tingletech

I use it with a local Qwen3.8 27B on llama-server, but I don't know what it means for it exceed any window.

you can use /slots on the llama-server if you want to get more up-to-date details on session token use.

I always run models in their default context, which is 262144 for Qwen3.8 27B. I've run sessions in hax where it hit that cap multiple times and compressed the context down to 15% and continued with no problem.

> and its “authors” are anonymous.

The blog posts are all attributed to "Lucas Vieira" and the dev group https://faisca.dev that is attributed as the author has 2 other projects. Lucas comes up in LinkedIn and looks like an actual person working in San Francisco.

Re: curl; this seems to work:

  cargo install fakecloud

SearXNG configured as in the OpenWebUI docs is pretty cool. My "Hello World" with a new agent framework is teaching it to use SearXNG. Hook this in as a tool and the agent can answer a lot of questions.

SearXNG is more of a metasearch, the dude who wrote it pops in on here and is working on a cool sounding project that is more like a local personal search engine, I forget the name, but I've been meaning to check it out.

There is also Common Crawl.

https://docs.openwebui.com/features/chat-conversations/web-s...

https://github.com/brian-learns/xng-agent

Neal Stephenson's Snow Crash came out in 1992. I guess the paper back became popular in 2000 though.

Language is the virus

"backbone" is what has me hung up, seems more like a scaffold or maybe a bionic exoskeleton.

maybe it's regional? I always thought "Could care less" was in the same class as "irregardless" or "daylight saving time" -- in common use yet technically incorrect

Don't get me wrong, I love `uv` -- I mean as far as being "the backbone of any modern Python library" -- shouldn't any modern python library work with any packing system that follows the PEPs?

they just implement the modern PEPs

in the US you can still buy hydrogen/ammonia/water gas absorption refrigerators such as "EZ Freeze". Propane/Natural Gas refrigerators use this cycle. I'm sure you can get them all over the world, the cycle is older than the one in the article.

very interesting, basically the agent is a python class, and state is saved right in the object.

the paper https://arxiv.org/abs/2607.20709

the code https://github.com/nvidia-nemo/labs-OO-Agents

I wish there were a toggle on google maps to show the internationally accepted names for these bodies of water.

(2020)

previous discussion in 2020 https://news.ycombinator.com/item?id=25023673

This 2025 discussion links to this too https://news.ycombinator.com/item?id=44117779

https://github.com/ryaker/SparrowDB is another one in this space

this opens some crazy "apple developer" dialog on my mac rather than a web page. I had to crank up chrome to read it. And it makes no sense, I have no clue what a sign on with apple address is. I guess I'm glad my hide my email addresses won't change because that would be a PITA.

I haven't looked at the book since the 80s, but what I remember is there is a pre-read, the fast read, and then the deep read as needed depending on the task.

pre-read is just looking at how long it is in the headings, and planning out what chapters to focus on if it was a text book. (its sort of iterative, you do a pre-read for the whole book, and then for each section you break it into)

The fast read you try to read only with your eyes, sweeping your eyes across multiple words at the same time, suppressing the urge to say the words to yourself in your head.

iirc the how to read better and faster book even had a cardboard mask you put on the page to practice the sweeping, and some pages that were laid out weird to try to teach you how to do it.

Some ai text just seems really easy to speed read, like if it's tuned for an easy reading level. In PRs some ai seems like it's arguing over weird flex technical details and really starts torturing the language in a way that makes it the opposite of easy to read.

https://felonybench.org/ and https://felonybench.com/ seem unrelated?

One's hosted on porkbun and one's hosted on namecheap.

UK has a shared print program

http://ukpbc.uk

I think under first sale doctrine you have a much stronger case with destructive scanning. Google Books, HathiTrust, and Internet Archive's book scanning project have had a lot of legal expenses.

I don't know what Liberia has to do with it.

Weeding (deselection works) is a fundamental part of collections management. Every trained librarian is going to understand this.

The interlibrary loan system has mechanisms in place to make sure the member libraries keep two copies of each work in each region. Collection managers consult these databases during weeding to make sure they don't deaccession the last copy.

If the monograph was never collected by a library and it gets caught up in a destructive scanning project then I guess it was pretty "rare" in a literal sense. "Rare Book" in library land is sort of a term of art and I'm not sure if the books in these destructive scanning projects meet the criteria.

In high school, a teacher gave me a copy of "How To Read Better And Faster" which teaches you speed reading. This came in very handy in college.

I find that when I try to speed read modern human writing, there are often errors (like missing or misused words) or awkward expressions that I do have to slow down and think harder a lot to really parse it.

With AI writing, it's sort of self redundant and the information density of each sentence seems to have more even information density. This makes it very easy to do a very high level speed read and get the full gist.

There are also what I'm assuming are bots on hugging face (or maybe non-native english speakers who are using ai for translation) that interact with me where I have no idea what they are saying until I read it very slowly.

I've got a script that checks out and builds the latest tag

https://github.com/brian-learns/llama_cpp/blob/main/update.s...

what's the significance of this? There is also a v0.1.1 from 4 hours ago, but these v0.1.x one's are on the releases tab https://github.com/ggml-org/llama.cpp/releases

seems like stray tags?

Weeding (the library term of art for selecting works for deacquisition) is a very import part of collection management. It's actually regionally and nationally coordinated so the interlibary loan network does not throw away the last copies.

https://cdlib.org/west/ https://papr.crl.edu https://eastlibraries.org

yes, it says "Now you can have beautiful clean workflows!" and links to a youtube video of a TV ad for a dermatologist.

they talk about the quants they tried in the article and settle on a Qwen3.8-27B-Hermes-iMatrix-NVFP4-Balanced.gguf which they calibrated on their own session traces and they pulled in 5 different llama.cpp pull requests to their local llama-server.

> If I don't have a fully kitted out test suit and a good type checker to verify the final product against

I have a template I use with generated python projects that sets up all the test info first.

https://github.com/brian-learns/testafize

I start the coding session in a directory where I've already set this up and maybe drafted a README.md, and usually it just automatically starts running the `make check` without prompting.

I've also used it for migrating old code. I put an old python 2 script in src and asked it get make check to work, and it even added test on its own initiative.

I'm mostly using Qwen3.6 (now Qwen3.7) running on my DGX Spark in llama.cpp using different harnesses, or the default model on opencode.

Google scholar thinks I'm a robot and won't let me look at the link

you still seem to be conflating layoffs with firing someone where you have to train their replacement.