Fantastic article, James. Loved your point about using these large language models to take notes for you in meetings so you can still productively contribute. Can you go into more detail as to how you do this?
And treating the AI tool as another person you can "talk to" and bounce ideas off can be really useful, particularly for someone like me who processes information by "thinking out loud".
With respect to using these tools as a writing assistant, or a writer I disagree with you though. While the temptation is certainly large (most of us engineers are terrible writers, and would prefer to focus on technical problem-solving, rather than writing), I'd make the argument Paul Graham does - writing is thinking. Sitting down and taking the time to clearly and concisely summarise what you're trying to communicate is also part of the problem-solving and thinking process. I can't tell you how many times I've written a paragraph that has then allowed me to realise some other insight.
I think you have many great tips to ensure you don't get crappy writing output, things like specific prompts, clarifying your key points, drafting then refining and doing a critical review of the output. But at the end of the day our clients are paying for our advice, whether that be written (in memos or reports) or on drawings. I'd be very wary of getting a LLM to do this for me.
Anyway, they're just my thoughts - love the work you're doing here!
I completely agree with you, writing is definitely thinking. This philosophy was a big part of why I started the Flocode newsletter. And that article by Paul Graham is excellent - I know it well.
That being said, sometimes a helpful push down the slide is all I need to get going. I often suffer from procrastination of the highest order.
For meeting notes, I use a system built around Voicemeeter Banana and a locally installed version of OpenAI Whisper. It’s essentially an automated script that records audio, transcribes it, and summarizes the transcription with custom prompts to maintain professional detail while cutting the fluff. It’s been incredibly useful, and maybe I’ll share the full setup someday.
And yes, caution with LLMs is absolutely necessary in our field. I’ve noticed more content like emails, proposals, even engineering documents that has clearly been exported directly from ChatGPT. That's low effort output. However, I still believe you can improve your thinking and your writing by leveraging LLM's thoughtfully.
Thank you kindly for taking the time to read and especially to share your thoughts. I really appreciate it. 👍
Fantastic article, James. Loved your point about using these large language models to take notes for you in meetings so you can still productively contribute. Can you go into more detail as to how you do this?
And treating the AI tool as another person you can "talk to" and bounce ideas off can be really useful, particularly for someone like me who processes information by "thinking out loud".
With respect to using these tools as a writing assistant, or a writer I disagree with you though. While the temptation is certainly large (most of us engineers are terrible writers, and would prefer to focus on technical problem-solving, rather than writing), I'd make the argument Paul Graham does - writing is thinking. Sitting down and taking the time to clearly and concisely summarise what you're trying to communicate is also part of the problem-solving and thinking process. I can't tell you how many times I've written a paragraph that has then allowed me to realise some other insight.
Paul makes the argument better than I can: https://paulgraham.com/writes.html
I think you have many great tips to ensure you don't get crappy writing output, things like specific prompts, clarifying your key points, drafting then refining and doing a critical review of the output. But at the end of the day our clients are paying for our advice, whether that be written (in memos or reports) or on drawings. I'd be very wary of getting a LLM to do this for me.
Anyway, they're just my thoughts - love the work you're doing here!
Thanks for your thoughts Luke.
I completely agree with you, writing is definitely thinking. This philosophy was a big part of why I started the Flocode newsletter. And that article by Paul Graham is excellent - I know it well.
That being said, sometimes a helpful push down the slide is all I need to get going. I often suffer from procrastination of the highest order.
For meeting notes, I use a system built around Voicemeeter Banana and a locally installed version of OpenAI Whisper. It’s essentially an automated script that records audio, transcribes it, and summarizes the transcription with custom prompts to maintain professional detail while cutting the fluff. It’s been incredibly useful, and maybe I’ll share the full setup someday.
And yes, caution with LLMs is absolutely necessary in our field. I’ve noticed more content like emails, proposals, even engineering documents that has clearly been exported directly from ChatGPT. That's low effort output. However, I still believe you can improve your thinking and your writing by leveraging LLM's thoughtfully.
Thank you kindly for taking the time to read and especially to share your thoughts. I really appreciate it. 👍