rookie numbers they need 5M now and 15M+ by 2028 they won't have it mainly due to memory
rookie numbersthey need 5M now and 15M+ by 2028they won't have it mainly due to memoryP Equity Research 📰: J.P. Morgan has domestic AI chip shipments…
Latest
Anthropic’s landmark $1.5B copyright settlement is approved
The final approval settles one case, but it doesn't resolve the broader issue of using copyrighted works to train AI models.
RT Whole Mars Catalog: A Cybercab with no steering wheel, a massive display, and high speed Starlink is going to be the best place to work or watch a …
RT Whole Mars CatalogA Cybercab with no steering wheel, a massive display, and high speed Starlink is going to be the best place to work or…
Benchmarked every spec-decode method on Qwen3.6-27B across vLLM and SGLang (single RTX PRO 6000 Max-Q)
https://preview.redd.it/wluwwp6s4heh1.png?width=1248&format=png&auto=webp&s=6e95d963645c5a0ef750bf81324a6d4dcbc0389e Spent the last few days measuring speculative decoding on Qwen3.6-27B (dense, NVFP4) on one RTX PRO 6000 Max-Q, comparing vLLM and SGLang across MTP, DFlash,…
RT Vaish Shrivastava: Great blog from @cwolferesearch on Agentic World Models highlighting our work on ECHO
RT Vaish ShrivastavaGreat blog from @cwolferesearch on Agentic World Models highlighting our work on ECHOCameron R. Wolfe, Ph.D.: I just published my blog on using world…
RT 500 Global: Sakana AI is doubling down on a thesis we believe will play an increasingly important role in AI development: the future of AI won't be…
RT 500 GlobalSakana AI is doubling down on a thesis we believe will play an increasingly important role in AI development: the future of AI won't…
RT 500 Global: Sakana AI is doubling down on a thesis we believe will play an increasingly important role in AI development: the future of AI won't be…
RT 500 GlobalSakana AI is doubling down on a thesis we believe will play an increasingly important role in AI development: the future of AI won't…
Scaling Agentic RL: High-Throughput Agentic Training with Tunix
Tunix is Google’s new JAX-native post-training library designed to eliminate TPU idling bottlenecks when training multi-turn, tool-using LLM reasoning agents. It maximizes hardware throughput by combining…
Anthropic is donating another $20 million to Public First Action
Anthropic is donating another $20 million to Public First Action
David Vélez and Robin Vince join the boards of the OpenAI Foundation and OpenAI Group PBC
David Vélez and Robin Vince join the boards of the OpenAI Foundation and OpenAI Group PBC, bringing global leadership in finance, technology, and governance.
How Anthropic Secures Its AI-Native Software Development Lifecycle
How Anthropic Secures Its AI-Native Software Development Lifecycle
How Datadog built a “universal machine tool” for Claude Code
How Datadog built a “universal machine tool” for Claude Code
Accelerating Text-to-Video Generation with Calibrated Sparse Attention
Recent diffusion models enable high-quality video generation, but suffer from slow runtimes. The large transformer-based backbones used in these models are bottlenecked by spatiotemporal attention. In…
Environment-free Synthetic Data Generation for API-Calling Agents
Training API-calling large language model (LLM) agents demands massive amounts of high-quality trajectories. However, collecting such data at scale typically requires fully implemented environments with executable…
Query Profiling: See Where a Slow Query Spends Its Time
When a Weaviate query is slow, the first question is where the time went. Query profiling returns a per-stage, per-shard timing breakdown, making query performance issues…
Papers for Stable LatentMoE and Gated MLA?
Four technologies used by Kimi K3 to make it SOTA: KimiDeltaAttention (used in Kimi Linear, essentially a more general gated delta net) AttnRes - described in…
b10075
hexagon: add CLAMP op (#25934) Website: https://llama.app macOS/iOS: macOS Apple Silicon (arm64) macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED macOS Intel (x64) iOS XCFramework Linux: Ubuntu…
nvfp4 kv-cache on 2×5060 ti, vllm
I can't take credit for this, someone else described the approach, and I just copy/pasted it into opencode/GLM 5.2 to get it working. It appears to…
No. As of now, in the air Mind you, American victory condition is basically monopolizing the capacity to exterminate the rest of the human species, an…
No. As of now, in the airMind you, American victory condition is basically monopolizing the capacity to exterminate the rest of the human species, and Chinese…
Those in the 1000+ prefill and 100+ decode range on Qwen3.6 35B at Q4, what hardware are you running?
Trying to see what I can scrounge together bare minimum hardware requirements to get up to that rough speed. Right now I'm running an RX6600XT and…
AI Daily Brief — 20 July 2026
Chinese labs topped AI leaderboards as Washington weighs restrictions on Chinese models and OpenAI discloses long-horizon safety risks.
Google has disappeared completely from the top 15
Google hasn't shipped a model recently that is capable of competing with Sol or Fable. The previous models were pretty disappointing and unreliable, it seems the…
For a long time, Meta's meh performance in AI products (despite good research out of FAIR) was a kink in the story about Chinese AI talent (Meta reall…
For a long time, Meta's meh performance in AI products (despite good research out of FAIR) was a kink in the story about Chinese AI talent…
China can only watch as its ally is being destroyed…
China can only watch as its ally is being destroyed…Tom Bike: Another large aircraft shelter destroyed in Jordan 🇯🇴🇺🇸 US al Salti AB by accurate Iran…