Likhwa Tech

The paper Reimagining Artificial General Intelligence: A Human-Centric Approach presents a new vision for the development of AGI systems. In this paper, I propose a redefinition of Artificial General Intelligence (AGI) as: a system capable of independently understanding, learning, and performing a wide range of intellectual tasks that a human can, while proactively interacting with humans in a natural, adaptive, and context-aware manner, across multiple modalities and environments. This summary provides an overview of the paper's key insights and takeaways The pursuit of Artificial General Intelligence (AGI) has long been driven by a singular question: what does it mean to create truly intelligent machines? As we stand at the threshold of a new era in AI development, it's clear that the answer to this question will have far-reaching implications for the future of human-AI collaboration. This paper proposes a human-centric approach to AGI, prioritizing interaction, adaptability...

Why We Need a Human-Centric Redefinition of Artificial General Intelligence A few weeks ago, I started working on Artificial General Intelligence (AGI). At the time, I wasn’t entirely sure where the road would lead—I was just curious. AGI, after all, is considered the next frontier in AI, something that goes beyond what we see in current systems. However, unlike previous technological advancements, I didn’t want to be a spectator in that future; I wanted to be involved, to be an active player—an architect—helping to shape what that future would look like. AGI promises to be something that can think, learn, and adapt across a broad range of domains—more like a human mind, as opposed to the narrow, task-specific AI we’re familiar with today. But as I dug deeper into AGI, I realized something important: as things stand, no one really knows what AGI will look like. To truly build it, we need to start with the basics—a vision. Most definitions of AGI out there focus on the idea of mimic...

The Self-Taught Developer's Dilemma: Overcoming Knowledge Gaps As most of you know, I'm a self-taught developer. I always mention this- not to brag, but rather as a disclaimer, and maybe, just maybe... to earn a little bit of sympathy for my shortcomings. Driven by curiosity, I've acquired most of my knowledge through online courses, tutorials, articles, and a lot of trial and error. Whenever I delve into a subject, I take a hands-on approach: breaking it down into its smallest parts, building from scratch, and trying to learn as much as I can about it. However, I recently had a humbling experience that made me appreciate the importance of formal, or at least structured, education. I was watching a video where Mark Zuckerberg discussed their LLaMA model, and at some point he mentioned 4.5 billion parameters. I know what parameters are, so I had an idea what he meant, but much of it sounded like Greek to me. Like, I can understand say, 100 or even a thousand parameters, but...

Sometime last year (2023), when the ChatGPT buzz was at its peak, I tried to visit the OpenAI site to check it out. Unfortunately, I couldn't access it due to location restrictions. Frustrated, I decided to build my own chatbot. I figured I'd focus on local languages, starting with my mother tongue. This sparked my curiosity about how such platforms work. I began researching and soon realized I needed a large dataset to train a model. However, there's a scarcity of isiNdebele content online. Faced with this challenge and my constraints, I decided to take a different approach. From the onset, I knew that, as impressive as current chatbots and AI tools are, they represent just the beginning of the AI race. These tools are currently at their worst and can only improve over time. The future direction is uncertain, and I believe no one truly knows where it’s headed. With my foot in the game, even on the smallest scale, I figured i could reimagine the future, create a vision of i...