When we think of AI, the first thing that often comes to mind is generative AI, like chatbots or content generators. However, in a recent article titled “How AI Works” published on January 9, 2024, Nir Zicherman sheds light on the workings of large language models (LLMs) in a way that’s accessible to everyone, not just tech enthusiasts.

Zicherman compares LLMs to sophisticated auto-complete tools. Imagine typing a message, and your device suggests the next word based on what you’ve written before. These models are trained on extensive datasets, operating through learned patterns rather than possessing true understanding or consciousness. Simply put, they don’t “know” what they are talking about—they just analyze data to make educated guesses.

To illustrate this concept, Zicherman uses a relatable meal analogy. Picture yourself at a restaurant: you’re hungry, but you haven’t decided what to order. The waiter, lacking psychic abilities, can’t pinpoint exactly what you want; however, based on past orders and an understanding of the menu, they can suggest options that might satisfy your craving. Similarly, AI predicts the next word in a sequence based on patterns derived from vast amounts of data it has encountered. It’s important to recognize that while AI can make relevant suggestions, it does so without genuine comprehension.

In another insightful article, “What Is ChatGPT Doing … and Why Does It Work?” the author dives deeper into ChatGPT’s mechanisms. At its core, ChatGPT generates text by predicting what word or token should come next based on a massive pool of human-written examples. By analyzing the data, it calculates probabilities and constructs sentences one word at a time. To keep the text engaging and less repetitive, a degree of randomness is introduced, allowing the model to select words that may not always be the highest probability options. This is controlled by a parameter known as “temperature,” which significantly influences the quality and variability of the generated outputs.

In essence, ChatGPT creates human-like text by understanding and predicting word sequences with the help of randomness. This approach results in a rich and dynamic output, steering clear of flat, monotonous prose.

### Points of Confusion: The Power of Speech

One question that arises is how modern AI can take inspiration from and reinterpret some of history’s most impactful speeches—like Martin Luther King Jr.’s iconic “I Have a Dream.” For instance, let’s consider a line from his powerful speech: 

“Let freedom ring from the hilltops of New Hampshire. Let freedom ring from the mighty mountains of New York. Let freedom ring from the heightening Alleghenies of Pennsylvania.”

Rewritten with an AI’s touch, it could become something like:

“Allow freedom to be expressed from the elevated points of New Hampshire. Enable freedom to be acknowledged across the substantial mountain ranges of New York. Permit freedom to emerge from the prominent Alleghenies of Pennsylvania.”

This raises an interesting point about the line between genuine inspiration and mechanical rephrasing. While AI can generate variations of powerful text, the emotional weight and historical context behind the original words are something that it cannot replicate.

The exploration of AI’s capabilities is a fascinating journey, illustrating both its limitations and its potential. As we continue to interact with this technology, it’s essential to cultivate an understanding that while AI can craft sentences and suggest ideas, it does so without the consciousness or emotional depth that defines human creativity. Let’s embrace the advancements while acknowledging the art that remains uniquely human.