Please share what operating system and model runtime you use? I have two and don't get close to that with AMD's own Lemonade. Thanks!

Lemonade is one of the worst performing options. Run any modern Linux distro and ask your current LLM to setup llama.cpp with dflash2 for you as an unprivileged container running from a systemd user unit.

Obviously only on a system you do not trust at all.

Shame the page does not tell you what HTMX is, even in one sentence.

The explanation is a single, obvious click away.

Pretty funny. 2013 though.

I didn't believe it till I tried to read it. Insane. Impossible to read.

I am a big Lisper. You lost me at the triple quote.

hey! indeed, it is a pattern more from Python/Kotlin/etc and on first sight it looks awful for those used to work in more lispy ways.

pngine is using a data language I made for fun with s-expressions and appropriately named "SJON". It is a data language (kinda like EDN, or er.. JSON) to declare things, not to run/interpret things, it has no reader macros, no quote/unquote, no eval, no cons or car/cdr or..., the only thing similar to Lisp is really the surface look, and on the surface triple quotes really feel like they don't belong in there.

But I need to paste WGSL into a config file and WGSL can be full of " and \, and I have no reader to extend.

In a way I'm glad you didn't spend a lot of time and got lost at the start ahah because if the triple quote bothered you the rest would too.

For instance, in SJON strings don't remember their quotes:

  (= """hello""" "hello")     ; true
  (= "a\nb" """a
  b""")                       ; true, escapes resolve before comparison

It is a data language, and one thing I wanted was for numbers to carry their units:

  (camera :ortho :zoom 2 :fov 90deg :shutter 250ms :scale 50%)
  ; unit is a suffix glued to a number with no space
  ; oh! and dates and times are also token kinds: 2026-05-19 and 23:59:59.999 lex as values
  ; these don't get interpreted, just carried over and then the schema decides what is valid

yeah... there is a schema, which is written in SJON too, that is validated ahead of evaluation, and docs have a binary IR, so a tiny runtime can consume one without carrying a parser.

For instance, the SJON schema for the WebGPU spec was sketched from the spec API.

Sorry for the long answer, yeah, it is not Lisp, just S-expressions to leverage a small part of their amazing versatility.

Needs a Lisp Shading Language.

ahah indeed! when I was young someone or something etched into my tiny brain that "thou shalt never abstract a DSL, it is wrong!".

I think I should have outgrown this by now, there is some truth to it but I feel it is certainly not absolute at all, and this is essentially a for fun project with no rules or greater intent beyond that.

The triple quote screams scrambled eggs with bacon.

hahah loved this one! so true, can I have a bit of onion too?

Surprising. I have hit its BS guardrails a lot lately, working on my vintage computers from the 80s and early 90s. Just about done with Claude.

You can’t be loyal to these things. I ditched ChatGPT during the peak Claude hype after Christmas.

I feel like Claude has shittified a bit and ChatGPT is good and fast. Gemini remains mediocre, although it seems Google AI energy is directed elsewhere.

Just sign up with something like openrouter and keep switching models until one completes the task.

Paying by the token is a much more expensive way of doing it than signing up for the various coding plans when you need them.

Not if the models are way cheaper. For many tasks you can do with DeepSeek Flash, and for more gnarly problems you switch to GLM or Kimi. But sure, if you do everything in large models like Kimi K3 or GLM it gets expensive quickly.

I’ve been doing this with Qwen 3.8 27B with success. Kindle, Android Tablet, and Raspberry Pi all working better and fully owned thanks to agentic help. No issues with hitting the guardrails here ofc.

What tools/skills are you using to do this with Qwen 3.8?

What about Raspberry Pi is not open?

The hypervisor running on the VideoCore, booting up the ARMs, and the OSes that run on them:

https://en.wikipedia.org/wiki/ThreadX

Initially at least. Had some changes of ownership and rebrands meanwhile.

Now playing: "Hot Chocolate - It's just an illusion"

The boot firmware.

What did you do with the Kindle?

kimi k3 is wonderful, I suggest you give it or GLM a try when you get the chance, I just use openrouter or cursor

What were you trying to do with vintage computers?

In this case, get an ancient VAXstation running with VAX LISP so I could port my Z-machine interpreter from CLOS to their CLtL flavors implementation and try to get Zork and other games running on it. But goodness forbid you ask Claude for help with 35 year old software without hitting their guardrails.

I am getting close to dumping Anthropic. I like their models in general, but boy I have hit their "F you, I ain't gonna help" too many times now on innocent things. Ain't nobody wanna deal with that. I have never gotten that from Antigravity and if I did I woupd tey Codex then go to Openrouter and leave US models behind.

I mean, who says "screw you" to requests to get 35+ year old vintage computers working? Claude, that is who. Its guard rails are so stupid. I hear people trying to do simple mailing list management hit it too.

I am just about done with them.

GLM 5.3 will probably do your vintage computing work without fuss.

I am still disappointed that the big 6502 is neither open sourced so you can build one nor commercialized so you can buy one.

Sorry, what does "everyone is sleeping on it" mean?

It's receiving less attention than it merits

It is under-utilized and not getting the attention it deserves.

I have a Flip 7 and did the opposite - turned off all Samsung stuff. Maybe that was wrong? Can you give some concrete examples of useful Samsung AI stuff please?

For example search and OCR in the gallery and camera app.