This NVIDIA remains the strongest platform for large-model inference at scale. Prefill/decode disaggregation, Blackwell-native quantization, custom kernels, and rack-scale NVLink turn GB200 into faster answers lower serving cost. Read the full paper here
LLMS
-
Performance Benchmarks: NVIDIA H200 vs GB200 for AI Workloads
By
–
The benchmarks show the gap. NVLS all-reduce latency drops from 586.1µs on H200 to 313.3µs on GB200. In MoE prefill at EP=4, combine falls from 730.1µs to 438.5µs. For decode, GB200 sustains much higher throughput at high token speeds.
-
Research on Serving Qwen3 235B Models on NVIDIA GB200 Racks
By
–
We published new research on how we serve post-trained Qwen3 235B models on NVIDIA GB200 NVL72 Blackwell racks. GB200 is a major step up over Hopper for high-throughput inference on large MoE models, not just a training platform.
-
Debating the economic ROI of major AI labs
By
–
that’s actual speculative; so far OpenAI and Anthropic have lost money and most businesses have not gotten significant RoI
-
Google Gemini Omni Video Generation Model Leaks Ahead of I/O
By
–
🤯Google vient de leaker son prochain monstre.
— VISION IA (@vision_ia) 12 mai 2026
Gemini Omni : un modèle de génération vidéo qui vient d'apparaître chez certains utilisateurs Gemini, juste avant Google I/O (19-20 mai).
Les premiers résultats sont bluffants, un utilisateur a demandé "un professeur qui écrit une… pic.twitter.com/U86extcaNFGoogle vient de leaker son prochain monstre. Gemini Omni : un modèle de génération vidéo qui vient d'apparaître chez certains utilisateurs Gemini, juste avant Google I/O (19-20 mai). Les premiers résultats sont bluffants, un utilisateur a demandé "un professeur qui écrit une
-
Evolution of Excel: From Early ML to LLMs
By
–
excel didn’t have LLMs in the 90s (it did have some ML, which I have praised, but not that far back).
-
Analyzing o3, Grok 4, and Claude Code
By
–
start with https://
open.substack.com/pub/garymarcus
/p/how-o3-and-grok-4-accidentally-vindicated?r=8tdk6&utm_medium=ios
… and my substack piece on Claude Code and then read my 2001 book The Algebraic Mind -
Claude Code and the Neurosymbolic Approach to LLM Limitations
By
–
claude code is an immense neurosymbolic effort to ward off the failure of pure LLMs. you sir are lost.
-
GPT-5.5 evaluation errors revealed in FrontierMath benchmark
By
–
The irony. Using AI (GPT 5.5), errors have been identified that would affect 1/3 of the problems in the FrontierMath benchmark Tiers 1-4. This could represent a shift in the evaluations of AI's mathematical capabilities, which may be underestimated.
-
Claude Code introduces Agent View for managing parallel AI sessions
By
–
🆕 Claude Code just dropped Agent View — and it's a big deal for devs running parallel AI sessions 👇
— Futurepedia – Learn to Leverage AI (@futurepedia_io) 12 mai 2026
Instead of managing 10 terminal tabs, you now get ONE dashboard to:
▪️Launch & background agents
▪️See what's working, waiting, or done
▪️Reply inline or jump into full sessions… pic.twitter.com/40WD7u84oYClaude Code just dropped Agent View — and it's a big deal for devs running parallel AI sessions Instead of managing 10 terminal tabs, you now get ONE dashboard to:
Launch & background agents
See what's working, waiting, or done
Reply inline or jump into full sessions