Ai on Tom Guernier's Blog

Links and Fable

Thu, 16 Jul 2026 07:56:37 +0000

Like seemingly everyone else with a Claude subscription, since the start of this month I’ve been squeezing as much Fable as I can¹ out of my little USD$20 subscription. And most of that has gone towards my little side project / informal frontier AI coding test case: a top down railway company simulator game a la OpenTTD. So what have I got out of the big code machine so far?

Some Links on LLM (and Human) Writing

Thu, 26 Mar 2026 02:51:14 +0000

I’ve had a crazy busy last couple of days and my sleep schedule hasn’t quite recovered, so this will just be a link dump post. At least it’s all ‘on-theme’, albeit a rather over saturated theme these days. If you’re tired of hearing about AI, I don’t blame you!

There have been quite a few good posts on the topic of using LLMs for writing recently: here and here. Both of them land pretty close to my position, which is that there’s plenty of background tasks that LLM-based systems could be useful for (e.g. grammar checks, collating notes¹) but using it for the actual task of writing as in letting the LLM put words in your mouth is too far for me personally. To be clear, every word here is strictly typed with keys on a keyboard, and I have no plans of changing that.
Speaking of writing aids using LLMs, there’s a skill (a.k.a. a Markdown text file, I still don’t quite understand why that took so long to figure out) which acts as a writing coach, providing questions to try to tease out the writer’s ideas and improve the clarity of their writing (while never actually doing any of the editing for them, thus keeping with the above bullet point). I tried it on one of my old (already published) posts and… though it’s certainly not overly sycophantic (it grilled my post pretty hard) it seems like it wants to squeeze out all the uncertainties and idiosyncrasies for the sake of a clearer, more streamlined post. I couldn’t help but think it wants to make my post more… LLMish. Perhaps there are cases where its guidance would be helpful, for instance if I wanted to tailor my writing for a specific audience, but for now I’d rather keep my human-generated idiosyncrasies. Hopefully you find them endearing, or at least sufficiently non-LLMy to seem authentic.
And for some more general reading on similar themes to above, here’s Dynomight on prose vs. formatting.

Have thoughts? Post them in my feedback form!

Not note taking with Claude

Mon, 09 Mar 2026 04:17:41 +0000

TL;DR: Don’t use AI to produce notes, but using it for semantic search and knowledge synthesis seems fine. Don’t use it until you have a well ingrained note taking habit and at least 1 month of daily notes. Keep AI and human generated notes as separated as you can.

So a little while ago I finally relented and started paying for a personal Claude subscription, mostly to get access to Claude Code. As an Obsidian vault is just a directory filled with plain text Markdown files, Claude Code can seamlessly interact with the vault just by starting the CLI application in the vault directory. So am I freed from the burden of writing my own notes, now that I have a text generator that can do it all for me? Obviously not, but I think there’s still some uses for it.

Pre-existing forms

Tue, 27 Jan 2026 05:39:33 +0000

Turns out I kinda misunderstood how Beeminder’s rate modification works, so I’m still on the hook for one more weekly post before I can shift my posting rate down to once every two weeks. Still, here’s an off the cuff mini post to keep the system satisfied.

While browsing around for AI predictions a little while ago I stumbled upon this post on Rodney Brooks’ blog talking about a rather prescient poem from 1961 by the poet Adrienne Rich. The whole poem is worth reading, but the most striking part is its last stanza, managing to sum up my thoughts on AI creativity right now in 2026:

Links for January 2026

Tue, 20 Jan 2026 04:23:16 +0000

Announcement: I’m going to be posting once every 2 weeks from now on. I’ll still have Beeminder to make sure I stick to the posting schedule, though. I feel like I’ve burnt through most of the topics I can write about in one big writing session (hence me punting with a link post), and I would like a little more leeway to write stuff that requires actual research, or perhaps even multiple drafts?

End Of Year Thoughts

Sat, 27 Dec 2025 00:02:00 +0000

Last blog post for the year! Merry Christmas, happy New Year etc. Here are some reflective end of year thoughts, some jolly, some less so:

1.

It seems for me that this year, especially the latter half of it, has been the year of agency. On the one hand it has definitely allowed me to do more in the whole ‘You Can Just Do Things’ spirit, not necessarily crazy out there stuff which I would’ve never thought of myself doing, but more so things I could imagine myself doing but at some unspecified future point which never seemed to solidify into any real action. For instance: actually writing my thoughts down and starting a blog, reading actual physical books in Japanese and actually getting through my thousands of unread emails and getting a clean and spam-free inbox. Not to harp on about Beeminder any more than I already have on here, but it helps a lot and neatly avoids having to rely on the slippery and unreliable concept of ‘willpower’.

Some musings on AI creative writing

Thu, 02 Oct 2025 06:39:00 +0000

Every so often someone (e.g. Scott Alexander here) posts an ‘AI Art Turing Test’ where people guess whether a picture was made by a human or by a Stable Diffusion-esque generative model. A little while ago fantasy author Mark Lawrence did the same¹ for creative writing with GPT-5. The results seem to show that on average people get identify about half the pieces correctly, a.k.a they don’t do any better than random chance².