Latest / Practice Ground / Richmond Alake: From Lagos to Oracle, The AI Engineer Building the Memory Layer of the Future | Ep11

Transcript

Richmond Alake: From Lagos to Oracle, The AI Engineer Building the Memory Layer of the Future | Ep11 · Practice Ground

• Nifemi Aluko: Go. Hey Richmond Nalake. nice to have you on my podcast today. thank you for joining in. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Thank you for having me and it's it's a pleasure for me to be here and to have a very good conversation. Nifemi Aluko: For sure, for sure, for sure. So, you know, just you know, to give you some background in case people don't know you, you're AI stack engineer, a strategist, an educator. currently you're director at of AI developer experience at Oracle. but before that, how I actually know you is you used to be at MongoDB and you taught taught a course while I was at Deep Learning. about prompt compression, query optimization. so that's that's the initial place kind of I met you. and you talk a lot about agent memory. would love to talk a lot about that with you. but yeah, let's go back a little back to July 2024. You launched a course with Deep Learning Andrewing prompt compression while I was out there. How did you get to this point? How did you go from, you know, that that that that boy that was born in Lagos, right? And then teaching a course at deep learning and doing all this AI, great great stuff in AI. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Yeah, yeah, from Lagos to UK to to power auto, right? No, it it's it's it's it's a pleasure anyway. so I guess I could education is typically the a good place to start. So I was born in Nigeria or Lagos and I moved to the UK, London when was ten. so I've been schooling in London ever since. Nifemi Aluko: Yep. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: So I went to university in London, just on the outskirts of London, and I studied software engineering for my undergrad. And right after that, I went into work as a web developer. So I built a bunch of websites for different companies. I worked in small companies to startups to large companies. I ended my three year career as a web developer working as a contractor for various different financial institutions. It was fun as being a web developer, but I very quickly got a bit bored. So I decided to to look into the space and see where I can transfer some of the skills I had. And machine learning was what I picked, because it you have a bit of coding into it and and so is I I got a bit of transferable skills. I I remember buying a textbook. Let me see. So I bought this book and blew the dust off it. Machine Learning a Probabilistic Approach by Kevin P. Murphy. And if you just open this book, right? It's just filled with a bunch of equations that when I opened it didn't make any sense to me. So I decided maybe I'm gonna go to university to do this. So I went to Nifemi Aluko: Yeah, yep, yep. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: University to do my masters in computer vision, deep learning and and robotics, which essentially is AI today, right? It's what you can class as AI today. I think that actual course has been renamed to artificial intelligence. Master Artificial Intelligence as opposed to that long name. But yeah, so I studied AI at the master's level after doing a fruit free year as a web developer. Then after my master's, I worked Nifemi Aluko: And who? Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: As a computer vision engineer at a startup for a game startup, then did that for around two years. Then I went to go work in a consultancy company as a machine learning architect, where I was put into some customers. and I learned a lot in being a computer vision engineer in a startup, I learned a lot about speed and wearing many hats. Then being a a machine learning architect at a consultancy, I learned about client engagements, I learned about communication. And then I then I decided and I was brought on to MongoDB as a developer advocate and it was almost a switch from very technical roles to more to advocacy. And the reason why I got reached out to by MongoDB was because in my career, even during my masters, there is one thing I did consistently and that is just learning public. Nifemi Aluko: Yeah. Mm-hmm. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: So when I was after my lectures, I was writing on my lecture notes and I was putting them out as medium articles. Exactly. So I was watching a lot of videos on machine learning and watched a lot of Andrew Ng's video as well on Coursera. and and the key thing was the couple things I learned, which is if you can't explain something as simply as possible, then you don't understand it. And writing helped me really Nifemi Aluko: Got it. Mm-hmm. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: really understand what I was learning in my lectures in my own and in a way that I could communicate it to people. So I I wrote a lot on medium. I have over 200 articles on Medium. I wrote for NVIDIA as well. In other I wrote for NVIDIA technical blog. I wrote for Neptune AI. I wrote for a bunch of different AI companies at the time. And I took courses as well because again, writing was just a method of learning. Now Nifemi Aluko: What happened? Yeah. Mm-hmm. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: And teaching was another map. I I was teaching a lot, I taught at Imperial College for the business school. I taught on O'Reilly. I did a bunch of courses on on O'Reilly. and yeah, so I wrote and I taught and I built as well. I'll put there. Do you have any questions? Nifemi Aluko: Interesting. Yeah. Yeah, yeah, yeah. I mean, I have several questions. I think, you know, as you're talking, I'm just curious, like, like where did the spark come from? Like a lot of people like go to school and I was talking to my professor, right, when I was doing my masters. I was like, you know, when I was doing my masters. I'll do research. I'll take it back to my professor. I feel like, okay, I found the answer. Like, let's start publishing. And the guy's like, cool, cool. Look at me. I wonder if we can investigate something else. And I'm like, bruh. then we know we go down some other thing and we just kept going. And I told him then I was just I was just 20 or 21 then. So I was telling him, I was reflecting. I like, look, at that point I was trying to graduate. So every time you said less investigate something else, I was like, let's just publish, right? and he was laughing about it. But he also said, you know, there's that fine line between staying curious and getting stuff done, right? Because I learned from him this like curiosity, which really helped me down the line. But also the school system is like, yo, you need an output, you need that degree, right? You know how also your Nigerian, your Nigerian size kits kicks in, is like get that degree, right? When for you, when did you this curiosity to learn, teach, share your stuff on Medium, where that come from? Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Yeah. I I'll be very honest, right? And I don't think I've ever said this before, but because we're both Nigerian, you probably understand what what I mean. In Nigeria, in the schooling system, you're ranked. it it the same as most places but one thing is very important is nigeria we we nigerians we place excellence at a high pedestal and you're not celebrated unless you're number one not number two if you're number two you might as well have have been number two hundred so in my childhood right growing up in Nigeria Nifemi Aluko: Yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: I was not celebrated unless I was number one. So I I could be number five and it could have been a hard work. And my mom would say this, and I'm sure you probably have experiences. The person that got number one, do they have two heads? You know, like and so l let me let me bring it to today. And what that did is, and there's obviously a negative aspect to that, and that's Nifemi Aluko: Two heads, yep. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: bringing but I take the positive from that which basically means that what my parents and my teachers in Nigeria at the time was trying to tell me was Richmond you are capable of anything anyone can do you just have to put the time and effort and you can be one of the best if not the best so don't stop trying don't stop being curious and just work hard and that means that I just subconsciously Nifemi Aluko: Mm-hmm. Yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Whatever I apply myself to be, I don't celebrate myself until I am considered like one of the best, or at least amongst the best. it's fortunate and unfortunately I'm doing that in AI, which is it's it's a lifelong it's a lifelong a journey of just being a a constant learner. You you if you Nifemi Aluko: Mm-hmm. Yeah. Journey. Yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Who signed up to just keep learning forever. and that's why when I was a web developer, I got bored, like I mentioned before, because I pretty much all what I pretty much knew all what I needed to be successful. I was earning a decent salary at the time for my age, and it and it was when it was in the six figures, right? And I was very young, and I was around a bunch of people that were in their 40s or late 50s, and Nifemi Aluko: Yeah. Mm-hmm. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: And I was in my twenties and I like, This this is not it, right? So hence why I got older went back to university. Nifemi Aluko: had this yeah yeah yeah dude I had the same experience so when I finished my master's I was 21 and the I started at Rockwell Automation and I think the next person in my engineering group was I think either 39 or 40, right? And everybody else was like in their 50s. So they're talking about they have kids. So we went for lunch I'm sitting there and just around these older folks which had its benefits because what I learned from them was so important. Like they were because we were y I was young, they were just willing to like teach and just pass on their knowledge, right? Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: I'll tell you one thing I remember you you just sparked a memory. I used to spend lunch alone because I couldn't even relate to these people. Like I didn't like it. so I just I remember I just used to spend lunch alone, but when it came to work, they're like, it it it was actually quite wild because this was back in a time in in the front end world, Angular JS and and was what what people were using and Angular was relatively new. Nifemi Aluko: mm-hmm. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: And I was the seen as the up to date front end and I was doing a bump bunch of migration from jQuery into using Angula and all of those front end framework. And and I remember I got to the point where I became the person people always come to for questions and I just thought, yeah, this this can't be it. let's pick up machine learning and scale but Nifemi Aluko: Yeah, right. Mm-hmm. Yeah, yeah. What's the next yeah? Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: curiosity, just to answer your question succinctly, it's for me that curiosity stems from trying to always be the best. And I've chosen a field where I I'll be very honest, I don't think I can ever feel like I am the best. Just so many geniuses in this space like I'm in and geniuses in different dimensions. We have very great educators, very great entrepreneurs, very great investors i in in the AI space. So Nifemi Aluko: Yeah. There's so many smart. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: It it's it's it's an interesting field that I don't think I'll get bored on. Nifemi Aluko: Yeah, I d you know, it reminds me of like when I went to Stanford I was like, yo, there's some smart people. You know when you're like you're around smart people, then then you're around like smart, smart people and you're just like, How? Like I remember one of my classmates, he went to Stanford like undergrad and he did mechanical engineering, then he became a doctor, then he was doing his MBA, right? So this is like he's an MD. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Yeah. Nifemi Aluko: Engineer MD now is doing his MBA and we're doing like a homework together and it was like, man, when he was in undergrad, he was humbled when the athletes at Stanford were having like a 4.0 GPA. So this that's like the athlete that is like on TV playing, still coming back, killing it in school. So this is like yeah, this place where it's just like is inspiring. It could be intimidating, you know, is like this combination. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: No very quick word this is it. It's in an environment like that and environment like the people that I surround myself with, excellence is average. So it's it it when you surround yourself with people where excellence is the average, the next level is just Nifemi Aluko: You don't even realize you don't realize that next level because you're just pushing. yeah. So Richmond, so let me let's talk, let's switch gears a bit into like stuff, something that you're interested in. Agents, agent memory, right? And you know, bringing it back to even human creativity, storytelling. I think like one of the Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Well I said. Nifemi Aluko: big things for humans, just through my research and my writing, is our ability to time travel, right? I don't know if a lot of animals have this ability, except for humans. We can go into the past, we could go into the future, we can think about different things, imagine, right? So it's a huge part of making us human, time travel. And time travel is possible because of memory, right? you can remember things, you can anticipate things, all that kind of stuff. So why is it why do you think why is it important for an agent to to have memory, in your opinion? Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Yeah, I I I'll take this. we can start at a basic, right? Or at least let's start for some fundamental objective, which is software is very quickly becoming agentic. And we can all agree, or we've agreed as an industry, that soft agent agents will be the primary will be the primary form factor for software going forward, which means that everything is gonna be an agent. The same way once upon a time, apps were not even a thing. then everything became an app. Every company had to have an app. Like you didn't exist if you didn't have an app. But for the app, there was a website. Every company had a website, right? So websites, mobile application, then agentic, right? That's the progression. And we're saying every software is gonna be agentic and everyone is going to have access to intelligent entities at their pocket, wherever it might be. Nifemi Aluko: Hmm? Yep. Uh-huh. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Where they can ask it any questions and it can complete tasks for them. Essentially, this intelligent entity becomes a part of everyday life of the user, right? And if we think about intelligent entities or even the promise of AI, which is the promise of AI is for us to create a computational entity that can surpass humans at most tasks. Then AGI is a computational entity, an artificial intelligence that can surpass humans at all tasks. So if we're trying to go down the path of humans, like you said, humans have memory, it only makes sense that if we look to biology, we take inspirations from it and then give an inspiration of what's made humans successful to computational entities. And this is not alien. Nifemi Aluko: Mm-hmm. Okay. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: To the way we approach technology in the sense that we today we can actually move across continents using plates, vehicles of flight. but in the initial days when people were trying to create vehicles of flight, like planes, aircraft, helicopters, they looked to birds. And they created vehicles where you had to like flap, flap right. Nifemi Aluko: Yeah, yeah, yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: We're using today is not flapping its wings, right? But we use it as inspiration to get us somewhere that surpassed nature itself in some form, right? Maybe not as efficient, but in terms of pollution. That's what I mean by agent memory. It's looking into what humans have, drawing inspiration from it, and using that as the seed to how we start to build reliable. Nifemi Aluko: Yep. Yep. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Believable and capable agents. Nifemi Aluko: Yeah, yeah, yeah. And and Richmond for you, what what would be the from a philosophical s sense, what would be the point of that, right? What would be the point of agents? Don't get me wrong, I I see the value of using it because I used I use it, but just from a broad sense, like what would be the point of An agent that can remember stuff that is close to human. I know the whole AGI argument, but I want to hear from your perspective. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: What would be the point of it, right? So I'll be very I'll be very transparent. I almost forgot that I had this call with you because it's Friday. Unfortunately, in my line of work, Friday means nothing. So Friday might as well be Monday. So bunch of meetings, bunch of stuff, and and this conversation we're having is not on my work calendar. Nifemi Aluko: Yeah. That's true. Yep. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: What would be the point of imagine if I had an agent that was my personal assistant that could just remind me five minutes beforehand and not just remind me, it says it's able to realize that two hours beforehand I'm not gonna have enough break for me to go make my food, my lunch, then it orders lunch for me and it comes down and I pick it up and then it it it literally puts me into this meeting. The agent remembers Nifemi Aluko: You Yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: would have to remember what I like for my food, would have to remember my schedule, it would have to remember other things about my day to be able to then really make me as efficient, as productive as I can. The point is technology is meant to at least the promise of technology is meant to free humans from from the I don't want to say work, because that is a from what we can describe as work. Nifemi Aluko: I think. Yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: And allow us to do more meaningful work. Nifemi Aluko: Mm-hmm. Mm-hmm. Yeah. And and Richmond, for you, what would that like meaningful work look like? Right. So let's say you're you're this stuff is remembering then let me take it back. I I remember I surveyed a whole bunch of people, right? Maybe this was twenty eighteen, about like creativity, what would you like to do and leave a more balanced purposeful life, right? And the biggest challenge, people want to do stuff. A lot of them have jobs. The biggest challenge was having the time to do it, right? And I remember someone write this was writing you know, in like the other tab, like, you know, type it out. The person said, I was like, what's the ideal solution? So people were like, an app that could do this or a community, blah, blah, blah. And someone just said, A clone of me, right? That could do other things, right? Okay. So I'm imagining like an agent that is like a clone that remembers things, remembers your family members' birthdays and all that stuff, tells you call your mom, tells you, you know, order this for this. If you have this thing that is doing all that, what would you do? How would you find what would you be doing with your time? Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Could be. Yeah. So in the spirit of honesty, right, because I chose this path, I don't feel like I've been working ever since I decided to go and do my master's degree. Right? I don't to me, what I this is fun. You know, writing about AI is fun, teaching it is fun, learn I I genuinely get excited in in ways that is unimaginable, right? So Nifemi Aluko: Mm-hmm. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Work is fun for me, which is very cliche and in some way very sad. But let me let me so what does that look like? So because I'm I'm a builder, I actually have tools that I build that embody what I think. in in in my kind of like practice of writing and learning. I've built tools that helps me op that helps optimize my learning that leverages technology today that we see in ages. So I have a tool called Open Speech. it is actually in production. So you can go to openspeech.io and I built a learning partner where it allows me to learn very quickly. So I can give this tool a bunch of papers and resources and I can chat with it and it can summarize everything for me. I can highlight and make notes and it will remember my highlights and notes and use that whenever I'm conversating with it. And just recently. Nifemi Aluko: Mm-hmm. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: I I I do a lot of writing, but I don't know if you do this when you write. Sometimes I write and when I'm editing, I make notes to myself, like in comments in Google Doc. Yeah. I'm like, when I'm editing, I'm like, okay, we write this in this. But one thing I realized I realized is wait, I can just delegate this to an agent that understands the seed material I'm using and also my writing style. Nifemi Aluko: for sure. That's yep. Research this, come back to this. Yep. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: And rather than me trying to resolve the comment, I could the agent could just do it. So I built this over a couple weekends ago and it's live in production today where I'm writing, but as I'm writing, I'm I I can edit and I just highlight it and I write a comment and I have a button that the agent can just then resolve the comment the way I would want it to. Nifemi Aluko: Really that's cool. I need to see that. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Yeah, open open IO, I'll maybe when this gets published it's still live. But the key thing is this, right? The way I see the way I utilize technology is the way I understand technology is by utilizing it in my own day-to-day life and understanding how I implement it, how I use it, how I market it, because my primary customers are AI developers. So I actually have to be an AI developer. It's this is just an extension. Nifemi Aluko: I'm curious. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: of the initial curiosity that I have that drove me to writing, it drove me to teaching, it drove me to to speaking. Now it's just driving me to build in public and really to understand this technology, apply it to my own life and process. Nifemi Aluko: Yeah. Mm-hmm. Yeah, yeah, yeah. Do you ever think that the the clone Richmond, if ever we get there, very close, the one that is resolving all this buzz, that's doing all this stuff, do you think he can ever become more curious than you? Right? Is that because if if it's curiosity surpasses yours, because I feel like the curiosity is like the the thing that's pushing the horizon. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Yeah. Nifemi Aluko: at least for you. That's what if something surpasses your curiosity, do you does it take anything away from me? I'm just cu curious. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: So to to be honest, curiosity is is is is such a broad term and I I don't even know how you would even define define it properly in the sense that curiosity you can build curiosity. So when when I was in and and we can to go back and forth on this, when I was doing my masters, we had a robotics module and Where and in one of this particular module, we worked in a virtual environment. It's called SLAM. And you basically program a robot to ex explore different paths. And your job is to create a path planning algorithm, right? And the I if I recall correctly, the tool was called SLAM. And if you define curiosity of ex exploring your environment and then maybe double clicking into interesting things or analyzing different interesting things, then we can engineer that in computational system. What I'm trying to say is curiosity is not just something unique to humans in a sense. So will if I was to build a clone of myself, will it be more curious than me? Maybe. But will it be curious in the sense that a human is? I don't think so. I think it'll be curious in the way that a computational entity is. Humans are driven by curiosity, which can be influenced by other humans and other emotions and things in our environment. Exactly. There's so many influence. Right? So we have a form of curiosity that is driven by that's why I was saying it's so hard to define curiosity and in the sense of how we're going to apply it because Nifemi Aluko: Yep. No. Culture, yeah, culture cultural, yep. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Robots can be curious when you buy a robot vacuum cute cleaner. It's curious, right? Exactly. Curiosity, right? So it's just more focus and specialized curiosity. Are we gonna build and and and and we mankind sent out spacecraft into space decades ago, what is it called? Voyager, I think is what it's called, and it was meant to explore space. That that is Nifemi Aluko: Yeah, yeah. Around trying to figure out like where where the trash is. Yeah. Mm-hmm. Mm-hmm. Yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Curiosity far beyond our own reach. So I will say curiosity i is is not human bound and it's different. There's different ways to look at it on wherever Nifemi Aluko: Do? Yeah, yeah, yeah, yeah, yeah. And and and and in a sense that's you know, that's sort of what this reinforcements learning is, right? Where it's like, okay, you give me a reward for so I'm just gonna figure out how to get more rewards, right? And less of right. So can you give us like a you know, for the most simple person in getting into the space, how do you think about memory per Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Mm. Yeah. Nifemi Aluko: Primarily, right? I think I saw you on a podcast talking about episodic, semantic, and procedural, memory. can you just give us like just broad overview of those things and then from an AI perspective, we've gone from prompt engineering to context engineering. Where do you think we're going in the future? Yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Yeah. so memory is very in terms of explaining the simplest form, we have long term and short term memory. So that's the most easiest way you can think about it. I could tell you my phone number right now, and if you don't write it down, you're not gonna remember it. That's your short term memory. but if I tell you if you write it down or you repeat it to yourself several times, in the when I tell you, it would then go into your long term memory. So that's the easiest way you can think of things. Nifemi Aluko: Mm-hmm. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: but you can dive a bit deeper because the brain is more complicated than that, it's more complex. we have different parts of our brain responsible for different types of memory. So humans have humans have episodic memory, which is basically any event that basically has some form of temporal substance to it. Like your birthday. Your birthday is an episodic memory, or maybe your birthday last year, what you did there, that's episodic memory. Nifemi Aluko: Mm-hmm. Yep. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Humans have semantic memory. This is memory that has an understanding of your environment, knowledge of your environment. Humans have entity memory. We have a memory where it's special is is it's focused on recording the people we meet or the things we meet based on their characteristics. So entity memory. Humans have we have some we have a part of our brain called the cere cere cerebellum, which stores Nifemi Aluko: Mm-hmm. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: information of our routines and skills. so that's stored in that part of our brain. And that will be some form of procedural memory for humans. And all of this can be represented in computational form in code. Right? We can implement these types of memory the same way people look to birds and they start we can look to humans and figure out okay this is how this part of memory works. Nifemi Aluko: Mm-hmm. Yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: How can I implement something similar using code, essentially? And using the agent, the LLM that we have in a genetic system more for the reasoning capabilities as opposed to relying on the LLM to remember or even remember how to remember, right? We just front-load all of that with our own abstraction and understanding of how memory should work and intelligent entities. It my Nifemi Aluko: Yeah. Yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: current conceptualization or understanding of memory might not be the ideal one because just the same way planes don't flap their they don't flap it they don't flap the wings it but now we have an efficient way of flying is the same way we might not we we need to discover other ways of actually implementing age of memory that is more efficient and that's why I am where I am today. So I work in Oracle Nifemi Aluko: Mm-hmm. Exactly. Yeah. Wings. Yep. Mm-hmm. Yep. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: I rock within the database division, there is no better place to be in in in a world where most of the 14500 companies data is sitting on Oracle. So I understand the the the heterogeneous nature of data on the internet. I get to play with it. I get to understand how we're building technology to be able to handle and manage those data. Nifemi Aluko: Mm-hmm. Mm-hmm, mm-hmm, hmm? Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Now, what that means for me is I'm able to take this abstraction of agent memory and apply it to some of the features we have in our database capability within Oracle. And I'm having fun, right? I'm building a bunch of projects, I'm building a bunch of demos, and I'm experimenting and I'm always staying curious. Nifemi Aluko: Yeah. Mm. I I mean, is is clear like from your explanation, like experimentation, yeah, having fun, man. And that's a good thing, you know, when you meet people that are doing things that they enjoy. It's just like, I mean, you can't take that away. So as you're explaining this, I have a question. Is there a case for no memory? Right? The the reason I'm asking this question is so similar to you, I build stuff too, right? I build I built a tool that helps me with my writing. I don't really like the LLM doing the writing for me. I like to write because of the practice of putting me in this like flow state, right? I don't want anything to take that away. But I do enjoy it as an editor. So similar to what you're talking about, right? So I kind of built this tool where you know I write, I send it into like my AI editor. And I say do not rewrite it, just give me feedback. It gives me feedback, tells me what to look at, right? Which is great. So I did it, I was like, this tool is cool, nice. So I did it one time, it gave me feedback, I changed it, I did it again, gave me a whole set of different feedback, right? Now, when I used to do this in Claude, by the time I, you know, do it once or twice, Claude is like, great, you addressed the last thing, right? And there's this it has this memory and within two or three iterations is like you're good to go. It's perfect. You know. It k kinda gives me like this form of in a way like the human version of a confirmation bias. Like, you know, it's it's like never me might Sometimes the L and M's are nice, you know what I mean? Like they're just like, let me tell him what he needs to hear right now, so he, you know, he keeps coming back. When I built this thing, because I didn't include any form of like memory, every time it gave me something new, but it then I was just it made me think like, is this a more objective way of editing in a way? Because every time I'm seeing something new. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Yeah. Nifemi Aluko: And by the time I did it like five or six times, then it was like, Okay, you're good. So this is a question of, is there are there cases where you're like, now let's not use memory for this application. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Yes, that will be the easy answer. it will be very very niche use cases and and this is the the the this way of working is not strange because you go on chat GPT on the interface, you can just press a button. It's called temporary chat. So what you're saying is don't remember anything from this. So that's a use case. It's almost like you're telling a secret or something or maybe Nifemi Aluko: Mm-hmm. Yeah, yeah, yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Unfortunately, this this is would be this would be the perfect use case for for an intelligent entity without any memory. A therapist, right? maybe like a one-time therapist. You just want someone to talk to and you don't want the person to you just want an unbiased perspective. So this person's not gonna know you, they won't know your history, they're just gonna look at this objectively and give you Nifemi Aluko: Exactly. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: some feedback and I don't want to say stuff that is sensitive and you just want them to forget. So it could be a use case for this. really in your particular use case, right, you had a situation where you were tapping into an age an LLM without any form of that didn't have any state persistence. So it could have then you're pro you're just using the probabilistic Nifemi Aluko: Mm-hmm. Mm-hmm. Yep. Exactly. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: probabilistic functionality of the LLM to its fullest without actually providing it with prior interactions or information. So you would actually always be in this loop of what right there's always be there there is always going to be something that as long as you told the LLM to give me improvement. Or you could say give me improvement and you probably have to prompt it and say if there is nothing Nifemi Aluko: Yep. Those painful. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Then say there is nothing. Right. Nifemi Aluko: Say yeah yeah yeah. Dude I'll be honest, it was a painful experience because it always found something. In fact I changed I changed something, he told me to change, and then he gave me some feedback like, why is this thing? Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Yeah. Well the the LM is actually the you could say the LM is being Nigerian because it is it's looking at you like, yo, you could do better, like you could do very you could do well, you you need to go back. The LM is Nigerian. It's not gonna it's not gonna give you thumbs up until you're the best. So Nifemi Aluko: What my god. Yeah. Yeah, yeah, yeah. I'm like, bro, I've changed this thing seven times, right? And yeah, I guess that that then goes even against my my because to be honest, after I did it, I was just like, Okay, I need to improve this thing where at some point it stops with the Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Okay. Yeah. And this is the importance of memory, right? Because perfectionism is the is the enemy of progress. And the LM is gonna always keep you in that loop and try to give you something perfect. And you might never get there because it it it's obviously following a set of instructions based and it's only gonna perform as well as you instruct it. So look. Nifemi Aluko: Mm-hmm. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Memory is very important in most of the use cases we find in production or we find we want to replicate in human we want to replicate human in human life. memory is very important. It's why it's why a lot of software today always ask for your cookies or because they're trying to personalize the experience. imagine if imagine if every time you open WhatsApp, all your messages were deleted. Nifemi Aluko: Yeah. Yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: You're gonna be like, no. So if we go back to the paradigm, right? We we we had a bunch of during the internet age, we had a bunch of websites, right? And every time we visit your websites, it's almost like a given. We expect the website to maybe remember some of our preferences that we've selected, or at least remember that I like dark. Nifemi Aluko: Mm-hmm. Mm-hmm. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: That's stored in your browser, in some form of memory in your browser. Then we got into apps. So we wanted apps to be more personalized. So we wanted data from different apps to be used in in in another app we're using. So if we go back to the beginning where we said most software is gonna be agentic, it only makes sense following that paradigm that we're gonna need memory for most of our agents. So Nifemi Aluko: Have seven. Yep. Yeah. Yeah. Yeah. For sure. For sure. So let me switch into some more, you know, practices. And Richmond, I hope you ha still have like ten minutes. I know we're cutting it close, but do you have like ten more minutes? perfect, perfect. All right. So for you, how do you stay aligned with your taste in such a noisy environment? Like things are moving quick. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Yeah. I got ten more minutes, yeah. Nifemi Aluko: Right. And how do you stay aligned with your taste? Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: There's an illusion in the space of AI, which is everything is moving quick and there is a lot. I'll be very honest with you. You can pick anything in AI in any dimension. So remember when I said AI is so massive and you can operate in it in any dimension and try to be the best. The illusion of the space is it's very broad and It's very hard to keep up with what's going on. But I'll tell you a secret that is not really a secret. Pick your what what makes you interested. Pick something you find interesting and just go as deep as you want to go. Because AI does one thing, it converges. Right? It converges all use cases, all application, all thoughts, and whatever you want to do, it always converges. So what I'm trying to say is. I studied computer vision when I was in university, right? And language models were primarily when they came out, were only used for text-based data. But very quickly, they started to get applied to computer vision use cases and you had VLMs. Again, you could have stayed in computer vision and eventually the space would have converged into what was, I guess, Nifemi Aluko: Mm. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Everyone thinks is popular. And people like Andrew has stayed in education and is holding skills in that. And it's still very relevant. There people that have been that are investors in AI, and suddenly they're everywhere, but they're just investors, they're just VCs, but they understand their practice. So the illusion of the space is there's a lot going on. So for me, I chose agent memory, and I could argue. Nifemi Aluko: Mm-hmm. Yeah. Yeah. Mm-hmm. Stay, yep. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: that agent memory is everything. We can argue for or against it. But the same way if I was a VC in AI, I would argue that none of this wouldn't happen without investing. Nifemi Aluko: Mm. Mm-hmm. Without the Yeah, without the capital. Yep. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: This is what this is what I say to people, right? Pick whatever you're interested in and go as deep as you want to go, but have an open mind to the different to the other perspective that things can go and other areas within AI. Nifemi Aluko: Yeah. And you said something very important because I tell people at least what I find works for me is having like this kind of self directed project, right? Where as I said, I I built this AI editor thing because I like writing, right? So I'm I'm always b because I'm always writing, because I'm always trying to publish, I'm always looking at technology from a perspective of how does this impact this? So it help it easily helps me to be like, can I use blockchain for that? Or can I use AI for that? You understand what I'm saying? I'm always is is almost like an anchor to yeah, to new stuff. Just having that, you know, one lame Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Yeah. Yeah. It's it's exactly so w one thing in Oracle and w I'm at a vantage point where we work with a lot of different customers and the space is very noisy. But wha whatever the noise is, when you talk to me, I'm gonna bring it to database and agent memory. So you can talk about quantum computing, best believe I'm gonna find a way to bring it and databases and agent memory. Nifemi Aluko: Yeah. Yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Because yeah, i i it it's still applicable, right? And that's like you said. Nifemi Aluko: Mm-hmm. Mm-hmm. Yeah, yeah, yeah. So, you know, going a little deeper in terms of practices, right? When do you have any like go to things you do or rituals when things start to get blurry f like in terms of like what I'm doing, a project not working, or even like purpose. Do you have any practices you have where you're like Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Yeah. Yeah. Nifemi Aluko: used to center yourself to bring meaning back to your your life. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: So I'm I'm very bad at this, I'm I'm not going to lie, and I wanna be very honest to your audience, which is a lot of people they say that they are able to take breaks from their work by going on walks. I'm gonna be very honest. Even though it works, it's so hard for me to find a bug or try to implement a feature that is not working, it's very hard to pull myself away from it. If it's not working, because I've always like wait, I've not tried this, but wait, I've not tried this. And I'm always going so it's very hard before I know I went like five hours and I'm like and I've got in nowhere. And I'm I'm gonna be very honest. When I step away and then I go for a walk, or I just do something else, and I come back and I solve it in like five, ten minutes. It's it's no matter how much I know this, it's Nifemi Aluko: No. Yes, you know. Yo. Quick. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Anyone that can step away from something that they're doing when it's not working, that person has a very strong willpower. Different techniques like time boxing and like I only have 30 minutes to solve this. If I don't, I'm gonna move on to something else. And but I'm gonna be very honest, I I still fall prey to a feature that you should is meant to work in a certain way and it's not, and you're spending five hours debugging. It doesn't Nifemi Aluko: We'll file yeah. Yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: If I'm using LLMs, it doesn't matter it was five years ago and I'm there just scrolling through Stack Overflow. I'm still spending hours on this. Nifemi Aluko: Mm-hmm. Mm-hmm. Ours. Yeah. Yeah. Yeah. Yeah. Yeah. Yeah. I feel you, man. Does does writing how does writing play a part in your I know you use it for like teaching. Does do you use it for any like personal use? Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Yeah. do I use writing for anything personal use? I guess Because like I said, work is fun and fun is work for me. So I mean I I write a lot of strategy documents and I've I I noticed that the skills I gain from writing on Medium and all of these blogs is applicable to the things I have to do in my in my line of work, which is not just explain the strategy we're taking, but really include the why and provide a narrative that is compelling. It's almost A secret weapon for anyone that is in a in a role where they have to form strategy. Writing can be a very secret, like a secret weapon to forming a narrative. Same way speaking can be. Or same way if you're a graphic designer, you can do nice graphic design and make things look flashy, then you put the the dollar sign at the end of it. The the thing is this. Nifemi Aluko: Mm-hmm. Yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Writing is a secret weapon for me in a corporate environment that's primarily that is primarily driven by documents and strategy and communication by email. So that is even in the world of AI writing, there is something that you you can almost still like bring your own originality and your own perspective and your own structure. And it it it always I don't think i I don't want to say it's something I don't think an AI could replicate because it it probably can, but there is a way you want the reader on the other hand to feel a certain type of emotion or understand a certain type of emotion, and there is a certain way you can construct and order your sentence and space them out to really get a point across. Yeah. Nifemi Aluko: Yeah. Yeah. And and and I think yeah, you're talking about I mean we could talk about that on like for hours because that's my thing where it's like there's the communication aspect, there's the trying to get inspire people to take action when they've we're done on that last line or even just trying to like transfer some type of feeling. So even when I'm writing, the first thing I do is I write like okay, how do I want them to feel? Question mark. I answer that. What do I want him to do? Question mark. I answer that. Do I want him to reply? Buy something. You you understand what I'm saying? And there was the one thing I want him to remember. And as I'm writing through the documents, I'm always going back to that. Like, is it doing this? So that's why, you know, we initially when we started talking, I was like, for now, I don't get the LLMs to do the writing. I get it to do the other part, the editing, the analyzing, and then getting it back to me. And not putting me in the loop. So that's why do need the the memory. So Rishma, two two more questions just before we end here. when we started, I talked about you know, being at deep learning, right? And one of the things when while I was there was you know there are all these instructors coming in, showing people what they can do, showing people what they can build, and you were one of those people. And while I was there, I was like, man, most of the Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Okay. Nifemi Aluko: people that doing the education. I hardly see black people, African people, Nigerians. You were the one person. and I was there I launched about 40 something courses. And this is not anything to do primarily with deep learning, but it's just some some sometimes the space that we're in. Yeah, that is just the space that we're in. I I know for a fact I go to some places, I'm the only one That looks like me in that space, right? what importance do you you know, you talk to some leaders and they're just like, diversity, I just want the best minds. And I'm like, okay, that's cool. Yeah, but we can get into that, what that means. But what how do you see it? Like you you go to these spaces. I was talking to Tony not too long ago. We had a similar conversation. Like, what do you think? Is it important to have diversity in building these these technologies? Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Yeah, it would be an understatement to say that it is very important. It's important to have diversity in this technologies for several several reasons. Firstly, it's beneficial for the performance of the technology. First of all, you need to have a diverse you need to have a diverse data set because the end user of your products you don't know what you're gonna expect. So you need to have a diverse and this was a back in the day, this was a massive thing for a lot of face detection and computer vision products, right? Nifemi Aluko: Yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: but really i in terms of diversity is important end of right in terms of how you go about it it is is what really matters. Like how do you try to implement diversity and make sure you're getting the best, but also allowing for opportunities for the best that actually arise from the crowd and realize that not everyone has the same starting line. And sometimes it's more important to recognize potential, right? Nifemi Aluko: Mm-hmm. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Than to recognize end results. Right. So for example, if if I if I see a kid that only had two hours, like again, Nigerian back, I remember days where I had to pay to go to the internet cafe to go to the internet, right? If you tell me that there was a kid that was able to learn coding that only had access to 30 minutes of internet a day and was able to build something. Nifemi Aluko: Mm-hmm. Yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: And only had thirty minutes and it's on parts with someone that had All of the internet access. I realize that this kid has potential, right? So it's always good to as a leader, it's good to realize where people are the background where people are coming from. And given I I I do understand that some leaders might not have the time to really go into the detail of your life story. but it's important as a leader to realize that the there is people come from somewhere and Nifemi Aluko: Yeah, yeah, yeah, yeah. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: There are certain questions you can ask to be able to identify potential in people. and that's what we need to really, really make, really, really bring, bring to the forefront is just be able to understand the potential of someone and understand where they're coming from and use that to be a judge of how they can perform. So, and for me, fortunately or unfortunately, Nifemi Aluko: Mm-hmm. Mm, mm hmm. Dude. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: I've made a pat to myself that representation is I have no option but to represent, right? So I have to get I have to show people that people that look like us can actually get to top of companies or be very prolific and really have an impact. Because I I I I remember when I was writing back in the day, there was a senior leader at Google, his name is James Maniakov. And he's not he's not well known, to be honest. He's not he's not as visible. I remember like maybe five or six years ago, I just saw a black person in tech at Google. And I've never spoken to this guy. He doesn't write, he barely speaks. I just saw his picture and his title. And his title said maybe director or VP at the time at Google. And that representation just made it possible for me to Nifemi Aluko: Mm-hmm. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: realize that I myself can be that in a massive complex Nifemi Aluko: Yeah, yeah, yeah. Dude, man, I mean that it is so important, right? This whether it's, you know, at a company, whether it's seeing someone's name that looks like yours on a book, whether it's like a repo on GitHub, and you're like, this name sounds familiar. Like it just allows you to feel like, okay, I'm not this imposter that's trying to be in this thing that I enjoy. Because sometimes your your people Don't really like nerd out on the stuff that you're nerding out on, you know what mean? Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: I feel you. But I I think in general, there is representation, there is more people like us, but we can definitely do more. And I think we could definitely do more. Nifemi Aluko: Mm-hmm. Mm-hmm. All right, cool. Last one, and I'll get you out of here. This is a usually a fun question that I ask everybody. So we're building all these things, man. Memory. We're building out, you know, agents. We're moving, technology is moving at at a good pace, right? If the world went one of two directions. If it went dystopian, right? You know what dystopian is, right? Like those sci-fi movies. pr pretty much we have to humans, we have to serve the the the machine overlords, all that kind of stuff, right? That's one scenario. Second scenario is utopian. Utopian is where, you know, we've built like this trustworthy AI. It's here to serve us. Man, in fact, it's doing most of the work. Now we just have to figure out like what do we do with ourselves? You know, there's there's universal basic income. You know, everything is good. There's abundance. So you have to choose one of these two. Whether it's dystopian or utopian. Choose one scenario and then tell me what you do as a practice on a daily basis. Like what will life look like for Richmond? Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: it will be obviously the utopian dystopian. I don't know. i it it will be the one where we we're we're in a prosperous society and environment. And what what would I do? I'm open technology at the time is very advanced and there's space exploration and space is infinite, so you just like the robot that we sent exploring spaces, we can just Nifemi Aluko: All right, let's go. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: be space explorers and we are in a world where it's ever expanding, hopefully. So there is infinite possibilities for us to explore. And it's that simple. We can never fully explore this universe in our lifetime. Nifemi Aluko: Mm-hmm. Man, perfect. Perfect. Richmond, dude, thank you very, very much for your time with me. I really appreciate it. I'm looking forward to following your journey and just staying in touch. My Nigerian brother, even fact to found out to you know, my Elisha brother to even even go deeper. But very happy of proud of what you're doing. very happy to to always see your face. As I said, while I was at Deep Learning, I was like, Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Yeah. Nifemi Aluko: we're we're in there, you know I mean. So thanks for taking the time to to chat. Untitled recording Richmond Alake webcam 00h 00m 00s 371ms StreamYard: Welcome. It's been a pleasure, Nifami.