Post

It's also useful for turning a page of corporatespeak into three actual pieces of information.

I use Chat GPT all the time I even gave mine a nickname, he helps me with ideas for dinner recipes most days. There is only the odd time he might help me write an email it’s ingenious.

I also attend conferences (where thousands of people fly in for the event), watch YouTube and go on road trips - AI usage is a tiny slither of my carbon footprint, even taking into account my share of training costs

omg. sure. nuclear. nothing could go wrong.

I have to do a lot of non-public facing writing. Think reports, executive summaries, emails. It's faster to let ChatGPT handle the editing and proofreading. It helps a lot with "I didn't have time to write a short letter so I wrote a long one instead."

Coding is communicating though

No. I never use it for writing things for me that way either. But I know several people in multiple companies who seem to mainly use LLMs to write their reports or emails for them. Wonder if that's skewing perception.

I do some version of this too. Very heavy use of Whisper for voice input. But it’s always just initial conversations, research directions, and Socratic questioning. Once it comes to writing the final words, I find LLMs can’t help that much. Other than as an expensive thesaurus.

Something I've tried a few times is to dump a long draft blog post into NotebookLM and have the weird fake podcast hosts do an episode about it - I find it useful for checking that the key points I'm trying to make come across, and their interpretation sometimes sparks ideas for clearer explanations

Oh yes I use them for that a ton. But for the final words - the bit where you have to find the exact, concise phrases that say the thing you want to say, in the right narrative tone, with the right structure to them - they’re sadly not much help.

People really hate writing.

I use it to help me write a little code, but I absolutely do not find it useful for writing for me. If I'm sitting down writing a long document, I'm doing so because I will learn during the act of writing. If you can use AI for writing a long document, then it probably just shouldn't be a long doc.

Not outlier. I do the same

A significant portion of the value of writing is being the tool for thinking. I don’t understand wanting something written that was not worthy of being a vehicle for human thought. Aren’t we then, generating content that isn’t worth being read either?

Not at all! I love to write which is why I’m doing it mostly myself, BUT: I use LLMs for research, critique/feedback, checking tone of voice if I’m writing for an audience I don’t typically write for, brainstorming, asking different perspectives, writing draft code for one-off things

I think using ChatGPT as a writing machine isn't necessarily the most popular use case but it's 100% the most high-profile and visible one. Nobody notices all the ways I use LLMs, they notice the slop

Haha ok, it’s not spectacular but it was decent color for the occasion

Do t use it at all for writing yet. Not sure I'd like that. I (maybe egotistically) like to think I have a unique voice (more humor, less formal) and don't want to give that up.

It's also really good at "what am I missing?" in analysis. Using it to "write" is the wrong use case in my view. Still haven't quite come up with a decent prompt to stop it congratulating itself the whole time.

I usually dump a bunch of my own words and thoughts into a document, then use an LLM (generally Claude) to reorder it into something that flows better, while admonishing it to not use style/language that isn't represented in my original content. It saves me a bunch of rounds of revisions.

If LLMs mean people who don't write at all start writing, even if it's mostly clumsy AI-assisted prose that feels a little bit "off", is that really such a bad thing? I'm not talking about spam here, I'm talking about people who have something to say but previously lacked the confidence to share

OpenAI says 1/3rd of all college students are using "Chat" as the kids call it now.

I use mine for rearranging text I already wrote. It's good for bouncing ideas around, create images and speed up projects. It helps a lot with coding (being my code experience Atari Basic and early HTML). And I love when it comes up with clever comebacks if I start pulling its (virtual) leg.

you know how people who abuse adderall or other cognitive performance-enhancing kinda drugs tend to be unshakably certain that everyone else does too? reminds me of that

I really wish they hadn't opened the piece with that guy - I nearly stopped reading two paragraphs in, but then kept going and found that the REST of it was genuinely really interesting

...the sources may not be as solid as they seem: bsky.app/profile/aedw...

Outlines are probably the most productive part for students to actually write themselves though :(

I use it (or rather gemini recently) all the time and I don't use it to write anything I expect others to read. Maybe I'd ask it to check something I've written for issues.

Me, too. I use it constantly and, if writing is involved, it’s for editing, brainstorming and thinking things through. The idea of having it just write for me never happens.

Breakout yes. Good? No

Surely you use the pop-up/keyboard shortcut. In my experience developing that muscle memory was what elevated my usage to “any time I have an interesting thought or question.” I think most people don’t even know there is a desktop app, let alone this feature

I'm GenX and a published fiction author and I can see what a useful tool AI is - and how it will be integrated into work and cause catastrophic job losses. The denialism is something to behold.

💯 for curiosity engine. (I like this name over llms) 😂

I mostly use it for exploring or optimizing code (less writing it from scratch), or for exploring learning paths (if I want to learn X, what should I study first). I’ve never used it for writing I’ve shared with others.

Curious if you get a sense for this from the spread of comments, I feel like 70% of my LLM use is for “thinking” and 30% generating code most non-tech people I’ve watched use them for answering questions and writing boring emails - they find my “thinking” usecase somewhat neurotic (which it may be)

I don't know about "most ChatGPT users", but businesses certainly are. When I freelanced for a marketing agency, they used AI as the primary "author" of marketing blog posts. The human content creators sometimes added extra material, but not usually more than 25% of the total length.

My biggest use of the LLM apps has been for market research and related stuff. For me, the deep research models are the real "killer app". This is coming from a person who developed chatbots during the pre-LLM era, so the concept wasn't completely alien to me, except for maybe code output.

One of my favorite things about LLM chatbots is working through language ideas. I can talk about & use examples from English, Polish, Japanese, German, and Ancient Greek and I don’t know where else I could find a conversation partner like that to work through my many silly ideas in polyglot fashion

I can recognize good writing but can't generate it easily. I can arrive at decent prose, but the phase from unstructured thoughts until first draft is 90% of the effort/time. I think I'm not alone. Me: write notes LLM: first draft Me+LLM: update until well-written, well-structured, and in my voice.

Yeah exactly! And I think using them in that way overlaps a lot with how I use them for more complex code; every response is not great, but it's easy to validate the correctness of the response so it's pretty helpful regardless

LLMs also regularly confuse verbosity for clarity or give sidebars too much importance when I ask them to suggest edits - for example, with this response from Claude when I asked it to look at my post about everyuuid.com

I also HATE writing published material as part of a job; I write as myself and that's not what you're supposed to do. I might use ChatGPT for a public job-related blog post specifically to undo my sense of authorship. I'm sure many people have that professional distance from most of their writing

Some of the popular wrappers when ChatGPT first came out seemed like they were mostly around writing. To write emails, blog posts, social media content, writing responses in dating apps, etc.

Recent survey of UK students found a fairly varied range of uses of genAI, more often things like explaining concepts, generating/organising ideas, summarising www.hepi.ac.uk/2025/02/26/h...

I much dislike the future of everyone's output being llm fluff. I followed a popular GitHub repo for a few months and read every issue to become more familiar with the codebase. One of the maintainers used an llm for every response. I hated it so much I stopped following the project.

Some folks maybe treating it like they would Grammarly 😀

Selection bias? Many of the people writing are writers.

This is a major part of most enterprise workflows I’ve encountered. It’s a bit better (especially with things like Instructor) than writing a billion different parsers for slightly differing inputs

(I kid, but statistically, you round to zero)

yeah i don't use it to produce writing like that... this is the only way i use it in my writing process:

Lets begin this is the reason why it became popular even among the normies back in 2022

Claude is much better for writing... I am exploring now some science fiction ideas and it is great to be able to describe the idea, get a draft, do modifications, ask for changes and iterate on the story with Claude. I have never written fiction and the blank page was too intimidating, not anymore.

I use it for my documentation tasks at work

A tool that destroys the world fundamentally by using it regardless of what you use it for is kind of evil. Just sayin'

Shakespeare is known to have remixed various works from other authors, and his derivatives were usually more successful. Other authors were fine by it. Others hated him. The same is happening today.

Feeding in larger chunks of code and using it to help analyze potential sources of bugs is another way to use it like that - I did that yesterday with some test failures and Gemini 2.5 Flash gist.github.com/simonw/d867a...

Yeah it's really interesting to think of code generation as a "translation" exercise - transforming a human-language description of a problem into a code solution to that problem

Now of course it's a terrible and unethical tool and if you waste another human being's time with its slop you should be dropkicked, but that *is* what it's designed and sold for

I'm a self-taught jerk with HUGE gaps in my self-education. Last week, I was explaining how I wanted my tiny implementation of the OBJ language to discover equational rules given a specified algebra. ieeexplore.ieee.org/abstract/doc... I don't trust it, but it does help me discover papers.

That being said, I haven't written about AI in a while, because the most interesting things I learn are classified as “NTK Confidential” and under very strict NDA, so I'd like to avoid mentioning them accidentally.