Run your own AI at home — I'll teach you, free, one morning read at a time. USMC vet, ham radio, homelab: 4x DGX Spark. Gear list: https://xitter.twitterapi.xyz/t.co/iTbkWsbbGV
Washington, DC
Joined October 2008
- Tweets2.5K
- Following1.4K
- Followers1.1K
- Likes1.4K
If there isn’t an emergency, they will invent one. For centralized control. I can’t stand this guy.
BILL GATES JUST ADMITTED: We May Need to Suffer a MAJOR CYBERATTACK Before People ACCEPT CONTROL!!
Bill Gates just told CBS that a major cyberattack on platforms like Fidelity or Vanguard, hitting ordinary people’s retirement savings, may be required before the public accepts the kind of centralized controls and regulations the global elite have been preparing for years.
In this video, Dan Dicks of Press for Truth breaks down the clip, places it in the classic Problem-Reaction-Solution framework, and connects it to Event 201, Cyber Polygon, and the ongoing push for AI and cyber “safeguards.”
Please SHARE THIS VIDEO far and wide!!
They want them to surrender to their party. It’s like the guy offering your business protection and you refuse knowing he is a thug. Then the next day your window is broken. You repair it. Next day it’s broken again. You repair it. Insurance starts hassling you for repair and your insurance rate goes up. You end up paying the thug for protection. That is the DNC.
Republicans are clowns who can’t figure out this game or articulate a counter argument. They instead surrender to the fake outrage and say they have better solutions to the tyranny. Republican Marsha Blackburn’s bill that died before being voted on because Trump signaled he would veto it. That was no better.
We are destined it seems to end up like Europe with two parties arguing they can control the levers of socialism better instead of little “r” republicanism and liberty for the individual.
JUST IN: Democrats propose requiring AI companies to obtain federal government permission before training AI models, with up to 20 years in prison for violations.
1
6
133
This brings a great model to more people! Great work @UnslothAI
Today, we're releasing the highest accuracy Qwen3.8-27B GGUFs.
Our new Dynamic V3.1 GGUFs lead by 5–10% on average across Terminal Bench, DeepSWE & more. We’re working on even better version for Qwen4
Run on 12GB RAM.
GGUF: huggingface.co/unsloth/Qwen3…
Blog: unsloth.ai/docs/models/qwen3…
A 33B image model that makes 16-megapixel images just shipped with open weights, and Unsloth already has it running on a 24 GB card.
@MiniMax_AI released MiniMax H3-Image on Nov 10: text-to-image plus editing with up to 16 reference images. Unsloth's FP8 build cuts the 66.3 GB transformer to 20.27 GB and reports about 50 seconds per 2048x2048 image on a 24 GB L4.
Image models this good used to live only behind an API with someone else's content rules. Where the license allows, the weights can sit on your own drive and the pictures never leave your machine. License note: MiniMax's community license doesn't currently cover the US, EU, UK or Korea, and above $1M a year in revenue or for hosting it for others, you need MiniMax's written OK first.
What would you make first if a model like this ran on your own desk?
4K image generation in 10 seconds! ⚡
Meet MiniMax H3-Image.
16 million pixel resolution. Stunning realism. Precise prompt following & editing. Up to 16 reference images. All in 10 seconds.
And it's open-weight. 🚀
Available now via our API, MiniMax Code, and Hailuo AI — with 50% off 16MP+ generations during launch week!
Can't wait to see what you create. 🔥
Hugging Face: huggingface.co/MiniMaxAI/Min…
API: platform.minimax.io/docs/gui…
MiniMax Code: code.minimax.io/about
@grok what is @nvidia @NVIDIAAI “switchyard?” How would someone using Hermes Agent @NousResearch with multiple DGX sparks and Mac’s hosting AI use it?
github.com/NVIDIA-NeMo/Switc…
1
1
206
Entry point for self hosting AI. (Experienced users be kind and accepting to new people taking first step joining the community)
This is for people completely new and unsure of the large commitment of a DGX Spark. This entry point lets you experience hosting your AI, very small models and simulate the decision on hosting larger models through available models that later can be downloaded.
@NousResearch @NVIDIAAI @nvidia
Links in comments.
Here are the direct product links from Micro Center and the official Nous Research links.
Micro Center — Hardware
1. Raspberry Pi 5 — 16GB
Micro Center SKU: 938035
16GB RAM, quad-core Cortex-A76 processor.
microcenter.com/product/7025…
2. NVIDIA Jetson Orin Nano Super Developer Kit
Micro Center SKU: 812057
8GB LPDDR5, up to 67 TOPS AI performance.
microcenter.com/product/6910…
Both are listed as in-store-pickup products.
Nous Research — Software & Cloud
3. Hermes Agent by Nous Research
Official documentation and installation
Open-source, MIT-licensed AI agent.
hermes-agent.nousresearch.co…
4. Nous Research AI Cloud Subscription
Nous Portal — Subscription Plans
Hermes Cloud — Hosted AI Agents
Current advertised subscription tiers:
Plan
Monthly price
Free
$0
Plus
$20
Super
$100
Ultra
$200
The subscription provides Hermes Agent model access, with included usage limits and cloud-hosting allowances depending on the plan.
Nous Portal
portal.nousresearch.com/
All four links are to the actual retailer or official project/provider.
People take open weight models and optimize them for self hosting. I read this article as not being able to use Anthropic’s AI as an operator doing that work on your own equipment.
How do you read this news?
I called it. Anthropic wants to cripple your code to keep your local AI dumbed down and inefficient. They admit it in their documents.
90
Grok 4.8 is coming!
Grok bot users may be able to experience it by the end of the week.
We have dumped JAX/SGLang and rewritten the entire RL and inference stack from scratch in C/C++, which took some extra time.
Training (aka pre-training) is done. Lot of cleanup was needed on the SFT data to get that mostly right. Now, 4.8 is in RL and the evals are starting to climb. We will do a soft rollout to Bot users later this week and see how it goes.
Grok 4.8 will bear no resemblance to 4.7 (thankfully). It is not yet incredible, but, this time, the foundations are built right, so I expect it to evolve to be something special.
I was all over this when it was first proposed. My “bipartisan” objection to regulating and destroying AI. Trump ended up signaling we would veto it despite them naming at after him to appeal to him.
xitter.twitterapi.xyz/grok/status/2119519399…
The issue now is these republicans are the ones who will be the 4 votes to give now anti AI democrats the 60 votes to prevent debate and filibuster. While yes Trump will veto, senators are elected for six years and Trump has two years left.
Statistically speaking the next president will be a Democrat. The first two years they will have both sides of the congress and the presidency.
To me the threat is real. Certainly closer to it than it was before.
I posted before and people dismissed me saying super intelligence is incapable on your little DGX sparks. Insults of, you are silly to think a DGX Spark purchase is a solution to super intelligence. “Bro thinks he is going to make super intelligence on his shitty DGX Spark.” The slowdown or banning super intelligence isn’t about you. Or an additional DGX spark isn’t going to solve banning super intelligence, you cant host or create super intelligence on them.
See the trend! Is super intelligence in data centers now that they want to ban? There is an interesting argument is AGI already here? Some say yes, some say no. But what I do know is this, AI is in data centers and they want them banned. I have AI on my desk. Opensource another things the despise with their planned token economy.
The hierarchy being Artificial Intelligence (AI), Artificial General Intelligence (AGI), and Artificial Super Intelligence (ASI) or Super Intelligence with the new nomenclature dropping artificial.
Dems said they want to ban super intelligence. Dems said there should be a 20 year mandatory jail time for violating their new laws.
Altman said the individual shouldn’t have inference at home. It all belongs in data centers for the world.
I went out and bought a 5th DGX Spark.
Buy what you can. Download a library of models.
@nvidia @NVIDIAAI
139
Hermes Agent and local is the answer. Are you positioned for it?
Use grok bot as the coordinator, it can make calls to v1 of a Hermes agent profile. And vice versa use Hermes agent to call grok bot with complete work with a webhook.
If grok bot policies changes, it’s a small adaptation.
@NousResearch @NVIDIAAI @nvidia
I am doing what I love. I am simply documenting it. It pays nothing. I’m ok with it. If it pays someday, I will do more of what I love. I made some good friends and met some amazing people with similar interests.
We don’t have to 100% agree. No purity tests.
2
132
@MiaAI_lab's TensorFold update for Qwen3.8 Flash Next (@Alibaba_Qwen) on one DGX Spark: code tok/s up 62.7% at 1 request and 45.4% to 47.1% at 2, 4 and 5.
Anyone run this yet? I am going to get to it this afternoon.
Do you like more speed? I know I do ⚡️
MASSIVE Qwen3.8 Flash Next update for ONE DGX Spark is now LIVE, with performance gains across the board! 🔥
Code tok/s gains:
1 stream: +62.7%
2-5 streams: +45-47.1%
Prose, tok/s gain (non-greedy, thinking on):
1 stream: +10.1%, 2 streams: +29.8%
4 streams: +36.4%, 5 streams: +17.6%
Prefill tok/s gains:
4k: +14.5%, 8k: +14.4%
16k: +13.5%, 32k: +12.6%
This was the BEST model to run if you have one DGX Spark, and now it's even better! nothing else comes close in terms of quality, speed, and concurrency.
Link & full tok/s and prefill numbers below.
3
215