Sublime
An inspiration engine for ideas
一扬
@yiyang-0b86268fc9034f3a
前20年的中国互联网浪潮(2000-2020),主体都是社交、视频、娱乐、电商、平台经济,不需要尖端技术,只需要 商业铺开和增长,再牛逼的ACM算法大神,也要给商业奇才们打工——中国就没有纯技术饭;
2013年后,deep learning浪潮来临,商业化落地不明显,一群大神挖出来,给AI四小龙IPO做嫁衣;
deepseek 是第一次由梁文峰花钱,给各种NOI和ACM冠军们在LLM的竞赛上公平竞争,这是他们经历的、在中国本土上的第一次全球竞争,智力优势一下子就显示出来了。
再牛逼的大神,像楼天成这种,给pony ai做self driving system,也要老老实实抽象设计一套rule based的框架,进了拼多多也要写抢红包,给各种中间件修修补补,在抖音快手里踏踏实实写搜广推屎山... See more
The Landlord Entrepreneur: Double Your Profits with Real Estate Property Management
amazon.com
In an industry built on open-source interoperable software, we believe that partnerships and community are the only sustainable moat. We've spent years helping to drive growth initiatives at leading companies in the space including Aztec, Celo, CoinList, Index Coop, and Solana. Seeing the challenges of growing crypto projects firsthand is what insp... See more
Lattice Capital • The Web3 Growth Playbook
Nine months after Li’s speech, China’s State Council—roughly equivalent to the U.S. president’s cabinet—issued a major directive on advancing mass entrepreneurship and innovation. It called for the creation of thousands of technology incubators, entrepreneurship zones, and government-backed “guiding funds” to attract greater private venture capital
... See moreKai-Fu Lee • AI Superpowers: China, Silicon Valley, and the New World Order
peipei liang
@poweroom
Dickson Liew
@cmp
Home
ondelta.io
过去一年,我一直跟大家讲,deepseek、moonshot、SRAM推理三家,都是可能挑战OpenAI的存在。
moonshot就不说了,他们解决的核心问题是context window,这是用户解决复杂问题的核心需求,moonshot最开始给出了非常惊艳的答案——可惜后来别人也追上来了;
deepseek v2走了mixture of experts的路, 这条路Mixtral 8x7B已经帮他们探明了,就是用一个零头的training cost去完成training,就是省钱;另一个是multi head latent attention,能让deepseek早期用一堆烂显卡以极低的成本 提供inference服务(0.28刀提供1M参数);
SRAM推理三家(Groq、Samba... See more