pieterk

Yup, I used this for a while and IME it may get you a few percentages more of useful context initially, so quality feels a bit higher, but things start breaking down in funnier ways when you do run out of that quality for any reason later, so definitely caveat emptor.

Yup, it's fantastically useful.

Maybe even more useful than Opus when I have all the constraints to an issue. There is less "knowledge" in the model (I get by with 48GB of RAM allocated to an 8b quant), so it has fewer things to hallucinate about.

I've been getting to know its limits pretty well over the last few weeks and would say it's an excellent code search/replacement/generation* engine.

It's got the "in-context script generation" flow down as well, so it will easily help automate tasks that you describe with text and perhaps example commands, or tools, or skills* that you provide.

*Think of it + Pi as an NLP abstraction layer over grep, or a shell, rather than a jack of all trades + world knowledge all-in-one.

FTA:

...pretentious as ZAMM may be, I really can't stop thinking about it now that we have to contend with the Maw.

What is the Maw? The Maw is the gaping pit of nihilism that has opened up in the middle of the tech industry. The Maw is the explicit or implicit subject of roughly 63% of blog posts now shared on link aggregators like Hacker News, the subject that nobody can resist writing about, even authors who typically write about SAT solvers or microservices. The Maw is the looming threat that has prompted such an outpouring of cris de couer in the blogosphere, which some might view as—though I hope events don't go this way—just the death braying of a highly literate professional class.

...

Quality is related to caring because once you care, once you are interested, you have a vantage point from which to make Quality judgments. These Quality judgments (e.g. "Is this good code?") are based in part on the romantic mode of understanding...

This resonates with me. I keep telling my team: engineering is finding the right problem, determining the constraints, and guessing a solution. Maaaybe the LLM contains a good enough guess, if you're lucky.

The other two aspects are unbounded search problems, and so by definition irreducible.

Good luck automating yourself out of a job, but be careful, you might become an engineer.

I guess I meant pre-4.5. Claude Code peaked at or before November 2025 IMHO.

This is awesome. Adopted it in my personal pi-mono repo :)

It had some bugs around whitespace replacements but the model seems happy with it now.

Thanks and keep it up! Shoutout to @jahala for tilth as well.