Reading time: 6 minutes
Alright people! Since you guys signed up for this newsletter, today is Day #1 of understanding AI.
First, who am I?
Before I begin, a quick introduction about myself - I’m Ankur, a product marketer, content creator and finance enthusiast. At heart, I consider myself a storyteller who likes to make boring concepts interesting and easy to understand. In the past, I’ve managed to make finance fun with my newsletter Money Simplified - read by 10,488 people as of today.
I’m falling in love with AI, and now want to try my best to make sure everyone understands the nuances of AI (and not just ChatGPT). So this newsletter is an attempt of helping non tech folks master AI, and grow in their career.
Okay. Enough about me. Let’s begin!
Today, I’m going to talk about 3 basic concepts - LLMs, RAG and AI Agents. Feel free to skip any part if you know it well. I just want to make sure everyone knows the basics before we move to more advanced topics.
So, here we go!
LLM (Large Language Model)
An LLM is like a super smart, well-read robot that can understand and generate text.
Imagine reading millions of books, articles, and conversations. Over time, you'd start noticing patterns in how words fit together and how people talk about different topics. An LLM does something similar but with the help of powerful math and computers.
There are a lot of LLMs today, but some well-known ones are:
GPT (Generative Pre-trained Transformer) is the most common that we all use. ChatGPT is a chatbot (or software program) that uses GPT as the base LLM
Gemini is another LLM that is made by Google
Claude, yet another model, made by a company called Anthropic is said to be better than ChatGPT (personally, I’ve used both and I agree - Claude is definitely better at generating content)
LLaMa (Large Language Model Meta AI) is made by..no prizes for guessing..Meta. Again, I’ve used it, and one thing it does a brilliant job with, is image generation
Essentially, for any AI program or software, the LLM is at the core. Think of the AI software as a car and the LLM as the engine. The car can’t run without the engine.
(Softwares can use more than one LLM too, but that’s a story for another day..for now, just understanding LLMs would be a good start)
RAG
Now, how LLMs typically work is that they are trained on a huge set of data, and it generates output based on this training.
GPT 3 for example, has been trained on 175 billion parameters.
The most sophisticated model of LLaMa is trained on 405 billion parameters!
But essentially, an LLM can only work well based on the data it has been trained on. And while 405 billion parameters seems huge, it’s not the entire universe.
For example, GPT has already been pre-trained. Now if a news came out yesterday, it won’t be there in GPT-3, and if a software that uses GPT needs access to that news, it won’t get it.
In comes RAG.
Retrieval-Augmented Generation is a powerful AI technique that, as the name suggests, retrieves data from external sources, augments it by integrating it with its own data, and then generates a response based on the data it has retrieved - Thus doing Retrieval-Augmented Generation
RAG is a very important concept, because even though there are billions of parameters that LLMs are trained on, it’s still a finite universe. To make it specific for your query, you can train the LLM with some more data.
AI Agents
Okay, so you may use Gemini to, say, write an email or a LinkedIn post. Or maybe even create a presentation.
All of which is Generative AI.
While it’s great, it only generates an output for you. It cannot take any action on your behalf.
Now imagine an advanced version of Gemini, where it will:
create the content of the presentation
design it in Google slides
send it to a prospective client via email
document the response received from the client, and
send the client a calendar invite for a meeting
All on your behalf.
That’s an AI agent - Where AI will not just generate an output for you, but also make decisions and take actions on your behalf.
And that’s not it. AI Agents can actually perform a lot more complex tasks:
Build your itinerary and book your tickets, hotels and transfers for a vacation you want to take
Manage your finances, make investments, rebalance your portfolio and become a personal investment advisor to you
Become a personal shopping assistant, browsing through your history, selecting items, making a payment and even returning items as and when needed
And a lot more..
Essentially, AI agents will be software programs that can run independently and take decisions on your behalf.
Yeah I know what you’re thinking. Isn’t it scary?
Well, not really. AI Agents will, at the end of the day, be controlled by us humans only. Unless we are able to make a Rajnikant-style Chitti or Gilfoyle’s “Son of Anton” HBO’s “Silicon Valley”. THEN it’s time to be scared, haha.
AI Agents are going to be all the rage this year (and maybe the next too). Today, AI is doing to technology what the Internet did to computers in the 1990s. In this decade, we’ll see a remarkable shift in the way work is done.
(And since you’re subscribed to this newsletter, you’ll be a part of it! Wink wink)
Anyway, that’s all for today. This is just the beginning, and I have a lot more content to share with you guys. See you next week!
Something cool I explored with AI this week
I used this tool called relay.app that lets you build your own AI agent with no coding. It’s pretty cool! Can help you create AI agents for a lot of tasks!
(Although it was a bit complex to use - I’d advise to wait for a few weeks before trying it out, it also seemed slightly buggy)
Couple of other things I’d like to ask (and tell)
As you know, this is a brand new newsletter (and this is the first email). So I’d love if you could tell me what you’re looking for, from this newsletter. It’ll help me write relevant content that will eventually help both of us. Please let me know HERE - Just 3 mandatory questions, won’t take more than a minute!
I also write a newsletter simplifying business and money management. It has more than 10,400 subscribers as of today. If you’d like to read more about startups, the economy or money management, subscribe to it here
I write smaller, more frequent content on LinkedIn - You can follow me there for byte-sized stuff
If you liked this piece, please share it with your friends (SHARE button below). It’ll help spread awareness faster :)
Until next time…