@taye

145 Followers
339 Following
327 Posts

I program interactive things. Currently working on video editing tools at https://miru.media. Sometimes I make art. Rarely I am spicy.

Somewhere around Berlin.

codeberghttps://codeberg.org/taye
githubhttps://github.com/taye
Thinking of updating and polishing up this Vite plugin for making GTK apps with Vue that I wrote way back, but with JSX and Adwaita. Then I could translate my browser based video editor into a #gnome app with #GStreamer instead of WebCodecs?

I feel this actively maintained Offline Translator app deserves more attention (available via Fdroid) (e.g. in combination with Whisper + voice input), comes quite far as a replacement for BigTech translation apps. This project was funded through the NGI Mobifree Fund, a fund established by NLnet. .

An Android translator app that performs text, PDF/ODT documents and image translation completely offline using on-device models.

Supports automatic language detection and transliteration for non-latin scripts. There's also a built-in word dictionary.

https://codeberg.org/davidv/offline-translator

RE: https://neuromatch.social/@lina/116779058863273132

If anyone wants to prevent LLM contributions, it seems like symlinking CLAUDE.md to /dev/urandom might do the trick: by default, claude code unconditionally reads CLAUDE.md if it is present in the repository root. The reading function checks if the files to load are block devices among a few other things, but does not mind at all if they are symlinks in general, including directory traversal to anywhere.

There are ways to run claude without reading CLAUDE.md, but they require manual intervention, and startup stalls before any of the "try whatever" self healing attempts of the LLM kick in. indeed there is no indication why startup stalls in stdout or debug logs: nothing for an external "agent" harness to diagnose either.

I have not confirmed this part: but it looks like the file read tool checks if a path is a system path like /dev/urandom, but it only checks the string of the path without expanding symlinks. Many of the tool and agent prompt texts say to read CLAUDE.md first, bypassing any config settings that would prevent that since it happens in LLM modality not program modality. so even if you do get it booted up, you would likely get inexplicable stalls at any point those tools/agents are invoked.

Went for a stupid little walk for my stupid mental health and brought you back some.

#Bloomscrolling #bloomstrolling

That sample video provided by @slomo in https://gitlab.gnome.org/GNOME/showtime/-/work_items/288 is 10/10 high-quality bespoke GStreamer-powered performance trolling of GNOME's entire graphics stack—UI toolkits, window management, compositing, all at once—I love it! 😂

It's like half of @juni's "Shaker" (https://mew.tv/projects/shaker.html) without even needing to invent a new protocol for it!

#Showtime #cats #CatsOfMastodon #DaftPunk #Wayland #GNOME #GStreamer

Window resizing is confusing on clips that change the resolution all the time (#288) · Issues · GNOME / Video Player · GitLab

Try e.g. attached video, which changes the height on every single frame. Looks...

GitLab

I have very conflicting feelings about this :D

#electronics #repair #righttorepair

We've tried making a build with a randomly generated 1GiB file included to make GrapheneOS about as large as the stock Pixel OS which fully works around the issue. We're not actually going to do that but rather we'll use the workaround forcing the fallback path for now and we'll find a proper fix.
When doing the little capchta tasks I always make small mistakes intentionally. Because to err is human.

New: #Small #AI: A #degrowth imaginary for designing with/for artificial intelligence

"This paper presents Small AI as a conceptual lever for designers to resist the ‘bigger is better’ trend in the discourse & development of Generative AI. We explore different types of Small AI: initially in relation to the size of models, as smaller models require less computational resources for both training & inference, thus implying a smaller ecological footprint. However, we extend the meaning of smallness to apply to political, epistemic and cultural domains as well. In this vein, the paper posits Small AI as an imaginary-in-the-making, & a rallying concept for an alternative innovation agenda that holds, in the spirit of E.F. Schumacher’s famous dictum, that it is small (AI) – not big (AI) that is beautiful."

Bendor, R., Murray-Rust, D., and Rehak, R. (2026) Small AI: A degrowth imaginary for designing with/for artificial intelligence. Design Research Society (#DRS26). https://doi.org/10.21606/drs.2026.1193

I'm appalled to see the European Commission apparently embracing closed source, commercial, ID and age verification mandatory, social media today - in the very face of the technology sovereignty and online freedoms debates.

I think they've just thrown away the greatest opportunity to be the champion of all things open and collaborative, and to live and demonstrate their own values. Incredibly disappointed.

🔥 🔥 from @_elena who nails it, and asks crucial questions.

https://blog.elenarossini.com/w-social-public-institutions-and-the-theater-of-european-digital-sovereignty/

W Social, Public Institutions and the Theater of European Digital Sovereignty

European public institutions are betting big on digital sovereignty and open source software. And yet they recently moved their Bluesky accounts to W Social, a private, for profit network owned by Swedish entrepreneurs that has quietly gone closed-source.

Elena Rossini