There has been a fair amount of discussion on what makes an A.I. agent an agent. |
Does it need to intelligently break-down a larger goal into tasks using an LLM?
Does it need to be able to run autonomously, without human intervention?
Does it need to self-reflect and self-adjust? |
These would all be neat but they’re not strictly necessary. |
An agent’s an agent no matter how small. (h/t Dr. Seuss) |
|
h/t Dr. Seuss’s “Horton Hears A Who” |
|
In my mind, an agent is just software that uses A.I. to accomplish a useful task by executing multiple steps. That’s it. Sometimes, those steps are determined by an LLM. That’s the basis for agent frameworks like CrewAI and LangChain (disclosure: I’m an indie investor in both). But sometimes, the “orchestration layer” can just be plain-old imperative code. Turns out, users don’t care as long as the agent does the useful thing it’s supposed to do. |
Here’s an example I’m working on now: |
Can A.I. Generate A Meme That’s Actually Funny? |
I am building an AI agent to make memes. I promise, it’s work-related, because I often need memes for presentations I’m doing. |
I creatively named it the Meme Maker agent. |
The v1 is super-simple. You enter a topic of interest and the agent does its best to |
1) Figure out elements of that topic that could be funny. |
2) Generate the concept and text for the meme. |
3) Create the visual for the meme |
There’s no fancy orchestration here. It’s not “coordinating” work with other agents (yet) and it’s not exhibiting any form of executive functioning where it has to maintain state or memory. You just give it a topic of interest (hopefully a trending topic that has humor potential) and it will generate a meme for you. That’s it. |
This is what I have called a “Minimum Viable Agent”. But chances are, I’ll iterate on it and make it smarter over time. Perhaps have it suggest topics based on current events. Perhaps generate multiple memes in different styles and let you pick which one you like. Maybe even have an integrated voting mechanism for RLHF (Reinforcement Learning through Human Feedback).
But for now, it’s simple and I hope you find it useful.
Here’s an example. You can see if you can generate a funnier meme yourself. |
|
If you like what it generated, post it on social media and tag me (@dharmesh). Would love to see how it’s working for you.
Thanks.
-Dharmesh
|
|
|
To make the most of A.I., check out agent.ai, the professional network for A.I. agents. Simple agents to do company research, find prospects and much more. |
|
Update your email preferences or unsubscribe here
© 2024 @dharmesh
228 Park Ave S, #29976, New York, New York 10003, United States
|
Terms of Service |
|
|