19 lines
1.6 KiB
Markdown
19 lines
1.6 KiB
Markdown
icon:: 🧠
|
|
|
|
- ***Agentic AI***: Systems that can *plan*, *reason*, and *take action* to *accomplish tasks* with *minimal human intervention*.
|
|
- AI agents extend the capabilities of LLMs with respect to 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 the 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
|
|
- **LangChain**
|
|
- |