Large Language Model (LLMs)
What is LLM?
An LLM is a type of GenAI designed to produce human-like text. The LLM's primary function is to generate contextually appropriate continuations of a given input prompt by predicting the most probable sequence of words.
A simple way to understand this is through an example. Given the prompt:
- “The capital of the Philippines is ______”
The LLM predicts and fills in the most probably next word. This predictive mechanism enables LLMs to perform a wide range of natural language processing tasks such as summarization, translation, question answering, content generation, and dialogue simulation with a high degree of fluency and relevance.
Examples of LLMs include ChatGPT, Google Gemini, DeepSeek, Microsoft Copilot, and other advanced AI chatbots.
Common LLMs in the Market
ChatGPT (https://chatgpt.com/) - Launched in November 2022, ChatGPT is a conversational AI developed by OpenAI. It quickly became one of the most widely used LLMs globally, marking a major breakthrough in making large language models accessible to the general public. Its widespread adoption was also driven by its user-friendly interface and strong contextual understanding.
Google Gemini (https://gemini.google.com/) - Originally released as Bard in early 2023, Google rebranded its flagship LLM as Gemini in late 2023 to align with its broader AI model family. Gemini integrates deep language capabilities with Google’s massive knowledge graph and search infrastructure. It intends to facilitate seamless integration with Google Workspace, making it a strong contender in the AI assistant space.
Microsoft Copilot (https://copilot.microsoft.com/) - Microsoft Copilot is an AI assistant powered by OpenAI’s GPT models but fine-tuned for productivity and enterprise tasks. Introduced in 2023, its deep integration with Microsoft’s productivity suite has positioned it as a practical, task-focused implementation of LLM technology.
DeepSeek (https://chat.deepseek.com/) - DeepSeek is a newer entrant in the LLM landscape developed by the Chinese company DeepSeek AI. Released in 2024, it gained attention for its open-source model, which demonstrated strong performance in both English and Chinese benchmarks. DeepSeek offers robust capabilities for research, enterprise, and educational applications, contributing to the growing diversity of open LLM ecosystems.
Unveiling where the Magic begins
How can LLMs seem to perform tasks so effortlessly—almost like magic?
LLMs are trained on massive datasets (often from the Internet), including books, articles, websites, and user-generated content. They learn to understand language by recognizing statistical patterns between words, phrases, and sentences. The "magic" lies in this: the more data they see, the better they become at mimicking human-like responses. Furthermore, there are two key processes that contribute to their accuracy.
Key to LLM Accuracy:
#1: Pre-Training
Pre-training is the process where the model is shown partial text and asked to predict the next word.
In other words, AI (essentially a computer program) learns to predict the next word by calculating the probability distribution of all possible words. It does this by being exposed to billions of textual data and repeating the prediction for trillions of times, making it accurate over time.
Essentially, an LLM is an auto-complete on steroids!
#2: Fine-Tuning
LLMs then undergo the process of fine-tuning, which is using feedback from human reviewers. AI companies hire teams of human reviewers to rate the quality of the model’s responses.
This technique is also known as Reinforcement Learning from Human Feedback (RLHF) and is crucial to making AI safer, more accurate, and more useful in real-world settings.
Reviewers evaluate AI model responses and indicate which ones are better or worse. Their feedback is then used to fine-tune (improve) AI, so that it can give better answers when it encounters similar questions in the future.
AI Is Not Sentient
LLMs can sound intelligent, empathetic, or emotional, but they are not conscious beings. They do not understand in the way humans do, nor do they have awareness, goals, or feelings. Their outputs are generated through complex pattern-matching, not reasoning or intention.
So, do you still believe that AI—like ChatGPT—is sentient or has emotions? It's no different from saying that a probability distribution can feel something. In reality, it’s all just pattern prediction, not consciousness.
In summary:
- LLMs are not sentient.
- They operate using probability distributions.
- No internal beliefs, desires, or feelings exist within them.
Key takeaway: This is not a dangerous magic genie. This is statistics executed on an enormous scale.
Source:https://the-decoder.com/genai-is-just-advanced-automation-not-a-panacea-or-an-existential-threat-says-stephen-wolfram
Introduction to Prompt Engineering
A prompt is the input or instruction you give to an AI model (like ChatGPT) to guide its response. It can be a question, command, sentence, or even a set of guidelines that tells the AI what you want it to do.
Examples:
- Vague: "Write something about climate."
- Better: "Write a paragraph explaining how climate change affects farming in the Philippines.
Therefore, prompt engineering is the skill of writing clear and specific instructions to get better results from AI tools. Think of it as giving the AI better clues.
Simple tip: The more specific your request, the more useful the output!
For now, take a moment to reflect:
- What would you want an LLM to do for you?
- Do you already have a go-to prompt (perhaps your favorite prompt) that works well?
We’ll explore more advanced and/or practical uses of prompt engineering in the next module.
Comments