Agentic AI – Đỉnh cao mới của AI sau GenAI | Trần Quốc Huy - WecommitAI

TL;DR

Key insights

The Transformer architecture was a pivotal moment for AI, enabling the processing of vast, unstructured data and leading to the AI boom.

• Before 2017, AI algorithms struggled with the sheer volume and complexity of human-generated data due to sequential processing limitations.
• The 'Attention is All You Need' paper introduced the Transformer architecture, which uses parallel processing and attention mechanisms to understand context and relationships between words.
• This breakthrough, combined with powerful GPUs (like those from NVIDIA), significantly accelerated AI development and adoption, leading to the rise of Generative AI (GenAI).
• The Transformer architecture is considered a foundational model applicable across various domains, including text, images, and audio.

Generative AI (GenAI) has limitations that necessitate further advancements, leading to Agentic AI.

• GenAI's core function is predicting the next word based on statistical probabilities, not true understanding, which can lead to 'hallucinations' or fabricated information.
• GenAI models suffer from limited 'memory context' or short-term memory, causing them to forget previous parts of a conversation.
• These limitations make GenAI unreliable for critical business applications where accuracy and consistency are paramount.
• Techniques like Retrieval-Augmented Generation (RAG) were developed to address these issues by incorporating external data sources.

Agentic AI represents the next evolutionary step beyond GenAI, enabling AI to act and interact autonomously.

• Agentic AI moves beyond simply generating content to performing actions, reasoning, and interacting with tools and data.
• Key concepts like 'Chain of Thought' (CoT) and 'ReAct' (Reasoning and Acting) allow AI to break down complex problems, plan steps, and execute actions.
• Projects like 'SmallVille' demonstrated multi-agent systems where AI agents with distinct roles could interact and simulate social behaviors.
• Agentic AI can be integrated with existing systems (databases, APIs) and can operate over extended periods, significantly enhancing productivity and capability.

Adopting Agentic AI offers a significant competitive advantage by enabling businesses to scale expertise and automate complex tasks.

• Unlike using off-the-shelf GenAI tools, designing custom Agentic AI systems allows businesses to leverage their unique knowledge, processes, and data.
• Agentic AI can automate tasks previously requiring experienced human engineers, such as database optimization and performance analysis.
• The ability to create multi-agent systems that collaborate and execute complex workflows revolutionizes how businesses operate.
• Early adoption of Agentic AI positions businesses to gain substantial advantages in productivity, innovation, and market competitiveness.

Key moments

• 00:01:16The AI boom is attributed to the 2017 introduction of the Transformer architecture, which solved the problem of processing massive, unstructured data by enabling parallel processing and contextual understanding, unlike previous sequential methods.
• 00:08:58Generative AI (GenAI), while powerful, has limitations like 'hallucinations' and short memory context, making it insufficient for many real-world applications. Techniques like RAG were developed to mitigate these issues.
• 00:20:53Agentic AI represents the next step, enabling AI to act autonomously by combining reasoning (like Chain of Thought) with actions (like ReAct), allowing them to interact with tools and data.
• 00:25:10The 'SmallVille' project demonstrated the potential of multi-agent systems where AI agents with distinct roles could interact, learn, and simulate complex social behaviors without explicit programming.
• 00:29:46A practical application of Agentic AI is shown in optimizing database projects, where a team of specialized agents, led by a coordinator, automates complex analysis and decision-making, significantly boosting efficiency and scalability.
• 00:34:35Transitioning from being a user of AI to designing and implementing custom Agentic AI systems provides a significant competitive advantage by leveraging unique business knowledge and processes.