Tootfinder

Indonesia’s native hornbills are being hammered by online and offline trade https://news.mongabay.com/2026/06/indonesias-native-hornbills-are-being-hammered-by-online-and-offline-trade/

Indonesia’s native hornbills are being hammered by online and offline trade
Among the many inhabitants of Southeast Asia’s dense rainforests are hornbills — a group of birds that stand out with their raucous call, large, ostentatious beak and colorful feathers. Indonesia harbors 13 species, the most of any country in Asia, three of which are found nowhere else. Hornbills are rapidly losing their homes as large […]

I'm conflicted about what to do

• Codeberg would screaming "I'm anti-genAI" which is something I really want
• git.gay looks cool and fun and gay, which is also something I really want
• self hosting would give me full control over the server configuration
• sereia.gay is a cool domain, but it mostly only works in portuguese (where git.da.sereia.gay is cooler). git.luana.dev.br is also an option (and cheaper, aka less likely to stop being mine), specially if I can set up more than a single domain to point to forgejo, but it isn't as cool.
• even if I set up a mirror, I'm undecided about whether to point it to codeberg or git.gay. Three-way sounds kinda abusive, idk?
• while I know forgejo is pretty lightweight, I'm concerned about having too much access in my NAS. Specially if I end up being ddossed by llms or if I ever do something people like. I have plenty of RAM from before the bubble, and it's modern but it's not the highest-end CPU out there. I'd also have to set up a reverse proxy, probably more reliable than my current oracle free tier one.
• I could get a VPS for the forgejo tho! Ive been wanting to get one anyway! (tho it would need a static ipv4)
• Tho, keeping it self hosted on my NAS would give me git push/pull even when offline!!
• Even if I keep a three-way mirror, which I might, I still need to choose a canon one to send links!
• Is it ok to accept PRs and stuff from either mirrors or do you need all but one of them to be read-only?

The Guardian's tech team share their shoe-leather experiences covering AI data centers, as field investigation becomes increasingly important in tech coverage (Gautam Malkani/The Guardian)
https://www.theguardian.com/membership/202

‘These are some of the most complex structures ever created’: how tech reporting moved into the physical world
The Guardian’s global tech reporting team are investigating the impact of the vast datacentres being built to power the AI revolution. We spoke to them about how their beat has become increasingly offline

Whenever some tech bro who only moves from one Silicon Valley hivemind to another adds a genius feature that requires being always online (or locks offline downloads behind a paywall), I am reminded about it daily on my commute.
I live in Germany, there is no 'internet' here. There is a crude system that hasn't been overhauled since the 90s that mimics what other nations call 'internet', but it is not real, nor is it your friend. It's an old German bureaucrat who files each pa…

The Brain That Goes Quiet: Serving a Large Model's Knowledge at 131 Tokens per Second on an 8 GB Laptop by Removing the Large Model from the Runtime Path
Myeong Jun Jo
https://arxiv.org/abs/2606.12154 https://arxiv.org/pdf/2606.12154 https://arxiv.org/html/2606.12154
arXiv:2606.12154v1 Announce Type: new
Abstract: In earlier work I showed that a 35B-class Mixture-of-Experts model can be loaded and executed on a consumer laptop with 8 GB of GPU memory. That result solved a placement problem and immediately exposed a different one: even correctly placed, the large model needed roughly four seconds to answer, because it was still being invoked at every query. This paper documents what happened when I stopped invoking it. During an offline phase, the large model reads source documents and writes verified answer entries into a structured knowledge store; at runtime, only a lightweight router, a deterministic renderer, and a 1B-class model are active. On the same 8 GB laptop, end-to-end response time fell from approximately 4,465 ms to 518 ms, effective end-to-end throughput rose from 15.7 to 131 tokens per second, and the small model's streaming decode rate held at 226-237 tokens per second with a time-to-first-token of 29-62 ms. The bottleneck is structural: three different large models (Qwen, Gemma, and GLM class) all showed the same multi-second runtime cost, and all three produced usable knowledge stores offline. On a 563-entry store built from seventeen real documents, keyword routing collapsed to 1.5% top-1 accuracy while BM25-based routing reached 92.8% (99.4% top-3), and a confidence gate raised effective top-1 to 98.0% by escalating 12.3% of queries. Exact-match fidelity of the small model ranged from 9/9 to 0/9 across envelope formats carrying identical content. A 16-case verification gate blocked all ten corrupted entries while admitting all six supported ones.
toXiv_bot_toot