You.com转发了
AI has a "last-mile problem" similar to self-driving cars. With self-driving cars, early demos impressed, but real-world deployment took years. It's easy to hack up a prototype, but making it work reliably at scale is hard. If each step of an AI agent is only 95% accurate, none of the 30-step workflows will work reliably.? Going from 95% to 99.9% accuracy is the real challenge. Full episode: https://lnkd.in/gmfbGwis
Jessica Kriegel this is why I was saying that people are generally over-estimating the ability of AI Agents. Richard's company (you.com) does deeper research into agentic abilities of AI than most other AI companies. A key comment from him is: "If each step of an AI agent is only 95% accurate, none of the 30-step workflows will work reliably. "
Don't build fully autonomous AI Agents that can/will go wild. Instead, only insert agentic behaviour in some parts of a more deterministic workflow. I call this paradigm "agents on rails".
I agree entirely, wish the hyperbole would stop and we could get to building. https://tyingshoelaces.com/blog/agentic-framework-lie
AI still wrestles with nuanced real-world scenarios. Progress comes through tenacity.
YES. The AI Last Mile is real. Data and especially data in domain matters so significantly. Anyone building real AI products has felt this.
Reliability is the important word here.
Nailing the last mile is key. Scaling AI from 95% to 99.9% accuracy is where the real magic happens.?
Leading AI Strategist | Expert Practitioner & Researcher | Operations & Change Management | Board Advisor | International Speaker & Author
2 天前Been watching the You.com journey for quite some time now, with excitement. I also keep mentioning you.com in my talks, all over the world. It's not all about Perplexity, at all (!), you know, but that's what the world still seems to think, so many months down the line. What's the next move, Richard Socher?