2.5 KiB
2.5 KiB
icon:: 🧠
- Agentic AI: Systems that can plan, reason, and take action to accomplish tasks with minimal human intervention.
- Certainly! Here's the revised text with grammatical corrections:
- Agentic AI: Systems that can plan, reason, and take action to accomplish tasks with minimal human intervention.
-
- AI agents extend the capabilities of LLMs in the following dimensions:
-
- Integrating memory to retain and recall information across interactions
-
- Tool use to utilize external tools, APIs, and databases
-
- Decision-making to plan and execute multi-step workflows
-
- AI agents will continue to refine their ability to reason, plan, and act.
-
- The development of agentic-based AI is a natural progression from statistical modeling to deep learning and now to reasoning-based systems
- This evolution marks a shift from predictive models to autonomous systems capable of dynamic decision-making.
- The main challenges with raw LLMs are the following:
- Context window limitations: The context window typically range from 2,000 to 128,000 tokens and this create the following challenges:
- Document processing: long documents must be chunked to deal with context limits
- Conversation history: A proper memory management is required to maintain infromation across extended conversations.
- Cost management: It is necessary to make efficient token use, because most providers charge based on token count.
- Limited tool orchestration: to execute complex workflows a proper infrastructure is needed to discover tools and manage tool interactions across multiple turns.
- Task coordination challenges: structured control mechanisms are require to manage with LLMs multi-step workflows
- Context window limitations: The context window typically range from 2,000 to 128,000 tokens and this create the following challenges:
- LangChain
- Open-source framework and venture-backed company based in San Francisco
- Main features:
- Composable workflows: the LCEL - LangChain Expression Language permits to break down complex tasks into modulare components that can be assembled. This enable the orchestration of multiple processing steps.
- Integration ecosystem: it provides interfaces for all generative components i.e., LLMs, embeddings, vector databases, document loaders, search engines. This will permit to switch between providers without rewriting core logic.
- Unified model access: interfaces to different language and embedding models
- Application development concepts:
- Memory and state management
- Agent architecture