Looking to better understand AI Agents

I just completed the Multi AI Agent Systems with crewAI course, but I don’t see a Q&A for this course so asking here.

In one of the later lessons, we created AI agents to perform financial analysis. We created a data analyst agent that had this back story:
backstory="Specializing in financial markets, this agent "
"uses statistical modeling and machine learning "
"to provide crucial insights. With a knack for data, "
"the Data Analyst Agent is the cornerstone for "
“informing trading decisions.”

I understand that these agents can use various tools that we provide. However, it says “uses statistical modeling and machine learning…” Fundamentally, these agents are just LLMs that can interact with tools to accomplish tasks. If someone were to tell me to use statistical modeling and machine learning I would actually create a ML model, but I don’t believe this agent created any statistical or ML models.

Can these AI Agents actual create/use a ML model to gain insight into a particular problem? In theory, you could create an agent to code a ML model, then run the model with traditional code, then have an agent evaluate the output and continue to iterate. But that doesn’t seem to be what happened in this example. Can anybody confirm?

All the Short Course forums are from this page:

Hey Connor. I also questioned this example.

The data analyst agent is clearly stated to “do statistical modelling” and ML. This is definitely possible if we build a more advanced agentic system where we give the agent access to a python tool that can execute code. Also, we’ll need some sort of data source interfaced with a tool that the agent knows about (Schema, etc).

However there is no way it can perform the statistical or ML analysis that is stated in the backstory, given the toolset that is fed into the agent (web scrape and web search). I wanted to ask the DL.AI team if this was a mistake? This example really doesnt make sense, it may just be a mix-up of backstories though.

Hey guys,

I’ve been rather interested in this too-- Unfortunately I have not yet gotten to the GANs specialization, or the short course in reference. That said I am on the verge of completing the DL specialization…

And I’ve just been really wondered, is this ‘multi-agent’ concept kind of just a ‘poor mans GAN’ ?

I mean that without offense, I mean yes you can fine tune, but so far very few can build a true LLM from scratch.

In any case, I was curious, is this the basic concept, or is it something else ?

That is to say it is not like people aren’t thinking about this, but I remain skeptical…

Sam Altman says helpful agents are poised to become AI’s killer function | MIT Technology Review

To increase the chances that the course staff will reply to your question, it really should be posted in the Short Course forum area.

This “AI Discussions” forum is really just for general topics.

your topic thread was missed because you had not selected the right category, I have moved it to the related category you are asking query from.

now comes to your question, surely the ai agent can create model using a python tool, but in this course, it is more related to the specific tasked assigned to the agent here based on the query and analysis, to get a better outcome of result than any usual chatbot.

what you are stating about would be surely coming forth with evolution of crewAI.