scirob

To elaborate on my usecase that herdr didn't so at all the the beginning but still does not do well: I have many vm's in different clouds working on different projects with different secrets that I want to keep isolated ( now also at home lan machines which when im travel get aliased to their tail scale address)

So my biggest requirements are around making isolation between environments convenient but safe from supply chain attacks .

I wonder what usecase you were optimizing for

I also built a similar thing a many. And in my experiments (even though iroh is great) iroh took significantly longer to reconnect vs mosh and ethernal. And actually plain old autossh was the best XP for me to close laptop get on traincopen again on mobile Internet.

So eventually I removed everything other than autossh.

Seeing as how many people are building similar things I'd love to share more. Maybe wanna ad each other respectively to the repo's ?I can share some of my benchmark results

64gb is the way

So for German Language this is now actually the #1 model from what I can see

https://dach.peerbench.ai/compare?models=meta%2Fmuse-spark-o...

So it does mean something to be the top western model if it means better performance for western languages.

sry late benchmarks failed first few runs.

this 3.5 Super is still below Qwen 3.8 next as they state themsleves

they also released an interesting smaller model https://news.ycombinator.com/item?id=50104922 3.5 bolt it uses 2bit quantization and beats the other small active layer models

No one is going to buy this from meta.

But there is a usecase to have always on listening device transcribing everything. I have a coworker that programmed an app for his apple watch pro so when the button is pressed it starts transcribing the meeting its super super handy for meetings on the go IRL without a google meets room. ^I looked into doing this in android its impossible in the SDK at the moment can't get access to the button.

I have personally build an always on local first setup in my office. It listens for me saying "send this to __employee_name__" and other similar things to do in parallel while i keep typing on my work instead of alt tabbing into w/e slack app that then might have an expired session one must login to again. (Not done but i think there are usecases)

i'm sure people have also heard of that pebble ring Pebble Index 01 many people are super stoked about it and I was about to click buy until i saw A) single use battery no recharging B) 15second timelimit or something else very short.

SO IMO there is a market but its hard to get the product right

Google continues to be the best in German language and Gemini 4 is a significant jump in our benchmarks

https://dach.peerbench.ai/compare?models=google%2Fgemini-4-a...

no hugging face link :( ... but hey its on openrouter yay

Here are my results

https://dach.peerbench.ai/compare?models=mistralai%2Fmistral...

Looks like a bit better than the recent Kolibri-1 but still below Qwen3.8 27B

"Rolling out soon" don't let them hype without any release

Nice we have a document we can point at of someone doing a space data center project pointing out all the flaws and benefits without hallucinations.

Putting data centers in Antarctica or extreme north Canadian islands would also put you out of the way of molotov-throwing joe public. Even in space your not safe from the voting public as long as your companies directors are sitting on earth somewhere.

Putting data centers in Antarctica or extreme north Canadian islands would also put you out of the way of molotov-throwing joe public. Even in space your not safe from the voting public as long as your companies directors are sitting on earth somewhere.

Just got benchmarks done for German langauge eval index i help maintain. Hy4 is a big improvement over h3 but still below Deepseek pro and Significantly below GLM 5.3 Flash .

hy4 ranks ~14th overall

https://dach.peerbench.ai/compare?models=tencent%2Fhy4-previ...

https://dach.peerbench.ai/models/tencent/hy4-preview

love the direction have asked myself manytimes why this feature sent doesn't exist somewhere; for sure would need to see some high profile people start using this to think its bullet proof

Maybe it's my dyslexic brain but "it's own frontier model" in my head converted to foundational model that was trained from scratch.

But this is qwen based.

But w/e I'm pro AI so more companies having more people with skills for more post training is cool

I tried to build a few non crypto decentralized apps. The killer was reliable always deliverability inside the browser. https://inbrowser.link/ was a huge jump in utility but came to late the ipfs.js just never worked consistently. In the end the only way to get good XP for yours users was that you provider the IPFS to http gateway for all content but then whats the point its just decentralization theater.

I remember in 2015 having the IPFS concept blow my mind its such a memorable moment when it really felt like someone designed something significantly different that current mainstream paradigms. But in the end it seems like it was still a case of a cool technology looking for a use-case not solving a real problem.

oh man and when Brave removed native IPFS support. https://github.com/brave/brave-browser/issues/37735

can this not be achieved through a system prompt adjustment ? Running yet another LLM call on top might not be the simplest solution.

Vampire survivor is an amazing game. If you can't accept the value of entertainment then the rest of your life is probably to easy.

But fair enough everyone can have their take.

We maintain German Langauge index as no one publishes or reruns these sepeartly.

Qwen 3.8 27B is a small improvement with some regressions in our benchmarks not a huge jump like benchmarks listed.

https://dach.peerbench.ai/compare?models=qwen%2Fqwen3.8-27b,...

German language has never been a big focus for asian models but they still outperform Gemma models https://dach.peerbench.ai/compare?models=openai%2FQwen%2FQwe...

So in production we have been using Gemini Flash Lite as primary and fall back to Qwen when gemini servers are overloaded or just giving us 429

Same question.

I usually have one local clause orchestrating multiple remote Claude in different tmux . And then another orchester and remote vm worker sin tmux for another repo etc...

It gets a bit hard to keep the overview but I don't want to give up my parallelism, your too might help

They had this before I used it to ship some stuff but binaries were big . How small did they get it with this update

Yea pls protect it feels mega oppressive how most EU countries require you to have the SIM card guy in his corner store scan your passport.

Quickly deployed it to check some benchmarks relevant for German language. These are results for CohereLabs/include-base-44 german only : Gemma 4 12B %61.9

  Gemma 4 26B (a4b MoE)    0.647
  Qwen 3 14B               0.621 
  Gemma 4 12B              0.618
  Ministral 14B 2512       0.604 
  Gemma 3 12B              0.547

The quwen 3 14B vs Gemma 4 12B difference is within random variance they same in some repeat runs they actually got the exact same score. Next step up Gemma 4 31B gets 0.676 on this. Or let in some reasoning Qwen 3 14B (reasoning) 0.676.

I'll run some cheat-proof benchmarks ones tomorrow see if qwen is still on top.

Idk n26 still not integrated. Twint in Switzerland has amazing penetration can pay at any terminal with qr code

I wonder why googles biggest European office is in the low tax high salary country of Switzerland

OMG just do, EU needs some balls. Upcloud, Scaleway, Hetzner, OVH are all production ready . EU business leaders are so afraid of not using the biggest provider they are blind to how you IT can absolutely run 100% on compute from these providers and opensource

Great, it's so easy to automate checking ref super bad to not check

It's in wasm so should be possible to code up the next language

I can confirm init7 and the whole setup in CH is indeed working well

haha for sure some one has made a little aggregator for this and saving tokens. I bet you gotta dig for a while though before you find a company exposing Opust 4.6 to customers and not flash 2.5 lite