[Framework] How AI Remembers You

Welcome to another edition of the best damn newsletter in AI.

We’re shaking it up a little bit today!

First, this newsletter is staying 100% free. And we’re doubling down on our commitment to make it an indispensable resource in your AI adoption journey.

Second, we’re going to be going deeper on the AI topics that matter for business. We’ll still cover news in a revamped “Here’s what everyone’s talking about” section. But instead of covering what everyone is talking about, we’re committed to covering what everyone should be talking about.

Third, we’re opening up Reader Questions. Our mission is to make using AI accessible to every business, and a huge part of that is making sure you’re not stuck and answering diverse questions. So, submit your question here. We’ll review every single one and cover answers here in the newsletter.

Here's what we're covering today:

• Understanding AI's advancements in short term memory, long term memory, and knowledge - frameworks for what AI knows about you

• Become a better presenter in 5 minutes by using ChatGPT to mimic an audience for feedback

• OpenAI's new developments in personalized AI memory and text-to-video models

Let’s get to it! 👇

THE TOP NEWS

• Google released yet another update to their most powerful large language model, Gemini. And people are freaking out because it can handle up to 1 million tokens (eg. 750K words) which means it can analyze more information, longer documents and more data to do your task.

• OpenAI is experimenting with a new way to personalize your ChatGPT experience with “memory”. As you chat, you’ll be able to tell ChatGPT to remember things about you, which you can then use in future chats.

• OpenAI also announced Sora, their text to video model. Chances are you’ve seen plenty of mind-blowing demos on social media; but in case you haven’t — check out their announcement page for examples of how crazy good it is.

TODAY’S PERSPECTIVE

Understanding advancements in what AI knows about you (a framework)

Every AI company is battling it out to find the best ways to make your AI as helpful as possible.

And one of the most important components is what AI knows and remembers about you and your work.

So today we’re going to give you our framework for how to think about the recent advancements in 3 buckets: short term memory, long term memory, and knowledge.

Advancements in AI Short term memory:

Many AI products are working to expand the amount of information that AI can consider at a time.

AI short term memory functions similarly to peoples: Imagine your brain's ability to hold a phone number in your head just long enough to dial it, or remember what you wanted to order just long enough to say it to a waitress. Short term memory is about the AI's capacity to "keep in mind" a vast amount of information temporarily during a chat conversation or a task.

Expanding AI’s short term memory helps it consider a larger volume of information or context as its doing a task, and we expect we’ll continue seeing this capability grow (Google Gemini Pro’s 1 million context limit being a great example!)

Advancements in AI Long term memory:

There are also many AI products that either let you create your own experiences with pre-defined context, or will start to remember this context automatically for you.

Where short term memory has a limited capacity and resets with each new conversation; long term memory can persist over time. But similarly to people, just because you store it in your long term memory, doesn’t mean its always easy or fast to recall that information (so it’s not 100% dependable).

OpenAI’s custom GPTs, custom instructions and memory features are all about long term memory.

Advancements in AI Knowledge:

Finally, we have AI knowledge. This is about giving AI access to specific information or data that it can use to complete a task.

Like a well organized Google Drive, they key to making the most of knowledge is to have a good structure that makes it easy for AI to find and use the most relevant information for the task.

“Chat with your data” products and the knowledge upload in Custom GPTs are great examples of adding knowledge to AI.

USE CASE

Become a better presenter in 5 minutes

Do you ever wish you knew what your reader or listening was thinking?

Well here's your way to get it.

One of our favorite hacks is to use ChatGPT to mimic an audience and ask it for feedback, before you deliver a presentation.

This applies to marketing, creating content and even writing emails too.

All you need to do is tell ChatGPT context about your audience, give it your material, and ask questions like:

1. What emotions are the audience likely to feel throughout this talk

2. If the audience only remembers one thing from this talk, what would that be

3. What tough questions might the audience ask me in the Q&A

LINKS

For your reading list 📚

A few more AI startups are tackling learning and predictive entertainment...

• Loora wants to leverage AI to teach English and we are very bullish on the use cases its tackling

• Qloo raises $25M to predict your favorite movies, TV shows and more using AI to find you better stuff to watch

Google is making waves in open source...

• Google Gemma: because Google doesn't want to give away Gemini yet so instead, they introduced a pair of open-source AI models for developers

Have a burning AI question? Or stuck trying to help your team adopt AI?

We read every single question submitted and use these to inform the topics we go deep on. 👇️ Submit your question below!

That's all!

We'll see you again soon! Thoughts, feedback and questions are much appreciated - respond here or shoot us a note at [email protected].

Cheers,

🪄 The AI Exchange Team