Yea, I suggested as much. Good to see them following through

I'm currently a bit unwell. But if anyone wants to talk to this person, they are responding here.

github.com/henriksson-l...

The point is to speed it up... But then it's slower. Lol 😹

And they aren't even using a single brain cell to think about licensing.

Should we ask them to stop?

After writing my own poa stuff for better consensus calling, this commit was extremely satisfying

"yeet the bio crate"

This one hit me right in the sore back

pretty sure this lined up with an outage/issues

I doubt 1 person can maintain this many repos, effectively, even with LLM assistance. I have to ask the question: "why tho?". Seems like flag planting rather than doing anything useful.

Yea most likely.

Pretty sure it will fail on these shorter, repeat reads too. It's more the consensus part that was not working.

Trying to write one complex tool, I have now written 3 other tools: A plotting library, a liftover type tool for repeats/SV, and now a POA library that actually does proper consensus building, even for repeats.
Maybe i'll finally be able to finish this tool 😅

Basically, never trust an LLM, and don't outsource your thinking to one either.

claude: It is definately A
me: it's B
claude: You're absolutely right, it's B
me: I was taking the piss, it's actually C
claude: oh you got me! it's C
me: it's actually D
claude: You're absolutely right! it is D
me: no, it's C. What the hell is wrong with you
claude: oh you rascal, you got me again!

Don't forget, Claude is also analysing the results of those tests. Marking tests as `#[ignore] and saying things like "this was a pre-existing warning/error, it's fine"

Yea, I do wonder how the testing is being done in these cases. Really complex software is a lot harder to do this with (ask me how I know 😅)

system76 system? I spy a CD?!?!

Need flamegraph to stop using perf especially on Ubuntu 22.04 with bugged perf that takes 10 years to do the simplest thing

Nice one mate!

Honestly I think LLMs for code are probably the least egregious use given the skills are mostly built from open source and freely available information. It's everything else they were trained on that I have an issue with, and how they are used in those areas.

me: I should add some interactivity to my plots

also me: *Shoves fully playable DOOM into an SVG file*

cargo build --features cli,doom
kuva doom -o doom.svg

open in browser. that's it.

Yes, it's April 1st, but this is real 😅
psy-fer.github.io/kuva/doom.html

big kuva update incoming 😎

Rip and Tear my friend.

I'll release it with v0.1.5 (next release, which is almost done), and embed the SVG in the docs so people can play it.

But if people wanna make their own, they can just clone the repo, switch to dev, compile with --features cli,doom and do kuva doom -o doom.svg

I got a little carried away making interactive plots using native SVG features. Then @jpelbers.bsky.social asked if it could run DOOM. It was probably a joke...but I took it as more of a challenge.

Anyway.
kuva doom -o doom.svg

Fully playable and fully contained DOOM inside an SVG

I may have gotten carried away...😅

New update should be done soonish. Lots of fixes and new stuff

Cooking up a cool feature for kuva with --interactive, where it adds a search, and group highlighting and coordinates on mouse, all within the native SVG file, when looking at it with a browser. This is just the minimal working prototype. What do you think?

depends on tool, it's popularity, and how complete the tool is on release. we don't tend to release things that don't have most of the features already implemented. Write good tools first, then maint time is low anyway 😅

our lab has made a lot of tools...

speaking of which, i need to flag some issues with the rust bio crate and their POA stuff crashing when encountering homopolymers. Wrote my own banded-poa as a fallback for now.