Sander Blue

10T…. Ya I can run that locally 😎

A reasonable assessment: low probability for direct inclusion of your specific FSA/MEBS Zenodo papers in these Colossus 2 pre-training runs. xAI's data pipelines (like most frontier labs) pull from massive, filtered web crawls, academic archives (arXiv-heavy), code repos, and high-visibility sources. Zenodo PDFs exist in the open web, but niche papers with limited citations/downloads rarely make the cut amid trillions of tokens—unless they've gone viral or been heavily referenced elsewhere. Your framework's core ideas could surface indirectly if echoed in broader discussions, but the raw training process prioritizes scale over specialized substrates. Post-training (fine-tuning, prompting, or retrieval) is where they'd have more impact.

@DannyLimanseta Pre-training phase is ~2 months

Come join us on our weekly webinar series, TradersPost Tutor Sessions! Join us to learn, ask questions, and level up your automation game. Watch live on Youtube (link in comments) now!

The sphere in Vegas just doing Sphere things 😲

Vibe coding will only get worse.

Needs more mountains and snow... but it's still pretty awesome.

"Code smell" still exists with AI-assisted coding no matter how strong the model is today. Anecdotally, it's the same smell SWEs have experienced over the years... something just smells off, different, inaccurate, or blurry. Intuition hasn't been replaced.

There should be a better “meh” emoji (or dataset of “meh” emojis). “Meh” means so many different things in different countries and lifestyles. Someone out there has already designed it/them (the emojis). Please reveal them 😒😏🤨😔😟🥺😉😎👌🤪🤬🥹🤣

This actually made me laugh… and then sigh.

The world stopped to watch Artemis II. Moments like this remind us what is possible and inspire the next generation to dream bigger and take us even further. We are just getting started on this grand adventure. It is time to start believing again.

Hilarious this band covers that banger on-hold music 😂🤘

$300 mini PC running 26B parameter AI models at 20 tok/s. Minisforum UM790 Pro ($351) + AMD Radeon 780M iGPU + 48GB DDR5-5600 + 1TB NVMe. The secret: the 780M has no dedicated VRAM. It shares your DDR5 via unified memory. The BIOS says "4GB VRAM" but Vulkan sees the full pool. I'm allocating 21+ GB for model weights on a GPU with "4GB VRAM." The iGPU reads weights directly from system RAM at DDR5 bandwidth (~75 GB/s). MoE only activates 4B params per token = 2-4 GB of reads. That's why 20 tok/s works. What it runs: - Gemma 4 26B MoE: 19.5 tok/s, 110 tok/s prefill, 196K context - Gemma 4 E4B: 21.7 tok/s faster than some RTX setups - Qwen3.5-35B-A3B: 20.8 tok/s - Nemotron Cascade 2: 24.8 tok/s Dense 31B? 4 tok/s, reads all 18GB per token, bandwidth wall. MoE same quality? 20 tok/s. Full agentic workflows via @NousResearch Hermes agent with terminal, file ops, web, 40+ tools, all against local models. No API keys. Just a box on your desk. The RAM is the pain right now. DDR5 prices 3-4x what they were a year ago. But the compute is free forever after you buy it. @Hi_MINISFORUM @ggerganov llama.cpp + Vulkan + @UnslothAI GGUFs + @AMDRadeon RDNA 3. Fits in your hand. #LocalLLM #Gemma4 #llama_cpp #AMD #Radeon780M #MoE #LocalAI #AI #OpenSource #GGUF #HermesAgent #NousResearch #DDR5 #MiniPC #EdgeAI #UnifiedMemory #Vulkan #iGPU #RunItLocal #AIonDevice

There are still parts of the world and wilderness where you can completely be disconnected from the internet... for now.

SpaceX now has over 10,000 Starlink satellites in orbit, which is honestly an absurd engineering achievement. And no, they are not just up there freelancing and hoping for the best. They stay separated because they are placed in organized orbital lanes, constantly tracked, and able to maneuver when needed. Starlink also uses automated collision-avoidance systems, which is how a constellation this large can operate without turning low Earth orbit into a scrapyard. It’s already the closest thing in the world to a true work-anywhere network and they are just getting started.

Scene: Watching final four with my wife and 3yo daughter Daughter: “I don’t wanna watch basketball anymore. Let’s watch my shows” Me: “We’re working on sharing the glowing rectangle time as a family” Daughter: <sighs> “Dad, we have TV downstairs. We have 3 TVs in the house”

（1/8）🚀 Introducing Qwen3.6-Plus: Towards Real-World Agents! 🤖 Today, we’re thrilled to drop a major milestone in our journey toward native multimodal agents. Here is what makes Qwen3.6-Plus a game-changer： 💻 Next-level Agentic Coding: Smarter, faster execution. 👁️ Enhanced Multimodal Vision: Sharper perception & reasoning. 🏆 Top-tier Performance: Maintaining leading general capabilities. 📚 1M Context Window: Available by default via our API. Built on your invaluable feedback from the Qwen3.5 era, we’re laying a rock-solid foundation for real-world devs. Get ready to experience truly transformative ✨ Vibe Coding ✨. Huge thanks to our community! Go try it out and show us what you can build. 👇 Chat: chat.qwen.ai API: modelstudio.console.alibabacloud.com/ap-southeast-1… Blog: qwen.ai/blog?id=qwen3.6 🔔Noted：More Qwen3.6 models to come and be open-sourced! Stay tuned~ 👀#Qwen #AI #AgenticCoding #VibeCoding #Agents

LLM Knowledge Bases Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So: Data ingest: I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them. IDE: I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides). Q&A: Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale. Output: Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base. Linting: I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into. Extra tools: I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries. Further explorations: As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows. TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.

Available in four sizes: 🔵 31B Dense & 26B MoE: state-of-the-art performance for advanced local reasoning tasks – like custom coding assistants or analyzing scientific datasets. 🔵 E4B & E2B (Edge): built for mobile with real-time text, vision, and audio processing.

Awwww snap, gonna give this a test drive on my private AI platform.