On Reading & Writing
Certain writing, as you’re reading, breaks the fourth wall. It takes you out of being immersed in the content and into the meta - much like this is doing right now. If you’re reviewing a résumé or a client presentation and spot a typo or an orphan word, the illusion cracks. If you’re someone who notices those things, you’re instantly pulled out of the experience. I see these as mistakes - they ruin the sort of perfection, the facade that’s been stood up. The white-glove delivery expected when you decided to look at this polish-promised piece.
Writing and reading through LLMs has a similar effect. There are certain tells that make me question the writing: groups of three, em dashes, and the “it’s not x, it’s y” structure. As I’m reading, if I come across an em dash, I’m taken out of the reading environment and put into a wary state: can I trust this content? I’m not sure yet how I feel about this. The more people publish content, emails, even IMs using LLMs, the more using AI is going to be okay - and maybe this is just how writing is now.
Does it benefit your content to look human-written, if it’s all still saying the same thing - maybe even more eloquently forming your thoughts for you? I’ve seen YouTube videos labeled “human-generated music” (not on my bingo card, but then again, many phenomena over the past five years have been unprecedented times). Clickbait, maybe, but it works.
Perhaps we need guardrails: only use AI for brainstorming, not the final edit. Or re-edit outputs so they have our fingerprints. I’m concerned about sameness - that we’ll all start sounding more homogeneous over time. I also wonder about those who haven’t even found their voice yet, who are already growing up in a digital fishbowl. Is it too pessimistic to think this will flatten their uniqueness?
There are silver linings. One positive could be bringing stylistic choices like the em dash back into favor. And if people are using ChatGPT or similar to create content, odds are they’re using it correctly. I’ve oscillated on whether to cut down on my own usage - I use them more than most - just so the reader feels this is a safe, AI-free space. But that would mean fighting my authentic voice, and more importantly, fighting a perfectly appropriate use of punctuation
On Speaking & Thinking
We’re turning into pop stars who lip-sync and auto-tune more than we sing live. That’s concerning, because conversation isn’t just something you prep for - it’s constant, it’s impromptu. If we delegate too much of our cognitive load, we may get things done faster, but we’re left exposed in unplanned situations. Many jobs require quick thinking and adjusting to a changing or unknown chain of topics. Consulting, for example, is all about responding in real time.
How do we keep our edge at work - that ability to think sharply, speak clearly, and operate without a script - while being encouraged to use and still benefit from AI tools? Here’s what I’ve started to explore and practices I’m implementing more regularly:
- Use AI to sharpen, not substitute. LLMs can help shape your thinking, but it shouldn’t be your thinking. I draft the slide, paragraph, or outline before turning to models and use the tool instead to reframe, stress test, or format. I fear not forming my own POV if I start with ChatGPT from the outset.
- Practice unscripted thinking. My job involves impromptu conversations, client meetings, leadership discussions - letting those muscles atrophy puts me in a dire spot. I plan to build “live response” drills into my week: answering a big question out loud in one minute, rephrasing an abstract ask into a tangible task. Talking through a deck without looking at the slides. Practicing without preparation to keep me on my toes.
- Don’t take LLM output as gospel. There was a point early on where I would let ChatGPT have the final word. I believed it came from an informed place, from having consumed good writing and valid research, as it practically knew the entire contents of the internet. Instead of accepting a first answer as the finished product, I’ve started to question and push back the more I use it - a helpful exercise.
- Do the hard thing. It’s tempting to skip the messy middle, the sausage-making. It’s easy to go as a first resort to LLMs and have it answer a question within seconds before I’ve formed my own POV. This ties to point one but is more about purposefully sitting in strategic thinking - it requires time, and we as a society only want things faster. Fight that urge where it makes sense. In a world where more of our work can be delegated, the real value comes from knowing when and what not to delegate.
- Journal. Write what comes to mind freely. Write uncensored. Write like no one’s reading. Mind-map different parts of your life to make connections across pillars. Sit in reflection. Let your thoughts be your thoughts, without outside input from Chat.
- Read. I’m more aware of what I’m putting in my mind, of tells to look out for. Of expanding my vocabulary and the input diversity of my diet by reading different types of content. Don’t let LLM outputs be the primary material you consume. There were writers and readers long before AI-generated content - good work, too - that’s what it’s all trained on to begin with.
I’ve used ChatGPT here as the primary example but it can be replaced with any of the models - Gemini, Claude, etc. If you have any tips or practices you implement to keep your edge, let me know!