“Inception Labs has launched the first production-ready Diffusion LLMs. Mercury Coder Mini achieves >1,000 output tokens/s on coding tasks while running on NVIDIA H100s – over 5x faster than competitive autoregressive LLMs on similar hardware Inception’s Diffusion LLMs (“dLLMs”) https://x.com/ArtificialAnlys/status/1894932634322772372
Nvidia’s earnings to be a test of AI chip demand as DeepSeek sows spending doubts | Reuters https://www.reuters.com/technology/nvidias-chip-demand-faces-scrutiny-deepseek-stirs-doubts-ai-spending-2025-02-24/
The IRS Is Buying an AI Supercomputer From Nvidia https://archive.md/oejjm
“Chat, @deepseek_ai just dropped DeepGEMM 🔥 > a lightweight CUDA-based library for efficient FP8 GEMMs on NVIDIA Hopper tensor cores > featuring JIT compilation, CUDA-core two-level accumulation > ~300-line core kernel > outperforming expert-tuned libraries like CUTLASS 3.6 in https://x.com/reach_vb/status/1894626368702304617
“Arc Institute and Nvidia dropped the largest AI for biology—in another open-source win! Trained on 9T+ DNA blocks from 128K species (entire tree of life), Evo 2 achieved 90% accuracy in predicting cancer-related gene mutations It can even design genomes https://x.com/rowancheung/status/1892507651000480049
Google Cloud launches first Blackwell AI GPU-powered instances — 72-way GB200 with 72 B200 GPUs and 36 Grace CPUs | Tom’s Hardware https://www.tomshardware.com/tech-industry/artificial-intelligence/google-cloud-launches-first-blackwell-ai-gpu-powered-instances-72-way-gb200-with-72-b200-gpus-and-36-grace-cpus
“If you could only learn one thing that will be relevant for the next 10-20 years, focus on learning how to deal with data. The future is not about faster hardware, smarter algorithms, or better ideas. The future is about DATA, and those who know how to deal with it will stay https://x.com/svpino/status/1895107722460438553
“Sorry, sorry nvidia, this is just funny https://x.com/teortaxesTex/status/1894234559530627240
“🚨Last week’s top AI/ML research papers: – Native Sparse Attention – Idiosyncrasies in Large Language Models – SWE-Lancer – Roadmap to fault tolerant quantum computation using topological qubit arrays – Qwen2.5-VL Technical Report – Scaling Test-Time Compute Without Verification https://x.com/TheAITimeline/status/1894067841302675802
“xAI has launched Grok-3, its new flagship model. Grok-3 sets a new record in compute scale: we estimate that it was likely trained using 4e26 to 5e26 floating point operations (FLOP). Significantly, Grok-3 is the first released model known to be trained on over 1e26 FLOP.” / X https://x.com/EpochAIResearch/status/1892671695535677745
“OpenAI plans to shift compute needs from Microsoft to SoftBank. 8 gigawatts of compute by 2030 and model costs will pass $320B https://x.com/rohanpaul_ai/status/1893107242750353531
“One-step Diffusion Models with f -Divergence Distribution Matching New work by NVIDIA; Generalize variational score distillation to the family of f-divergences to learn a single-step student generator model. When using Jensen-Shannon divergence, f -distill achieves current https://x.com/iScienceLuvr/status/1893858785426587730
“# The *actual* LLM scaling law. Adding more compute and data to LLMs makes them: – Linearly more expensive, and – Logarithmically more useful. Therefore: – Scaling becomes less useful the more you do it (once you reach a point where cost is non-trivial).” / X https://x.com/jeremyphoward/status/1895237652137509066
“GPT-4.5 is big and compute-intensive—it’s not a replacement for GPT-4o. A typical query costs on average $68 / 1M tokens, with cache discounts. We’re evaluating its long-term API availability as we balance it with building future models. If GPT-4.5 plays a unique role for your https://x.com/OpenAIDevs/status/1895220435823808687
“Huawei CEO Ren Zhengfei “Our previous worries about lack of core technology have eased, I firmly believe that a greater China will rise faster” “China’s advanced semiconductor manufacturing process can produce chips equivalent to 5.5nm, but TSMC and Samsung have already achieved” / X https://x.com/dylan522p/status/1893904701550453090
“Arc Institute and Nvidia dropped Evo 2, the largest publicly available AI model for biology to date It’s trained on 9T+ DNA blocks from 128K species and achieved a 90% accuracy in predicting cancer-related gene mutations Another open-source win https://x.com/adcock_brett/status/1893708370709741788
“In today’s competitive product landscape, scientific understanding of models often lags behind speed of model deployment. If the goal is to train a deployable model (especially when bottlenecked by compute), it totally makes sense to make several changes at a time without” / X https://x.com/_jasonwei/status/1894821797000028357
“NVIDIA introduced CorrDiff for Earth-2, delivering kilometer-scale weather forecasts at 10,000x higher energy efficiency than traditional high-resolution modeling. ☁️ Generative Model CorrDiff uses diffusion-based AI to upscale coarse 25 km grids to 2 km resolution. It
https://x.com/rohanpaul_ai/status/1894894037582389763
Report: OpenAI plans to shift compute needs from Microsoft to SoftBank | TechCrunch https://techcrunch.com/2025/02/21/report-openai-plans-to-shift-compute-needs-from-microsoft-to-softbank/
Vertex AI Pricing | Generative AI | Google Cloud https://cloud.google.com/vertex-ai/generative-ai/pricing
Apple Says It Will Add 20,000 Jobs, Spend $500 Billion, Produce AI Servers in US – Bloomberg https://www.bloomberg.com/news/articles/2025-02-24/apple-says-it-will-add-20-000-jobs-spend-500-billion-produce-ai-servers-in-us?embedded-checkout=true
Microsoft Dropped Some AI Data Center Leases, TD Cowen Says – Bloomberg https://www.bloomberg.com/news/articles/2025-02-24/microsoft-cancels-leases-for-ai-data-centers-analyst-says?embedded-checkout=true
“Data centers could consume 12% of global energy by 2026. 🎙️ Fascinating chat with @SashaLuccioni on CNN’s Terms of Service podcast with @claresduffy. She breaks down AI’s environmental footprint—from data centers using country-sized energy to water cooling challenges. Important https://x.com/fdaudens/status/1894852549586588022
