๊ฐ€์ž… ํ›„ ์ดˆ๋Œ€ ๋งํฌ๋ฅผ ๊ณต์œ ํ•˜๋ฉด ๋™์˜์ƒ ์žฌ์ƒ ๋ฐ ์ดˆ๋Œ€ ๋ณด์ƒ์„ ๋ฐ›์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

Jesus Rodriguez
@jrdothoughts
Co-Founder, CTO-CPO of @SentoraHQ (fmr IntoTheBlock), Co-Founder of @layerlens_ai, @neuralfabric( acq by Cisco) and The Sequence, Teaching at Columbia-Wharton
๊ฐ€์ž… September 2011
1.6K ํŒ”๋กœ์ž‰ ์ค‘    6.6K ํŒฌ
The AI demo gets the applause. The infrastructure gets paged. ๐Ÿ“Ÿ Behind every blinking cursor, GPUs crunch tokens, sandboxes run code, and some poor engineer makes sure the agent can't email the production database to a stranger. In this week's The Sequence Learning Loop, we pop the hood on last week's agent stack: โš™๏ธ DeepSeek's DeepInfer opens up its production inference engine, including DSpark, a scheduler that decides how many draft tokens are actually worth checking. That's a 4โ€“7% throughput gain near production, plus a refreshingly honest regression at low load. ๐Ÿ’ค Google's Agent Compute Engine gives agents computers that know when to sleep: checkpoint the workspace, free the compute, wake up right where you left off. ๐Ÿ›ก๏ธ NVIDIA builds the infrastructure around the whole agent, from running models to enforcing the rules. The big takeaway: the next gains in AI come from running models cheaply and trusting where they act. ๐Ÿ‘‡
๋” ๋ณด๊ธฐ