Lmst

Endre Stølsvik (@stolsvik)

DeepSeek이 자사 AI 네트워크, 학습 데이터, 인프라를 완전히 오픈소스로 공개했다는 주제의 트윗입니다. Anthropic이 이를 참고하거나 도구를 활용했다는 주장도 있습니다. AI 모델 개발 투명성과 오픈소스 전략의 파급력을 보여줍니다.

https://x.com/stolsvik/status/2026235056302284990

#deepseek #opensource #anthropic #aiinfra #transparency

My first time at #SCAsia/#HPCAsia this week. I don’t have a crazy preset agenda as I would at SC/ISC, so looking forward to having time to explore new experiences, hopes, and perspectives on #HPC and #AIinfra, meet people, etc.

FutureLivingLab (@FutureLab2025)

LLM에서 가장 어려운 부분은 모델 자체가 아니라 인프라라는 관점의 글. Andrej Karpathy가 언급한 'vibe coding'처럼 빠르게 개발하는 방식은 초기 속도에는 도움이 되지만 AI 인프라에서는 확장성이 떨어져 '빨리 배포 → 더 빨리 리팩터'하는 문제로 이어진다는 교훈과 대규모 시스템에서 먼저 깨지는 부분들에 대한 논의를 시작한다.

https://x.com/FutureLab2025/status/2008866617787343206

#aiinfra #llms #engineering #devops

Neocloud Economics: CoreWeave vs Nebius – Vertical AI Infra Crushes Hyperscalers (60-70% Margins) ⚡

Neoclouds own stack (chips→racks), dodge AWS debt/leasing. Nebius edges CoreWeave on costs; $10B+ ARR potential. AI training explodes demand

Why vertical? 2-3x cheaper GPUs vs cloud giants.

VCs: Next hyperscalers? Founders: Build atop. GPU wars incoming. 📈

#Neocloud #CoreWeave #Nebius #AIInfra #GPUEconomics

https://buff.ly/XJsjWqp

⚡ Cut GPU costs by 68% without slowing inference.
No hype. Just real infra lessons from shipping AI in production.

👉 Read the full story:
https://medium.com/@rogt.x1997/why-my-ai-startup-cut-gpu-costs-by-68-without-slowing-down-with-runpod-58f71733e86f

#GenAI #AIInfra #MLOps
https://medium.com/@rogt.x1997/why-my-ai-startup-cut-gpu-costs-by-68-without-slowing-down-with-runpod-58f71733e86f

La IA tiene hambre de energía y la solución es nuclear. ⚛️🤖 Analizamos el proyecto de Last Energy para crear mini reactores. ¿Sostenibilidad real o un riesgo necesario para el progreso tech? https://www.glitchmental.com/2025/12/reactores-nucleares-last-energy-ia-infraestructura.html #Nuclear #AIInfra #MastodonTech #CleanEnergy

ARBITER: what it is / what it isn’t

IS

semantic scoring
geometric fit
negative answers
offline 26MB

ISN’T

LLM
vector DB
embeddings
retrieval

getarbiter.dev
#AI #NLP #RAG #AIInfra #SemanticSearch

ScaleOps' new AI Infra slashes GPU costs by half for self‑hosted LLMs while giving full visibility into pods, model behavior, and even a helm flag for easy tuning. Curious how you can cut spend and keep control? Read the full breakdown. #ScaleOps #AIInfra #GPUcosts #SelfHostedLLMs

🔗 https://aidailypost.com/news/scaleops-ai-infra-cuts-gpu-costs-50-selfhosted-llms-adds-full

🚀 Tired of burning weekends fixing infra? RunPod 2025 makes GPU deploys boring (in the best way). Pods, endpoints & MCP turn ideas into live projects faster than ever. ⚡

👉 Read the full guide:
https://medium.com/@rogt.x1997/pods-endpoints-and-a-smoother-future-the-hidden-simplicity-of-runpod-f9bace9e1a8c

#RunPod #AIInfra #AIBuilders
https://medium.com/@rogt.x1997/pods-endpoints-and-a-smoother-future-the-hidden-simplicity-of-runpod-f9bace9e1a8c

🚀 Tired of burning weekends fixing infra? RunPod 2025 makes GPU deploys boring (in the best way). Pods, endpoints & MCP turn ideas into live projects faster than ever. ⚡

👉 Read the full guide:
https://medium.com/@rogt.x1997/pods-endpoints-and-a-smoother-future-the-hidden-simplicity-of-runpod-f9bace9e1a8c

#RunPod #AIInfra #AIBuilders
https://medium.com/@rogt.x1997/pods-endpoints-and-a-smoother-future-the-hidden-simplicity-of-runpod-f9bace9e1a8c

🇹🇼 #ASUS subsidiary #TaiwanAICLOUD building NVIDIA‑powered supercomputer (80→250 PFLOPS).
https://www.scmp.com/tech/tech-trends/article/3322882/taiwanese-laptop-maker-asus-subsidiary-builds-supercomputer-boost-islands-capacity
#NVIDIA #HPC #AIInfra #Taiwan

🔀 LiteLLM @LiteLLM

Open-source LLM gateway + Python SDK for 100+ providers in OpenAI format. Integrates with Langfuse/LangSmith/OTEL for unified views.

https://www.everydev.ai/tools/litellm

#OpenSource #UnifiedAPI #LLMOrchestration #AIInfra #DevTools

🛰️ Martian @withmartian

OpenAI LLM router with a willingness_to_pay knob—adjust the base_url to select the best model (cost × quality × latency). Includes open-source adapters for easy multi-model use.

https://www.everydev.ai/tools/martian

#LLM #ModelRouter #AIInfra #DevTools #OpenAIAPI