Image created with gemini-3.1-flash-image-preview with claude-sonnet-4-5. Image prompt: 1980s NORAD war room with large CRT monitor displaying glowing wireframe silicon wafer diagram in amber and blue with red warning zones, dark silhouetted operator in foreground, large bold red retro sans-serif text reading CHIPS overlaid on screen, cinematic lighting, high contrast, foreboding atmosphere
This is a huge step backwards for OpenAI: OpenAI’s ambitious $500 billion Stargate data center venture with Oracle and SoftBank stalled after internal disagreements and leadership gaps. After missing its 10 GW capacity target for 2025 and raising its projected compute spend
https://x.com/kimmonismus/status/2025851041242087901
An interactive world model developed by NVIDIA in collaboration with academic partners. – DreamDojo turns egocentric human video data into physical intelligence. – Human data is more scalable than robotics data but lacks action labels. – To solve this, a dedicated action model
https://x.com/TheHumanoidHub/status/2025368793321799909
Announcing DreamDojo: our open-source, interactive world model that takes robot motor controls and generates the future in pixels. No engine, no meshes, no hand-authored dynamics. It’s Simulation 2.0. Time for robotics to take the bitter lesson pill. Real-world robot learning is
https://x.com/DrJimFan/status/2024895359236051274
Meta and AMD Partner for Longterm AI Infrastructure Agreement https://about.fb.com/news/2026/02/meta-amd-partner-longterm-ai-infrastructure-agreement/
NVIDIA has open-sourced SONIC, a humanoid behavior foundation model that gives robots a core set of motor skills learned from large-scale human motion data. https://x.com/TheHumanoidHub/status/2024935738362765677
SONIC: Supersizing Motion Tracking for Natural Humanoid Whole-Body Control https://nvlabs.github.io/GEAR-SONIC/
We have seen rapid progress in humanoid control — specialist robots can reliably generate agile, acrobatic, but preset motions. Our singular focus this year: putting generalist humanoids to do real work. To progress toward this goal, we developed SONIC ( https://x.com/yukez/status/2024639427788857707
We trained a humanoid with 22-DoF dexterous hands to assemble model cars, operate syringes, sort poker cards, fold/roll shirts, all learned primarily from 20,000+ hours of egocentric human video with no robot in the loop. Humans are the most scalable embodiment on the planet. We
https://x.com/DrJimFan/status/2026709304984875202
What can half of GPT-1 do? We trained a 42M transformer called SONIC to control the body of a humanoid robot. It takes a remarkable amount of subconscious processing for us humans to squat, turn, crawl, sprint. SONIC captures this “”System 1″” – the fast, reactive whole-body
https://x.com/DrJimFan/status/2026350142652383587
At 17,000 tokens per second, AI output outruns human perception! @taalas_inc literally baked the 8B Llama model into the chip to achieve this speed. Here is an explanation of how LLM-as-a-chip works and the real trade-offs of “”baked in”” intelligence
https://x.com/TheTuringPost/status/2026805719463292973
day ruined because orin cpu doesn’t support sve instructions (it’s 6am)
https://x.com/vikhyatk/status/2027017720466985437
First, @wschenk told me: “”Did you see that speed?! It’s mindblowing. And it’s all over X now.”” So I had to understand how @taalas_inc – and their concept ‘model-as-hardware’ works. Are GPUs in danger? As usual, history (this time about semiconductors) helps us see what this
https://x.com/Kseniase_/status/2026801100725137545
Fourth drop on @fal performance blog. Our resident GPU communications expert Ismayil is revealing one of the many interesting things we do on comms-computation overlap in our inference engine. Async Ulysses. https://x.com/isidentical/status/2026000340873777419
Meta announced a $100 billion mega-deal with AMD for a 6GW of planned data center capacity. One day before NVIDIA’s fourth-quarter results for FY26.
https://x.com/kimmonismus/status/2026279386681356704
Today, we release our largest LFM2 model: LFM2-24B-A2B 🐘 > 24B total parameters > 2.3B active per token > Built on our hybrid, hardware-aware LFM2 architecture It combines LFM2’s fast, memory-efficient design with a Mixture of Experts setup, so only 2.3B parameters activate
https://x.com/liquidai/status/2026301771539202269
We’re building an LLM chip that delivers much higher throughput than any other chip while also achieving the lowest latency. We call it the MatX One. The MatX One chip is based on a splittable systolic array, which has the energy and area efficiency that large systolic arrays
https://x.com/reinerpope/status/2026351870852358492
We’re building an LLM chip that delivers much higher throughput than any other chip while also achieving the lowest latency. We call it the MatX One. The MatX One chip is based on a splittable systolic array, which has the energy and area efficiency that large systolic arrays
https://x.com/reinerpope/status/2026351870852358492?s=20
with responses api, they started storing the state in upstream (prev response id). i think this was a storage thing. for websocket change, they probably retain state in gpu vram. so the infra buildup might be the cause chat completions api just didnt have the design for infra
https://x.com/dejavucoder/status/2026223111021220265
With the coming tsunami of demand for tokens, there are significant opportunities to orchestrate the underlying memory+compute *just right* for LLMs. The fundamental and non-obvious constraint is that due to the chip fabrication process, you get two completely distinct pools of
https://x.com/karpathy/status/2026452488434651264
Meta 🤝 AMD Today we’re announcing a multi-year agreement with @AMD to integrate their latest Instinct GPUs into our global infrastructure. With approximately 6GW of planned data center capacity dedicated to this deployment, we’re scaling our compute capacity to accelerate the
https://x.com/AIatMeta/status/2026266818789454057
Sam Altman says the GPU capacity the world needs for digital work and home humanoid robots will be the most expensive and complex infrastructure project the world has ever taken on. He also says the idea of an orbital data center is ridiculous given the current landscape.
https://x.com/TheHumanoidHub/status/2024900239283683819
Diffusion just rewired the AI speed game. Inception’s Mercury 2 hits 1,000 tokens/second, that’s 10x faster than Claude 4.5 Haiku and GPT-5 Mini, not through custom chips, but through a fundamentally different architecture borrowed from image generators like Midjourney.
https://x.com/kimmonismus/status/2026662718321897974
Energy is becoming a huge domestic political problem: Donald Trump is bringing Amazon, Google, Meta, Microsoft, xAI, Oracle, and OpenAI to the White House to sign a “Rate Payer Protection Pledge,” committing them to generate or purchase their own electricity for new AI data
https://x.com/kimmonismus/status/2026720759163298282
Stargate is the umbrella brand for our compute strategy. It’s about mobilizing the full ecosystem to deliver a step-change in global AI compute — and over the past few quarters that vision has become reality. Since announcing Stargate in January 2025, OpenAI has expanded from a
https://x.com/sk7037/status/2026067771394838629
NVIDIA just released a Blackwell-optimized Qwen3.5 MoE on Hugging Face 397B parameters quantized to NVFP4 for 2x faster inference with SGLang.
https://x.com/HuggingPapers/status/2025825405836648849
Leave a Reply