Under the concept of “you can only fear what you don’t understand,” I’m starting a blog series to discuss AI, particularly Large Language Models (LLMs), often referred to as “ChatGPT.” Join me as we explore what LLMs are, how they function, and whether they should be feared.
Large Language Models are heralded as the “invention of the century” and a milestone towards Artificial General Intelligence. While I don’t fully agree with this label, I acknowledge their importance in propelling us towards the future. Many use ChatGPT without grasping its underlying mechanics, hence my decision to write about it.
We’re forging a vast gap between those who comprehend Technology & AI and those who don’t, much like the financial literacy disparity. Through these posts, I aim to demystify AI in layman’s terms for entrepreneurs, curious individuals, and anyone eager to learn. My approach remains human-centric, avoiding the use of LLMs to draft this content, rather conveying pure ideas as I would in lectures.
One common inquiry I hear is whether AI will outshine humanity. I believe understanding diminishes fear, but AI is often termed “Black Boxes” for a reason. We input data and observe outputs, but the intermediate processes remain obscure. For instance, though we know “1 + 1 is 3” is incorrect, the logic behind certain AI conclusions can elude us unless enough users accept the outcome.
In the coming weeks, I’ll delve deeper into AI, LLMs, their functions, and agents, advocating for comprehension over fear. By democratizing understanding of AI, we unlock collaborative potential to refine models and technology. Interested in participating? Follow us on Medium or connect on LinkedIn to deepen your understanding of AI’s capabilities.
Now for the main content: when you type messages on platforms like WhatsApp, you might encounter word suggestions derived from trigrams, using three words to predict the next one. Despite the basic function, the technology hinted at a larger potential, realized when Google’s “Attention is all you need” paper revealed the importance of context for understanding word meanings.
Some ambitious individuals, backed by significant investments, managed to train a model on comprehensive internet data, leading to the creation of LLMs as we know them today. It wasn’t until they integrated LLMs into a chat interface that their potential became widely recognized.
Most LLM users today engage them for fundamental text generation tasks, yet their potential extends to executing actions by choosing from various tools, like picking the hammer over a screwdriver. This ability to act on information suggests infinite possibilities, such as using AI in investment strategies.
LLMs are trained on the extensive internet, utilizing projects like Common Crawl, which collects web data globally. While public data is exhausted for training, companies are striving to source more private data, even incurring significant costs to acquire and apply it.
Using APIs is a crucial aspect to note. Despite free access, users inadvertently pay by sharing personal data, improving models over time. To assure privacy, it’s recommended to opt for cloud providers like AWS, which add layers of data protection.
Are LLMs the future? I believe they’re integral, albeit not the sole solution. Our team’s extensive AI and LLM usage highlights their impact, and we plan to share our insights and experiences in upcoming posts.
Armed with a foundational understanding of LLMs, we can now explore advanced features like function calling or agent mode, moving toward AGI, a vision embraced by industry leaders. If you appreciate this content, show your support to guide our focus on future posts about AI, data, and innovations.