What are AI Agents? | Highlights and Annotations by Gistr.

This video explains the shift in AI from monolithic models to compound AI systems and then to AI agents. Compound systems use multiple components (models, databases, programs) to solve problems, improving adaptability over tuning single models. AI agents take this further by using large language models to control the logic of the compound system, enabling more complex problem-solving through reasoning, action (using external tools), and memory. The video uses the example of calculating sunscreen needs for a vacation to illustrate the power and flexibility of AI agents, contrasting them with simpler, more programmed systems. and when we put a LLMs in charge of the logic, this is when we're talking about an agentic approach. so let's break down the components of LLM agents. the first capability is the ability to reason, which we talked about.. so this is putting the model at the core of how problems are being solved.. the model will be prompted to come up with a plan and to reason about eachstep of the process along the way.. another capability of agentsis the ability to act. and this is done by external programs that are known in the industry as tools.. So tools are external pieces of the program, and the model can define when to call them and how to call them in order to best execute the solution to the question they've been asked.. So an example of a tool can be search, searchingthe web, searching a database at their disposal.. another example can be a calculator to do some math.. this could be a piece of program code that maybe might manipulate the database.. this can also be another language model that maybe you're trying to do a translation task, and you want a model that can be able to do that.. and there's so many other possibilities of what can do here.. so these can be APIs. basically any piece of external program you want to give your model access to.. third capability, that is the ability to access memory. and the term "memory" can mean a couple of things.. so we talked about the models thinking through the program kind of how you think out loud when you're trying to solve through a problem.. So those inner logs can be stored and can be useful to retrieve at different points in time.. But also this could be the history of conversations that you as a human had when interacting with the agent. and that would allow to make the experience much more personalized.. So the way of configuring agents, there's many are ways to approach it.. one of the more most popular ways of going about it is through something called react, which, as you can tell by the name, combines the reasoning and ACT components of LLM agents.. so an alarm, the alarm is given a prompt. so the instructions that's given is don't give me the first answer that pops to you..08:35think slow planning, your work. and then try to execute something. tried to act. and when you want to act, you can define whether. if you want to use external tools to help you come up with the solution.. once you get, you call a tool and you get an answer.08:55maybe it gave you the wrong answer or it came up with an error. you can observe that. so the alarm would observe. the answer would determine if it does answer the question at hand, or whether it needs to iterate on the plan and tackle it differently.09:09up until I get to a final answer. Monolithic models are limited by their training data and are difficult to adapt, requiring significant investment in data and resources. Compound AI systems, conversely, are modular and combine multiple components (models, programs, databases) to solve problems. This modularity makes compound AI systems much more adaptable and easier to modify for different tasks. They can be adjusted by changing components or their interactions, unlike monolithic models which require retraining. This allows for quicker adaptation to new problems and improved accuracy. Add my notes... The "React" approach combines reasoning and action capabilities in LLM agents. A user query (e.g., "how much sunscreen for my Florida vacation?") is given to the model, which is instructed to plan its solution methodically rather than providing an immediate answer. The model then executes its plan, potentially using external tools (like weather forecasts or sunscreen dosage websites). If the initial attempt fails, the model observes the outcome, refines its plan, and iterates until a satisfactory answer is found. In the sunscreen example, the model might first determine vacation length, then research Florida's sun hours, find recommended sunscreen application rates, and finally calculate the number of bottles needed. React is a way to make AI agents smarter. Instead of giving a quick answer, the agent thinks through the problem step-by-step. For example, if you ask "how much sunscreen?", it won't just guess. It will plan: find out how long your trip is, check Florida's sun levels, learn how much sunscreen to use, and then calculate how many bottles you need. If it makes a mistake, it tries again until it gets the right answer.