Avi Schiffmann

later on um if i ask like what is my favorite car it'll you know see the word car and like look for something kind of similar and you know it's not going to be a perfect match right you know porsche 911 is not the word car but it's close enough maybe it's like a 0.8.9 match right something like that like a 90 match um so it's close enough so you would just choose like some of these closest memories right now let's say we only have that one memory um we would take that memory and we might put that in let's say the

later on um if i ask like what is my favorite car it'll you know see the word car and like look for something kind of similar and you know it's not going to be a perfect match right you know porsche 911 is not the word car but it's close enough maybe it's like a 0.8.9 match right something like that like a 90 match um so it's close enough so you would just choose like some of these closest memories right now let's say we only have that one memory um we would take that memory and we might put that in let's say the

later on um if i ask like what is my favorite car it'll you know see the word car and like look for something kind of similar and you know it's not going to be a perfect match right you know porsche 911 is not the word car but it's close enough maybe it's like a 0.8.9 match right something like that like a 90 match um so it's close enough so you would just choose like some of these closest memories right now let's say we only have that one memory um we would take that memory and we might put that in let's say the

I don't know, let's say the user message that gets sent to this bot where the system prompt is, I'm a helpful assistant. Then your user message is, what is my favorite car? Then maybe you have underneath that, you augment this message and you say, in brackets, the user's favorite car is a Porsche 911. Or you maybe don't even need to augment the way that's written.

I don't know, let's say the user message that gets sent to this bot where the system prompt is, I'm a helpful assistant. Then your user message is, what is my favorite car? Then maybe you have underneath that, you augment this message and you say, in brackets, the user's favorite car is a Porsche 911. Or you maybe don't even need to augment the way that's written.

I don't know, let's say the user message that gets sent to this bot where the system prompt is, I'm a helpful assistant. Then your user message is, what is my favorite car? Then maybe you have underneath that, you augment this message and you say, in brackets, the user's favorite car is a Porsche 911. Or you maybe don't even need to augment the way that's written.

You can just say, my user chat message that I sent ages ago. my favorite car is a Porsche 911, right? And so when the system prompt sees, you know, your question and then, you know, the little context that you have there, it'll be able to say like, oh, you know, your favorite car is a Porsche 911, right? So that's like, kind of how it works.

You can just say, my user chat message that I sent ages ago. my favorite car is a Porsche 911, right? And so when the system prompt sees, you know, your question and then, you know, the little context that you have there, it'll be able to say like, oh, you know, your favorite car is a Porsche 911, right? So that's like, kind of how it works.

You can just say, my user chat message that I sent ages ago. my favorite car is a Porsche 911, right? And so when the system prompt sees, you know, your question and then, you know, the little context that you have there, it'll be able to say like, oh, you know, your favorite car is a Porsche 911, right? So that's like, kind of how it works.

Um, you would kind of store all of these memories and probably some kind of like vector database, right? Um, you don't really need to use a standalone tool and you know, a lot of people are probably using super base as their database and you can just kind of, you know, use like PG vector in there. They make it all pretty simple. But, um,

Um, you would kind of store all of these memories and probably some kind of like vector database, right? Um, you don't really need to use a standalone tool and you know, a lot of people are probably using super base as their database and you can just kind of, you know, use like PG vector in there. They make it all pretty simple. But, um,

Um, you would kind of store all of these memories and probably some kind of like vector database, right? Um, you don't really need to use a standalone tool and you know, a lot of people are probably using super base as their database and you can just kind of, you know, use like PG vector in there. They make it all pretty simple. But, um,

You know, to set up a system like this that can extract those memories and put those memories back into your chats alone is already, you know, pretty technical. But then, again, building a true memory system, like I was mentioning earlier, your memory is a lot more adaptive, right? Like, you know, you need to update existing memories. What if now my favorite car is a, you know...

You know, to set up a system like this that can extract those memories and put those memories back into your chats alone is already, you know, pretty technical. But then, again, building a true memory system, like I was mentioning earlier, your memory is a lot more adaptive, right? Like, you know, you need to update existing memories. What if now my favorite car is a, you know...

You know, to set up a system like this that can extract those memories and put those memories back into your chats alone is already, you know, pretty technical. But then, again, building a true memory system, like I was mentioning earlier, your memory is a lot more adaptive, right? Like, you know, you need to update existing memories. What if now my favorite car is a, you know...

Don't know a Pagani something whatever right now now that needs to be updated and if you think about the way your brains working to write like you have Like the importance of these memories matters a lot and that will degrade over time as well You know recency matters a lot for some of these memories because if it's trying to maybe it maybe you're trying to build a system that updates existing memories and maybe I said my favorite car and

Don't know a Pagani something whatever right now now that needs to be updated and if you think about the way your brains working to write like you have Like the importance of these memories matters a lot and that will degrade over time as well You know recency matters a lot for some of these memories because if it's trying to maybe it maybe you're trying to build a system that updates existing memories and maybe I said my favorite car and

Don't know a Pagani something whatever right now now that needs to be updated and if you think about the way your brains working to write like you have Like the importance of these memories matters a lot and that will degrade over time as well You know recency matters a lot for some of these memories because if it's trying to maybe it maybe you're trying to build a system that updates existing memories and maybe I said my favorite car and

It was a Porsche 911, but that was five years ago or five months ago, whatever. And only yesterday did I say that my favorite car was, you know, I guess I don't know. You know, fuck cars. I like motorcycles more. I'd say like a Ducati 916. you know, is now my favorite transportation thing, right?

It was a Porsche 911, but that was five years ago or five months ago, whatever. And only yesterday did I say that my favorite car was, you know, I guess I don't know. You know, fuck cars. I like motorcycles more. I'd say like a Ducati 916. you know, is now my favorite transportation thing, right?