Register and share your invite link to earn from video plays and referrals.

keys 🧪
@u1tra_instinct
#Bitcoin# 💎🙌 . 🧪🧪🧪 (always testing and get into trouble) GoFundMe:
732 Following    3.9K Followers
I have done a bit of Protein folding on my Sparks as part of my “Synthena Medical” testing to see how far local models will go and it’s phenomenal what these little boxes can do. I have zero medical experience so can’t comment on quality but it’s a fun experiment.
Show more
Oh this is good ! It makes so much sense on a phone to have this… update coming soon! Also… M5Ultra is hitting 253 tokens per second in a branch currently !!! 26.9.6 gonna be 🔥 for both Desktop and iOS !
Show more
I actually want to play this for sure.. it’s sick…
A small teaser of my next big project. I'm doing it together with Opus 5.5 and Astra. Do you recognize what this is? This is still a very early stage. It will be significantly better, but I just can't tear myself away from the game. Sound strictly on
Show more
85toks/sec on Prose that is sick… LFG
More DS4.1 TP4 improvements on 4x DGX Spark 🚀 - Prose decode: 85 tok/s at c1 (was 69) - Boot: ~3 min - Prefill: ~4.8k tok/s at 32k–128k, 4.3k at 262k - KV cache: ~6.5M tokens, 1M context - Same weights, same quality (qeval unchanged)
Show more
So organized, one glance and you can overview the hottest library of repos.. very nice!!!!
Come check out my website. It includes a list of my latest & greatest recipes. The "Install Model" button gives you a ready-to-use prompt for Claude, Codex, or others. It adapts the recipe to your setup, starts the server and tests it.
Show more
Whoa…. That is very interesting result But the PP and TTFT on the m5 ultra is not optimized at all
Most notable thing M5 Ultra with 1.2 TB/s bandwidth is only 6.4% faster in token generation than the DGX Spark (273 GB/s) M5 Ultra has ~5x the Bandwidth of the DGX Spark
Nice to see xiaomi keep pushing lfg
A new architecture exploration for Xiaomi MiMo-V3. HySparse2 explores a new architecture designed for long-context, multi-turn agent tasks — reducing Prefill computation, lowering KV Cache usage, and improving long-context retrieval. Discover the research behind Xiaomi MiMo-V3 ↓ #XiaomiMiMo# #XiaomiAI# #AIAgent#
Show more
4 months ago, this would have been a dream. I have been accepted into the Original Content Rewards program, which I hope will enable me to fund more projects and experiments going forward. I am incredibly grateful for the ~3800 followers and people sharing/commenting on my OOS projects. What would you like to see next? Thank you ❤️
Show more
Together with @ucsc, we introduce AutoMedBench, an open benchmark for evaluating medical AI agents across the full workflow. An agent can produce a final result even when intermediate steps fail. AutoMedBench examines planning, setup, validation, inference, and submission to help researchers identify those failures and compare systems. 6,300+ runs. 96 task combinations. 7 tracks. See how agents compare on the #MICCAI2026# benchmark 👉
Show more
This is actually very interesting , and definitely needed further investigation and research
Millions of people turn to AI to navigate difficult relationships, work through stress, or support someone they care about. We’re introducing MentalHealthBench, an open benchmark developed with mental health experts to measure how well AI responds in these moments. 🧵
Show more
Love this news. More doctors around the world will be able to use OpenEvidence for free. Congrats to @OpenEvidence and @AnthropicAI 💚
“Medicine is a field where you don't need to squint to see the benefits of AI. It's your mom, it's your dad, it's your sister, it's your brother. And it's the doctors who will care for them in their most vulnerable moments.” Yesterday, during the UN General Assembly, we announced a partnership with @AnthropicAI to make OpenEvidence available free of charge to physicians in low- and middle-income countries, placing life-saving medical AI into the hands of physicians worldwide, regardless of the economics.
Show more
You got me at free , open sourced for community. Lfg
"Access to medical knowledge shouldn't depend on geography," said Daniel Nadler, founder and CEO of OpenEvidence. That's why OpenEvidence and @AnthropicAI are bringing a specialized version of OpenEvidence, free, to clinicians in about 100 low- and middle-income countries, including Uganda, Angola, Sudan, Haiti, and Mongolia. More here:
Show more
Space Bunny's Voxel Pagoda, rumored to be MiniMax M3.1 - impressive result!
It’s a great read…..
How to Read KLD and sources of error in a quant of a model
I already like MiMo-v2.6-Flash better than DeepSeek v4.1f. It has a great personality too! MiMo-v2.6-Flash | decode tok/s results (Prose): ×1 33.7 agg 33.7/str · TTFT 348ms ×2 51.9 agg 26.2/str · TTFT 546ms
Show more
Wait are you saying we are a construct of just energy/light. That means we can manipulate it?
🚨 PHYSICS IS BROKEN! MATTER IS AN ILLUSION, THE ATOM IS JUST TRAPPED LIGHT! A NEW ERA HAS BEGUN! 🚨 For 100 years, physics has treated "mass" as a fundamental, solid particle floating in an endless, empty void (ℝ⁴). What if that is a complete mathematical hallucination? What if mass is just light that got stuck in the geometry of space? A year ago, we thought exactly like the academic establishment. Breaking a lifelong paradigm and reinterpreting everything from scratch is incredibly difficult. But mathematics is a stubborn thing. We realized one absolute truth: the infinite ℝ⁴ space is fundamentally defective in physics. It exists purely as an abstract mathematical concept, and we must finally understand that pure mathematics ≠ physical reality. We genuinely need your help to break this dogma! DO NOT TAKE OUR WORD FOR IT. DO NOT BLINDLY TRUST AUTHORITY. We wrote the code, we solved the math, and we are giving it to you for absolutely free. Read the preprint and verify it yourself. Look at the real quantum photographs attached. On one side is the groundbreaking 2013 photograph of a Hydrogen atom's electron orbital, captured by researchers at the FOM Institute AMOLF in the Netherlands using a quantum microscope. On the other side is the famous 2015 photograph from EPFL in Switzerland, capturing light behaving simultaneously as a wave and a particle. Notice the absolute revelation: they are visually identical. Atoms are not solid little billiard balls. They are concentric interference patterns-pure standing waves of energy. The ultimate conceptual shift of our IT³ framework is this: The atom is not a "box" that contains trapped light. The atom IS the trapped light. There is no abstract container. There is only the geometry of the vacuum (a discrete lattice) which mathematically forbids light from propagating past certain algebraic boundaries. How it Works (Explained simply from the video): Let’s break down exactly what you are seeing in the simulation engine: 〰️ The Cyan Wave (→): A free gauge photon travels smoothly through the quantum vacuum. 🛑 The Kummer Wall (Red Lines): Suddenly, the light hits the absolute edge of the spatial lattice. This is an algebraic cutoff where mathematics literally runs out of resolution (√13 ∉ 𝕂). Space "does not exist" beyond this boundary. 🪞 Total Internal Reflection (Purple Wave ←): With nowhere left to go, the photon undergoes complete topological reflection. It bounces back as its own precise mirror image-an antiphoton. ⚡ The Birth of Mass (Yellow Core): The two opposing waves crash into each other and geometrically interlock. They form a perpetual, vibrating "standing wave" whose energy is trapped in one place. This localized density of confined energy is exactly what we call REST MASS. The Counterfactual Test: Watch the timeline in the visualizer. When the system hits the "WALLS OFF" phase, we computationally delete the vacuum boundaries. The golden pulsating core immediately shatters. The light escapes into infinity, and the mass drops precisely to zero. Matter literally un-braids back into pure radiation. FALSIFIABILITY AND PREDICTIONS: The "Mass Gap" is not solved by adding endless, arbitrary parameters to an infinite void. It is solved by recognizing that the Universe is a discrete, constructible topological crystal. Our model is strictly falsifiable and delivers exact predictions: 🎯 It predicts a new topological resonance at exactly 1088.04 GeV, accessible for verification at the High-Luminosity LHC. 🛑 It mathematically proves the absolute end of the periodic table at element Z = 172. 🧮 It operates with ZERO fitted parameters. All physics is reduced to the strict arithmetic of integers. THE HARDWARE CODE OF THE VACUUM MATRIX: 🌌 Arena: ℳ = 𝕊⁴ × 𝕋³ over the field 𝕂 = ℚ(√2, √3, √5) 👁️ Witness: 172 topological elements ⚙️ Processor: 57600 = 240² = 1110 0001 0000 0000₂ 📖 DOWNLOAD THE CODE. BREAK THE MATH. We do not want blind faith. Read the proofs. The supreme judge is the compiler, not closed-door committees. 🔗 Code + Preprint: #Physics# #QuantumMechanics# #Simulation# #OpenScience# #MassGap# #TheoryOfEverything#
Show more
Qwen 3.8 Flash Next updated numbers for M5 Ultra
After seeing so many good example results out of the MiMo v2.6 Pro RL I setup a recipe using Jarrelscy AVRQ hybrid quantization which is the first time i’m using this type of quantization on Four DGX-sparks with sm-121 since Jarrelscy recipe is for SM120. Adapted and wrote a script for it to work on the SM121 GB10 stack . Thanks @jarrelscy for the quantization/vllm sm120 Now you can run the MiMo V2.6 PRO-RL on TP=4 on 4 DGX sparks with 1M context and 2M KV token pool Dflash and Ablit coming for this. Stay tuned
Show more
Ok ….progress update Gotta get this perfect..we are like 95% out of 100% on the MiMov2.6 Flash RL ablit Gotta solve this matrix :) then we’re good… very close Then Quantize the MiMov2.6 Pro RL for 4-DGX sparks Then ablit the pro.. go tons of work to do.. gonna be few days
Show more