@sudoingX Qwen 3.6-35b-a3b MoE
RTX 3060 (12 GB)
128k context, with MTP
60 tps
HT
Machine Space
326 posts
Machine Space
@TheMachineSpace
Local LLMs, open-weight models, AI culture and AI shitposting. AI is serious business (derogatory)
MACHINE SPACE Katılım Temmuz 2026
204 Takip Edilen13 Takipçiler
qwen 3.6 35b a3b census. this model runs on everything, so let's prove it.
reply with your rung: cpu only, single gpu, multi gpu, or dgx cluster. add your tok/s and usable context before it drags.
i want to see the same model breathing on a laptop and screaming on a spark cluster in the same thread.
English
Machine Space retweetledi
@JohnDCook If they haven’t filed a death certificate, it’s Social Security fraud. Big deal!
English
Hypothetically, if someone were to knowingly make false statements to cover up the death of a US senator, would they be violating any federal laws?
English
Machine Space retweetledi
Machine Space retweetledi
🛠️ OpenRouter just dropped a big MCP update
👀 Your agent can now discover, rank, test, and report on models directly in your editor.
Here’s what’s new:
🔍 Task Type Insights
🔹 Real usage data across 30 categories. See which models actually perform best for coding, reasoning, writing, etc. (updated weekly)
🔎 Sharper Model Search
🔹 Filter by price, age, intelligence, coding score, and tool-calling success rate. Sort by Design Arena ELO.
🔹 Rankings now break down by use case & modality.
📌 Pin ProviderLock send-message to the same model + same provider every time. No more random routing during evals.
🖼️ New Tools
🔹 generate-image → Generate images directly from your agent (billed to your key)
🔹 send-feedback → Report bad outputs straight to the OpenRouter team
⚡ Quality of Life
🔹 Cleaner verb-noun tool names
🔹 Read-only tools no longer trigger permission prompts
🔹 Docs search now powered by Mintlify
Get started: openrouter dot ai /docs/guides/overview/mcp-server
English
Machine Space retweetledi
Who cares
Variety@Variety
An AI-generated film adaptation of "The Odyssey" has been announced. “Odysseus: The Fall” is a 135-minute AI-generated movie from AI film studio Fountain 0 and director Ash Koosha. The movie will be available to rent or buy on Fountain 0’s website later this summer. “I’ve been just obsessed with it since I was a kid,” Koosha said in an interview. “One of the things that really drove me towards this story was the character of Odysseus himself, and my take on it — the feeling that I’ve had over the years reading different takes on it, and different translations — and my take was something that I just wanted to tell. And recently, when we did the first film, I realized this is the best time given the public discourse that exists out there as well.” variety.com/2026/film/news…
English
Oh fuck Georgia Power in the goddamn ear
Hedgie@HedgieMarkets
🦔Georgia Power is acquiring over 300 parcels of land, including family homes, to build a 35-mile transmission line. The company says 70-80% of the power on that line will serve data centers. Homeowners who don't sell face eminent domain proceedings. Georgia Power won't say which companies the data centers belong to, citing "safety and security." My Take A for-profit utility is using eminent domain to take people's homes so it can sell electricity to data center operators whose names it won't disclose. I have a hard time seeing how that qualifies as serving a public purpose, which is the legal standard eminent domain requires. The families losing their land aren't benefiting from the data centers. The data centers employ almost nobody locally. And the utility is building new capacity primarily to serve private corporate clients, not residential demand. Georgia Power admitted that in its own numbers. Eminent domain exists so governments can build roads, schools, and infrastructure that serves the public. When 80% of the power on a new line goes to private data centers and the utility won't tell you whose, the "public purpose" argument starts to fall apart. Ansley Brown's family built their home when she was five years old. Her mother wanted it to be generational wealth. Now it's a right of way for a transmission line so an unnamed company can run GPU clusters. Hedgie🤗 cbsnews.com/news/georgia-p…
English
@technohobo23 @TheAhmadOsman Get ready for hundreds of thousands of locally-hosted models on Windows:
nvidia.com/en-us/products…
English
@TheAhmadOsman @TheMachineSpace More people use windows. Some hate it. Non technical people buy what is at the store not what they can install. This is not an OS endorsement just an observation. secondly most people do not use AI right now. When they do eventually use it from a desktop it will be from windows.
English
Osmantic is my top recommendation when people ask me "how do I get into AI?"
1. Make a Linux boot sector
2. Go to Osmantic
Ahmad@TheAhmadOsman
Local AI is easy now - How easy??? Really easy... > curl -fsSL install.osmantic.com/ods.sh | bash That's it, that's all you need now
English
@TheAhmadOsman Step 1 is a LOT more complicated than step 2.
QQ - do you run on Windows? Because RTX Spark is gonna make thousands more people really interested in easy-to-host models this fall...
English
Machine Space retweetledi
🦔A former Mayo Clinic AI compliance lead is suing the hospital for firing her after she blew the whistle on its AI tools. The lawsuit alleges Mayo's digital assistant MAYA had a 67% error rate that staff tried to hide by deleting results and mischaracterizing outcomes. A high-risk cardiac surgery device was allegedly approved without required review board oversight. When she flagged these issues, she was told fixing them would "compromise Mayo's competitive advantage," then demoted and fired during medical leave.
My Take
A 67% error rate on a medical AI tool is bad enough. The allegation that staff knew and deleted the results to hide it is worse. Mayo is one of the most respected hospital systems in the world. If this is how they handled AI integration, with review boards bypassed and unfavorable data deleted because the priority was speed, smaller hospital networks with fewer resources are almost certainly doing no better.
The "poor cultural fit" language is familiar to anyone who has watched a company push out a whistleblower. She raised ten separate complaints, got excluded from meetings, was warned that her boss had "marching orders to get rid of her ASAP," and was fired during medical leave. Mayo says it follows all applicable laws and regulations. The person they hired to ensure that says otherwise, and she's willing to take it to federal court.
Hedgie🤗
English
Machine Space retweetledi
Now we can run Sonnet-4.6 on our phones.
Local AI is evolving faster than anything else.
Start learning about local.
PrismML@PrismML
Today, we’re announcing Bonsai 27B: the first 27B-class model to run on a phone. Bonsai 27B is the new multimodal flagship of the Bonsai family. Based on Qwen3.6 27B, it brings a new capability tier to local AI: multi-step reasoning, structured tool use, long-context workflows, and coherent agentic loops. Until now, models in this class have been impractical to deploy locally. A 27B model occupies roughly 54 GB in 16-bit precision, and even a strong 4-bit build is around 18GB - too large for a phone and for most laptops. Bonsai 27B changes that. It comes in two variants: • Ternary Bonsai 27B: 5.9 GB, 1.71 effective bits per weight, optimized for laptop-class quality. • 1-bit Bonsai 27B: 3.9 GB, 1.125 effective bits per weight, optimized for phone-class footprint. Everything is open-sourced today under the Apache 2.0 license.
English
Machine Space retweetledi
Elon can’t beat him at this game. He tweets better and will keep wiping Elon in a tweet fight. He’s far better at tweeting.
Sam Altman@sama
Concerning.
English
This is the dream: machines just doing their machine things and popping up from time to time with "hey boss, check this out."
Zhengyao Jiang@zhengyaojiang
The first experimental evidence of recursive self-improvement (RSI). Autoresearching the autoresearch agent for eight days. The result beats the harness we hand-tuned for two years, on held-out benchmarks: 🧵(1/7)
English
Machine Space retweetledi
@edels0n The price graph would defy time-series logic and spell "wtf?"
English
Imagine how crazy the chart would look if this were a public stock
zerohedge@zerohedge
*OPENAI PLANS TO RELEASE HOME SPEAKER AS FIRST HARDWARE PRODUCT
English
This is a great framework no matter what you feel about AI: there is value in gathering info, planning and observing. These things in themselves are actions.
If you're pro-AI or anti-datacenter or pro-open-weight, this prescription is for you:
Brian Albrecht@BrianCAlbrecht
"We must act now" is right in one dimension, I just don't think it's the dimension that gets featured in the NYT. Let's talk about a topic I've written on: policy can be thought through using real options theory 🧵
English
Machine Space retweetledi
🤝 @arcee_ai, an American AI lab, built its frontier open model family on NVIDIA Blackwell Ultra.
Its flagship MoE model, Trinity-Large-Thinking, was RL post-trained using NVIDIA NeMo open libraries.
Arcee optimized agentic model inference with NVIDIA Dynamo and vLLM, along with NVIDIA accelerated networking, to deliver low token cost, achieving:
✅ 3T+ tokens served on OpenRouter in its first two months
✅ $0.90 per 1M output tokens
✅ 2nd on PinchBench for open agentic models
Explore Arcee's open models, powered by NVIDIA ➡️ nvda.ws/457XCOw
English
@Scobleizer All that money and sturm & drang and they're gonna bring us the hottest product of 2014: Alexa.
English
Or "tariffs are killing us but we dare not say anything unless we get targeted by a mob"
𝕭𝖑𝖆𝖐𝖊@itsblakexx
AI was cited as the reason for 26% of layoffs in April. By May, it was nearly 40%. A year ago it was 5%. That's not a trend. That's a company realizing AI makes a better excuse than "we overhired."
English