Emily (@IamEmily2050)
Grok 에이전트 아키텍처의 채택 방식에 대해 설명하는 글로, 대규모 단일 모델을 수개월에 걸쳐 훈련·사후처리하는 방식 대신 소형·효율적인 모델들을 사용해 수주 내에 개선 가능한 구조라고 주장합니다. 향후 몇 달 내 광범위한 채택을 예측하고 있습니다.
https://x.com/IamEmily2050/status/2023948296616575017
#grok #agents #modelarchitecture #efficientmodels #llms
Emily (@IamEmily2050) on X
It seems many people still don't understand how the new Grok agents architecture will be adopted by everyone in the coming months. Instead of one big model taking six months to finish training and post training, it uses small, efficient models that are easy to improve in weeks,
Bindu Reddy (@bindureddy)
가격 대비 성능이 뛰어난 소형/저비용 모델들의 가치가 강조되었습니다. 추천 모델로 Gemini Flash 3(가성비 우수), Kimi K2.5(벤치마크 우수), Haiku 4.5(속도 우수)가 언급되어 일상적 업무 자동화와 실사용에서 작은 모델의 중요성을 주장하고 있습니다.
https://x.com/bindureddy/status/2021059401180905856
#efficientmodels #gemini #kimi #haiku
Bindu Reddy (@bindureddy) on X
Cheap models are extremely under-rated!
These models will automate routine tasks and will play a key role in achieveing
Best cheap / small models in the world today
Gemini Flash 3 - excellent price for performance
Kimi K2.5 - very good on benchmarks
Haiku 4.5 - faster and
#Google’s
#AIinfrastructure boss,
#AminVahdat, stated that the company needs to
#double its
#AIservingcapacity #everysixmonths to meet demand. This is a challenge, requiring a 1000x increase in capability over the next 4-5 years. The focus is on building more reliable, performant, and scalable solutions, leveraging
#customsilicon and
#efficientmodels.
https://www.cnbc.com/2025/11/21/google-must-double-ai-serving-capacity-every-6-months-to-meet-demand.html?eicker.news #tech #media #news 
sayzard
eicker.news ᳇ tech news