The teacher said: “Your grammar is appalling, but the story is lively and inviting.” I wonder if my grammar has improved a little bit? I have a feeling it has. But sadly, I also have a feeling that my storytelling ability has decreased. I wonder why. Perhaps nowadays we really are consuming too much content, and imagination is dying as a consequence. Or is there some small trick, more coffee? Or perhaps less. Writing early in the morning, or late in the evening.
In either case, one thing is clear: you cannot improve if you don’t try. Writing a lot, even if you keep getting disappointed by the end product, is the only path to getting better. And so I do. Really, mainly for my own soul.
The fact that I’m not a native English speaker makes grammar an additional challenge. I am also learning Russian. And exploring Latin, at least a little bit. And German, at least to be able to explain that I don’t speak German when I hop across the border. It’s all great, as I am on a mission: making my grammar suck in many languages.
LLMs
I hate AI-written content. It has a synthetic feeling. It takes the ‘you’ out of it. Reading AI-written posts is soul-killing. But I do utilise it sometimes. Maybe for a small, soulless email that I have to write. I am not happy with myself when I do it, but hey, I am not perfect, and my day has only 24 hours, like everyone else’s. Let’s optimise, right?
For posts like this… it would be pointless, though.
But for grammar, it can work.
So my usual prompt would be: “Spellcheck, grammar!” That is sufficient.
I’m not happy with the bot just auto-correcting and calling it done, though. For my Ukraine post, I took the following approach: paragraph by paragraph. Let’s talk about what could be done better. What’s actually the proper way to construct dialogue, place quotation marks and use capitalisation in English?
We went through the whole thing, and it took lots of time. It was somewhat productive. I learned. We even went so far as to analyse the recurring grammatical mistakes I was making.
The end result came out somewhat okay. But I felt it lost some of its uniqueness, the part of me I wanted to imprint on it. Even with just the grammatical restructuring we’d done, replacing a word here or there, fixing the flow, something was still lost. Or so I feel.
Due to all the messing and moving, reading and re-reading, the structure also got broken. I need to go back and repair at least that.
My feelings, hence, are mixed. Let me make concise statements.
Conclusion
Purely AI-written content is a no-go for me. I feel dirty. I feel like my brain cells are dying and I am not improving, and most importantly: it’s not me speaking. Again: it can work for a small, quick email here and there. That’s it. I wouldn’t subject my friends, family or co-workers to reading tens of paragraphs of AI-generated text.
Grammar and spelling: absolutely yes, preferably in a way where I actually re-type the corrected version, so that I learn. Maybe I’m a zealot, but I truly want to learn.
LLMs as an extensive learning partner: yes! With some caution. LLMs can be quite seductive. “Hey, this is much better.” And, “This flows better.” And, “This word would be more appropriate here.” In the end, the result is synthetic again. The bot does not understand that pure optimisation is not the end goal, and that the author’s voice is a better option than the word with the highest weight.
LLMs as a learning tool, then: with caution, but yes, that’s the best way for me.
They’re also good for fetching cover images, moving things around, writing one-paragraph post descriptions, and other boilerplate, non-creative tasks.
Finally, as a research partner. Basically an inline Google. My article about Arch to macOS has the sentence: “…thriving ecosystem of launchers, like Alfred and Raycast.” What I actually wrote was: “…thriving ecosystem of launchers, like Alfred, and (what’s the other one, very popular right now?).” Then I instructed the bot to take care of the statements in brackets. Really useful. Needless to say, double-check those sections, as LLMs do get things wrong.
To conclude, I’m using AI the way we all hoped it would be used: I want to do creative things, I want to learn, and the bot can take care of the boring parts.