“(1) Scaling data centers: This still scales for ~2 years. (2) Scaling through dynamics: Route to smaller specialized models or larger/smaller models. (3) Knowledge distillation: I believe distillation behaves differently than other techniques and might have different properties.” / X
“@craigsdennis made a nice proxy server and video here!
Taiwanese law prevents TSMC from producing 2nm chips overseas, Taiwanese govt official confirms | Tom’s Hardware
The Great American Microchip Mobilization | WIRED
Amazon
“Anthropic released Claude 3.5 Haiku on their API, Amazon Bedrock and Google Cloud’s Vertex AI The new model outperforms GPT-4o on SWE-bench Verified and even surpasses Claude 3 Opus on many benchmarks But comes with increased pricing
AMD
AMD unveils Versal Premium Series Gen 2 for data center workloads | VentureBeat
AMD will lay off nearly 1,000, or 4% of staff, as AI competition heats up | VentureBeat
NVIDIA
Elon Musk’s xAI raising up to $6 billion to purchase 100,000 Nvidia chips for Memphis data center
“balanced thoughts on the “ai is hitting a wall” meme: – gpt5, 3 opus, gemini 2 delays are real. no >2T models have made it past release 1.5yrs after GPT4 – its poignant that Nvidia is hyping FP4 perf just as research comes out saying FP4 is quantizing too far – ilya, noam,” / X
“@Nvidia benefits from AI/GenAI (from the tools of Cadence, Synopsis and Applied Materials) and Nvidia uses AI/GenAI themselves as well. Most of it is in design time (which is majority of work in chips business)! I have attended few chip design conferences in the heart of https://x.com/sarbjeetjohal/status/1855727272822796343
Leave a Reply