正在加载视频...
视频加载失败
Ex-Google Jeff Dean just released the best 1-hour lecture on AI engineering: from basics to Graphs 1:45 - LLM from scratch 17:22 - How to use AI models 30:03 - Prompt engineering 52:35 - One human coordinating 100 agents 27 years of building AI at Google, compressed into one... show more
16 条评论
Here is the Youtube video if someone is interested:
27 years of experience condensed into an hour... this is an absolute must-watch for any AI engineer. 🧠
keep posting more
hundred agents one human, sure
One hour covering everything from LLMs to 100 AI agents.That’s a serious masterclass.
The 52:35 part on one human coordinating 100 agents is the one people will skip. In practice the hard part isn't coordination, it's deciding when to trust the output.
The useful cut is the last chapter.LLM from scratch and prompting are table stakes. The hour only gets interesting at 52:35 — one human coordinating 100 agents.That’s the job now: orchestration, not another prompt trick.
jeff dean's a genius, love watching his talks
one of my fav video with him, gonna watch it again
27 years in one video, saving this
context engineering beat prompt engineering as a term because the input isn't just words anymore — retrieval order, tool results, session state, timing. the prompt is the smallest part.
any slides mate ?
The jump from “how to prompt a model” to “one human coordinating 100 agents” is the whole story. At 100 agents, the hard problem isn’t prompting anymore 🤓 It’s state, ownership, delegation, verification, authority, recovery - and giving one human a control surface that doesn’t turn them into the message bus. That’s where agent engineering becomes runtime engineering.
AI engineering is shifting from building models to orchestrating intelligence
nice
Once one person can coordinate 100 agents, the bottleneck isn’t throughput — it’s decision rights and quality bars. Who sets the definition of done, who catches silent failure, who owns the call when agents disagree. That’s the new operating system.
