tarasglek

doesn't asking llms to do diffs result in a similar problem? Any additional complexity due to formatting will degrade reasoning

why not have more granular versions of whole file editing, eg whole function? I find function at a time coding yields best results for me when I write code in chatcraft.org

I really wish to find ts tooling that would show me deltas in ts checker memory usage and tie them to diffs in ci

Their ui is also hosted on the nvr, they serve ui assets over webrtc

Doing http over webrtc is how https://camect.com works to let one access cameras own private server via their ui. They have a centralized bit for auth and then use webrtc and a physical nvr to serve your videos maximally efficiently...so there is low risk of their cloud becoming a financial burden that they cancel ala google nest cams

It's a super nice architecture

I found it easier to use ssh over websocat over cloudflared. Then you just need websocat again on clientside and can use regular ssh client with it

Been experimenting with aider for a while. Really fun using it for basically free with deepseek. Thanks for getting that to work.

My question is re your whole file fallback approach. Since you already have everything mapped out with treesitter, have you considered doing function at a time instead? It's just as easy for a model to ask for a function as it is for a file

This worked really well for clickhouse and sqlite. Any plans to open source?

Is this official? How does one use this. I'm a very newbie whatsup so sorry for dumb q

This is the first timer i've seen tinygrad backend in the wild. Amusing that it's supposedly more stable than llama.cpp for this project.

Weird to be comparing vs rsync. Rsync is notorious for maximizing latency as it has no pipelining.

Better baseline would be transferring data via ssh + tar. Or serving a directory over http and concurrently downloading a file with axel

That's so cool. I just recently saw a more recent impl of this: copying fb over spi on rpi. This approach is much more clever and equally useless

I don't get it. How does one use vga or HDMI to hook up i2c?

pikvm solution is a Raspberry Pi Zero 2 W + HDMI CSI cable + pi pico for atx control + usb c splitter. It's actually cheaper. And once you put it together, you run stock pikvm image with smooth upgrades, etc.

Latest prices for highthium cells are $40/280ah

They can be used to replace ups.the voltages are close enough with lfp cells. I've done it. So long as you have a bms, there is nothing to worry about

What's a good cheap invertor vendor from china these days?

We are looking at ordering a lot of these systems in Ukraine to deal with russian attacks on power grid

I proposed this at firefox when I was there, was deemed as too niche a feature...I'm still sad that I can't have this feature

Took me a while to realize that I can't execute JS blocks. Kind of unexpected that go works and js doesn't in JS a web app :)

Yes, there needs to be some legal mechanism to release specs and diagnostics if manuf can't provide service...eg cos they went bankrupt or where all licensed technicians are 6mo backlogged. Some sort of a legal documentation escrow where the info is released every time the manuf isn't living up to commitment

I think turso in memory replicas might work here. They have some code to capture pg cdc to do that.

Likewise using something like pglite to run a pure pg in memory replica would be very cool

I love that this dataset is accessible via cors from web frontends. Got me into experiments with clickhouse

Same with github, much better than than throttled apis

It was the most robust way to write c++.

Ownership model in Symbian lead to really good memory efficiency. The testing tools made it really hard to import open source code cos it brought out all the memory correctness issues in addition to the usual porting problems. This was the safest systems code before rust came along.

Most devs did not understand that it was a "take your medicine for healthier code" situation and hated the dev experience

you will love yaml since its a similar improvement in token use over json

1. For reading. I simplify all documents to mdown then render back to html on my readers...e-readers and rss

2. Saves context for llms and they are often trained on markdown so work best with it.

3. For search. Can search markdown much better than html with postgres

Wrote https://markdown.download to help me with these

Not a hardware guy,how does one crimp and secure those custom ends?

Unclear from docs, does your solution support inferring number of speakers from audio? Found it a bit frustrating that this wasn't automatic in diarization algos I tried last year

This is interesting.

Newb question, how do use opensk/tock as an hsm? A quick search points back at this thread.

gof3r looks amazing, thank you!

The metallica-esque lightning logo is cool

Can you give a concrete example of GNNs helping?