rcarmo

Good. I have been poking at rcarmo/go-264 but was hesitant to tackle encoding :)

Depends on what you are using. I have been experimenting with various compaction techniques in piclaw (https://github.com/rcarmo/piclaw/blob/main/docs/pipelined-co...) and Codex has native "opaque" server-side compaction at a dedicated OpenAI endpoint.

I do hope they provide optimized A3B quants--that's been the sweet spot for usable local inference for me, at least.

Yeah, diarization is the real feature these days. STT needs uniformization, but quality of diarization is what is setting personal solutions apart in this field.

And yet, downvoting is still working within seconds of my commenting! Rock on HN, never change.

I find it fascinating that I submitted this yesterday and it failed to get any traction then - is it the AI spam that has turned submitting stuff less visible?

Still doing https://rcarmo.github.io/projects/piclaw (which is my main “IDE” now), ended up forking https://github.com/rcarmo/tau-prime to have something I could run on my iPad (and on the Mac, with a simple sandbox), hoping to get back to 3D printing (https://taoofmac.com/space/notes/2026/07/12/1230) and playing with hardware.

Oh, and migrating most of my stuff to microvms on https://rcarmo.github.io/projects/pve-microvm/

well, I created https://github.com/rcarmo/piclaw, and as a result of that I have several permanently running agents and entirely too many projects going at once. That's _my_ "messing with AI"...

I have a checklist to go _back_ to reading 30-odd books a year, and right now the top 5 items are:

1. Stop messing about with AI

2. Stop doomscrolling/interacting on social networks (HN is within my 15m allocation)

3. Stop watching _any_ Youtube video that doesn't teach me anything

4. Gloss over my 200 RSS feeds, don't be a completionist

5. Put on classical music, not indie or radio

It almost works. Almost.

Actually, not really. The vocabulary is sophisticated, but there isn’t a single “the honest value” or A bunch of itemized bold fluff.

This is a thing of beauty, no matter what the general feeling in the comments seems to be against LISP. And yeah, you can do it in JS/TS/Python/etc., but somehow it doesn’t feel as elegant.

Very nice, but needs pi.dev / rcarmo/piclaw support :)

This is pretty much what I did, except that since I use Proxmox I used that to manage the volumes and then just did a pass-through mount to an LXC that runs Samba - that way I get a decent, simple GUI to monitor and manage volumes and can do user management inside the LXC.

People overestimate what LLMs can see vs what they tell you they can “see”

Funny to see it took the same IR approach as I did with https://github.com/rcarmo/go-joker - although I did it somewhat based on the .NET IR and this seems a bit more AI-ish.

Yes, but automatically generated content doesn’t need to do that - like I wrote in another comment, my agents know how to check bounding boxes.

That’s probably because you have (or implicitly created) a skill that does that. My agents know how to check bounding boxes.

People want to generate corporate content. For years now, it comes and goes.

AI has zero need to see the renders (you are just wasting tokens) and that is why I built the Go version - I patched a lot of the PPTX code in my MCP server, as it happens.

Nice, but I don't see a lot of ECMA 376 test cases. Both https://github.com/rcarmo/python-office-mcp-server and https://github.com/rcarmo/go-ooxml are ECMA 376 compliant (I made sure), because for headless generation and handling that's kind of important :)

Oh, and you're not the first, I started this a year ago. :)

Nope. If you use pi and its extremely minimal harness you can see a _lot_ of difference in models.

Yep. I spotted the same thing in piclaw (which relies on the pi runtime) but did not have time/energy to do a lot about it—and fable does the same, as far as I can tell, with one out of five or six edits failing. But I prefer OpenAI models for coding, so it wasn’t a real problem.

Years of regular use here, has been great even before I started using it as an agent tool.

As someone who also works for a US company with very large EU customers and partners I can attest this is completely true. Most European people on HN seem to be in the startup/SME space, so this won’t resonate, but the key point is that people who work for US companies also have zero incentive to switch to an EU company due to the mindsets we see locally.

I like the idea, but without a cross-platform OSS browser it’s hard to consider adopting it (and I am primarily a Mac/iOS user…)

We could fix a lot of this by just making sure .local (which is used in Bonjour/mDNS) could coexist sanely in mixed resolver environments _and_ could support subdomains. I built https://rcarmo.github.io/projects/mdnsbridge to “fix” it for my particular use case, and if it wasn’t for TLS shenanigans and the lack of subdomains, my issues largely went away.

Why isn’t there a black bar at the top of HN yet?

Oh fuck. I used to swap notes with him every now and then, drifted away over the past couple of years. Had no idea.

You should just integrate with pi.dev, like I did for https://github.com/rcarmo/piclaw (which has replaced Obsidian for me). I too integrated a terminal and a WYSIWYG Markdown editor (as well as plugins for a mindmap, kanban, etc.)

Nice, but as usual if you want a 3-step “getting started” example you have to wade through the docs, and even then…