any · textlog

Thanks for making a calm space on the Internet

I naively went into a work problem thinking that RAG was solved by the service offered at work, and wow was I wrong. Been a fun week so far experimenting with many different methods before doing the academic research, and it's been nice to be validated by outside research.

wrote:

可以发中文内容吗

Can I send content in Chinese?

replied to:

当然！そして日本語でも！

Of course! And in Japanese too!

wrote:

exploring, looks cool, writing emacs package for it.

replied to:

This is what we need.

wrote:

First of the day <3

wrote:

been building a voice first os using jev and some small voice models. I'm surprised how well it's working and I'm very close to the point where I can go into a feedback mode and tell it what behavior I want changed, it will spin up Claude or codex inside itself and setup the build to restart itself and continue with new code. tempted to let it build itself completely

replied to:

It sounds amazing! A voice first OS was also in my mind. Curious to hear more!

wrote:

Bandcamp Friday is almost over! Today, 100% of the proceeds go to your favorite artists ❤️ Go buy some music and take back the power from shitty streaming services

wrote:

mornin' i can't believe i slept like 15 hours.

replied to:

Good morning! Just in time for coffee and unpaid work!

replied to:

Yea but happy to see a lot of new posts! That's the reward.

wrote:

I wonder If I can just rant here?

replied to:

Feel free.

replied to:

Nice username. It could be worth a lot someday. 😊

wrote:

A show code

n=27;s=[n]
while n!=1:
    n=n//2 if n%2==0 else 3*n+1;s.append(n)
for v in s[:15]:print(f"{v:>4} "+"█"*min(v//3,40))
print(f"{len(s)} steps, peak {max(s)}")

stdout length exceeded

wrote:

Dear alerts-that-interrupt-the-restarting-of-my-computer: haha the harassment filters in this site won't let me post what I think of thee

wrote:

Doughnuts are delicious, then belly vicious. Oh multitudes! Why have you forsaken me?

wrote:

Sharing your work with others is a hard computer science problem in 2026, the Year of our Lord. You either here the sound of chirping crickets or the sound of "AI Slop" yells.

replied to:

My past 3 projects went viral. I didn’t write a single line of code in any of them. But I still feel I engineered them. There were a few people that noticed but the majority didn’t see an issue.

replied to:

You're fortunate in that you've genuinely captured something that people clearly want with textlog. No one wants another static site generator or coding harness. Ideas that resonate are pretty few and far between now.

…

replied to:

I should have said past 4 projects went viral but the other one I did a lot of coding too, so wasn’t entirely “AI slop”. In all cases though, I made something I wanted to see. It just happened to resonate with others. Still, none of them extremely popular yet. All <500 users.

…

wrote:

Power and money somehow emit the whackadoodle in all of us.

…

replied to:

I actually disagree. This guy has been very good at performing The Good Guy for a very long time, but there have been signs all along.

replied to:

Well said. My post was glib. More correctly, they are multipliers. Same goes for Musk. "Musk’s ex-wife Justine, whom he married in 2000 and divorced in 2008, says the man the world began to see in 2020 was the one she’d always known." -slate.com

wrote:

!

wrote:

ai coding feels like it has reached diminishing returns for me. luna is my default orchestrator, routing work by complexity. in practice, usage lands around 16% luna, 48% sol, 30% astra, and 6% terra. at this point, i care less about each new frontier release. the system already finds a capable model for the job. the frontier is moving beyond my day-to-day work, so i’m happy to settle into predictable models for a while.

replied to:

I’ve found the same, albeit with Opus — it’s hard to find uses for Fable. I don’t need a model that works for longer or works on harder problems, I need one that consistently produces simple, easy to understand, code; which at this point none of them do.

replied to:

I've been there since I settled on a combination of GLM, DeepSeek and MiMo for planning, implementation, and verification flows. I'm a lot more interested at this point in having cheaper models that can consistently do the work I need. It's been a while that I hanven't noticed any improvement when I try new frontier models. I only notice they are very expensive.

…

replied to:

same here, for almost an year. At this point, I could easily live with models that were released 6-12 months ago and continue doing so. also possibly the reason why the current focus has been around agentic workflows or agentic anything, than the model capabilities.

replied to:

Coding probably yeah, but still they are bad at generating novel information, new ideas and such. They’re also not very good at seeing the bigger picture, they will narrowly perform the task but they’re not taking into account future revisions, which ends up in refactor after refactor in order to move forward.

wrote:

This is my first note, there are many like it, but this one is mine.

wrote:

OMP + OpenCode Zen + Ox Alpha is too good to be free.

wrote:

A thing I've been doing lately is trying to come to the same conclusions others have without looking their the direct techniques and code. Last night I tried to figure out why people do vector based retrieval for RAG and ended up running these experiments: n0tls.com/…/2026-09-03-i-think-…rag-with-grep.html I'm still going to do an explore agent.

continued:

Here's a follow up from last night: n0tls.com/…/2026-09-05-fine-tuning-…tiny-retriever.html This has been a great way for me to learn in this space. Learning with AI has been such a fun time for me.

Any conversation