use the best LLMs. Ask on Slack for access to API keys or our hosted ChatGPT-like clone if you need.

Some advice on pairing with an LLM while writing

Everyone's process is different, so this might not work for you, but here's how one of our writers uses LLMs, and we recommend that you at least try this process for yourself as well.

Focus on your own voice and ideas first, then go to the LLM for help. Not the other way around. Anyone can use LLMs, so the valuable part you're adding is a bit of yourself, not augmenting what the LLM can do.

  • Start with research (LLMs can help summarise posts and resources if needed, transcribe videos, etc).
  • Then form your own opinion on one or multiple points (use LLM to sanity check if you want to, but remember they are sycophants to the core, and will often happily confirm your own opinions)
  • Write your own outline. This is important. LLMs tend to go too broad and don't have your opinion in mind. After writing your own outline, you can use an LLM to help expand it or add points you didn't think about, but keep to your original plan where possible.
  • Write the body. Only if you get stuck, use an LLM sparingly to bounce ideas off of, or to help you get unstuck. Don't use the LLM's words, or you'll "delve too deeply into the rich tapestry of adjective hell."

Channel Larry McEnnerney and write your introduction yourself putting as much value and personality (for blog posts) on display as you possibly can.

Also see our post on valuable writing.