Andrew Ng On AI Agentic Workflows And Their Potential For Driving AI Progress

In this Luminary Talk given during Dev Day at Snowflake Summit 2024, Landing AI Founder and CEO Andrew Ng talks about AI agentic workflows and how they could drive more AI progress than even the next generation of foundation models. He describes major agentic workflow design patterns — such as reflection, tool use, planning, and multi-agent collaboration — and why they're a powerful tool for developing LLM-based applications. Introductory remarks are offered by Snowflake CEO Sridhar Ramaswamy.

00:00 -- Opening remarks by Snowflake CEO Sridhar Ramaswamy

06:09 -- Conversation between Sridhar and Landing AI Founder and CEO Andrew Ng

13:38 -- Presentation by Andrew Ng

To view the Opening Keynote, click here:

To view the Platform Keynote presentations, click here:

[yoyartube]

13:39 is when Andrew's actual presentation starts.

[bichhubiswa]

Andrew is such a humble person I always loves to hear him ❤

[willblumrosen]

What a fantastic presenter, thank you Andrew!

[varunagarwal]

Wow, what an introduction to the actual presentation. I can only imagine how motivating it must be for the employees who are lucky enough to experience this level of insight on a regular basis. :)

[maksimhajiyev]

Building AI applications with agentic workflows introduces significant challenges, especially when these workflows rely on large language models (LLMs). The problem is that LLMs aren’t deterministic—they don’t always produce the same output given the same input. This unpredictability becomes even more problematic when we try to layer additional complexity, like autonomous agents, on top of an already unstable system.

In many enterprise settings, where reliability and consistency are key, these complex agentic workflows can cause more harm than good. Most tasks can be handled without the need for this extra layer of abstraction. By using LLMs in a more controlled and straightforward way, you can get the job done without introducing unnecessary risks. Instead of over-complicating the architecture, focusing on simpler, tightly managed LLM-based solutions can lead to more stable and reliable outcomes—exactly what’s needed for critical enterprise operations.

[sirishkumar-mz]

With its creative processes, SmythOS is leading the way in advancing AI. For those who are serious about utilizing AI, it is an essential tool. #TechTrends #AI #Innovation #SmythOS

[technologyandinnovation]

Wish Andrew had talked about more of activities in the field but ran out of time. If the first dude have just introduced Andrew and let him have a stage - but he needed to waste everyone's time.

[The_AI_Adoption]

"The Level 2 of AI Adoption is arriving before AGI"

The Last AI of Humanity (New Book )

[MarcosSanchez-rj]

🎯 Key points for quick navigation:

00:00:14 *🌟 The event focuses on bringing together the builder community, irrespective of whether they are Snowflake customers or not, aiming to foster idea exchange and inspiration.*
00:00:54 *📸 The speaker humorously shares excitement about meeting industry luminary Andrew Ng, highlighting the enthusiasm for technology development.*
00:01:36 *🚢 Snowflake encourages developers to innovate through examples like building applications without the need for deploying servers.*
00:02:18 *🏗️ Snowflake is increasing its openness by transitioning to a platform with more community-led development and open-source contributions.*
00:02:43 *🎉 The company recently concluded its first international AI hackathon, showcasing its commitment to fostering innovation and supporting developers.*
00:03:25 *🤝 Snowflake collaborates closely with startups to help them scale applications, highlighting partnerships with startups earning millions on the Snowflake Marketplace.*
00:04:18 *🏆 The startup challenge hosted by Snowflake awarded significant investment, with Signal Flare emerging as a winner.*
00:05:02 *🚀 Snowflake's accelerator programs launched with VC firms invest up to $100 million in early-stage startups.*
00:05:45 *🌍 They've launched the NSTAR education program to provide free training resources globally, advancing skills related to data and AI.*
00:06:09 *🌟 Andrew Ng is introduced as a luminary speaker, hailed for his extensive contributions to the AI field, prompting reflections on AI's transformative potential.*
00:07:05 *📈 Andrew Ng emphasizes the potential impact of AI agents and agentic workflows, encouraging exploration of scalable and efficient AI systems.*
00:07:59 *🌐 Ng appreciates initiatives like open-sourcing models, promoting broad technological access to prevent monopolization.*
00:09:09 *⚖️ Ng expresses concern about regulatory proposals in California which might stifle open-source innovation in AI.*
00:11:12 *🚦 Ng advocates for thoughtful regulation focusing on harmful applications rather than stifling technological advancement.*
00:11:55 *🔍 The conversation explores the transformative potential of AI agents in enhancing the capabilities and benefits of large language models.*
00:13:52 *🤖 Ng introduces the concept of agentic workflows, comparing them to traditional zero-shot prompting, highlighting iterative improvements in AI performance.*
00:16:12 *💡 Agentic workflows showcase significant improvements in results, suggesting their potential to surpass advancements between different model versions.*
00:17:07 *🏄 Ng’s team demonstrates the use of vision agents to automate tasks using visual datasets, exemplified by tracking distances in shark-surfer videos.*
00:18:02 *🔬 The vision agent automatically generates code to perform complex visual tasks, demonstrating its efficiency and utility.*
00:19:41 *⚙️ The innovative use of vision agents can streamline computer vision tasks, making application development more efficient for users.*
00:21:12 *🧪 Testing mechanisms enable vision agents to correct and enhance code generation, improving accuracy and efficiency in application outputs.*
00:23:00 *📄 Automatic code revision and testing through vision agents demonstrate significant potential for automating complex visual analyses.*
24:50 *🚗 The speaker expresses excitement about agentic AI for various applications, although current systems have limitations such as common failures in object detection.*
25:36 *🧠 The current AI system struggles with complex reasoning, often requiring precise prompt tuning to achieve accurate results.*
26:17 *🛠️ The visual agent software is in beta and sometimes yields impressive results, though it remains sensitive to prompt phrasing.*
27:00 *🌐 The vision agent team released the core engine as open source, aiming to contribute to the advancement of AI agents and improve developer capabilities.*
28:16 *⚖️ AI agents are being utilized across various applications, including legal work and complex document analysis.*
29:22 *📚 AI research agents engaging in tasks such as web searches and document synthesis are gaining traction and becoming more practical.*
30:19 *🚀 Recent advancements have transformed many AI applications from novelties into functional tools, highlighting a shift towards practical utility.*

Made with HARPA AI

[outofbody]

Before you know devs will be out of job before anyone first - automate what you know first - devs know coding

[FullChick-hl]

Wilson Elizabeth Lopez Sharon Martin David