Amazon делает ставку на агентов, чтобы победить в гонке искусственного интеллекта

## Key Takeaways from the Interview with Amazon's AGI Research Lead This is a fantastic, in-depth interview! Here's a breakdown of the key takeaways, categorized for clarity: **1. The Current State of AI & Agents:** * **Current Agents are Limited:** The speaker stresses that *most* of what's currently marketed as "agents" are just chatbots with added automation—they don’t truly *act* and learn from consequences. They often take longer and are less reliable than direct human action. * **LLMs are a Foundation, Not the Solution:** LLMs are good at processing text, but they lack understanding of consequences and the ability to interact with the real world. They *hallucinate* and can be dangerously unreliable when asked to *do* things. * **The Promise of True Agents:** True agents will fundamentally change how we interact with computers and software. They will *execute* tasks autonomously, leveraging all available tools and resources (like a lab's equipment in the drug discovery example). This unlocks *massive* leverage for human workers. **2. Amazon's Approach to AGI & Agent Development:** * **Focus on General-Purpose Agents:** Amazon is betting on a *general-purpose* agent that can handle a wide range of tasks, rather than specialized agents. * **Training Agents Through Simulated Environments ("RL Gyms"):** Amazon is building realistic simulations of work environments (Salesforce, ERPs, CAD programs, medical record systems, accounting software) to train agents. This allows them to learn consequences and refine their actions *without* real-world risk. They're essentially learning to "solve problems" within these simulations before being deployed to the real world. * **AGI as a "Universal Teammate":** Amazon’s vision of AGI isn't about replacing humans, but augmenting their abilities. The goal is to offload the execution of tasks, freeing up humans to focus on higher-level thinking and creativity. * **AGI as a New Building Block of Computing:** Amazon sees AGI as shifting the foundation of computing: from servers, storage, and code to *intelligent agents* that *do* things. **3. Google's Approach (Comparison):** * Google is focusing on building *specialized* agents for specific tasks, while Amazon is aiming for general-purpose AGI. **4. Adept & The Reverse Acquisition:** * Adept was a pioneer in focusing on agent development. They created the first production-ready agent (focused on computer-use automation). * Amazon acquired Adept to accelerate their agent development efforts. **5. Definition of AGI (Shift in Focus):** * **Early OpenAI Definition:** A system that outperforms humans at economically valuable tasks. * **Current Amazon Definition:** A model that can help a human do *anything* they want to do on a computer. This is more practical, grounded, and focused on human augmentation. **In essence, the interview paints a picture of a significant paradigm shift in AI. The speaker is advocating for moving beyond LLMs as the primary focus and towards building truly *acting* agents that can learn, execute, and unlock new levels of productivity and innovation.** Amazon’s strategy is ambitious and focuses on the long-term potential of general-purpose AI agents.