AI and large language models

What is AI and machine learning

Artificial Intelligence (AI) and machine learning are distinct concepts.

1. Machine learning is a subset of AI.
   

2. AI can be broadly categorized into two types: 2.1 Symbolic AI (which is model-driven and relies more on mathematical formalization), and 2.2 Non-symbolic AI (which is data-driven and includes machine learning).

3. AI can be described as the systematic development of machines capable of representing and interacting with knowledge about the world.

Machine learning

1. Machine learning is the study of techniques for identifying patterns in data.
   

2. Arthur Samuel defined it as "the field of study that gives computers the ability to learn without being explicitly programmed."

3. Since 2010, machine learning has gained prominence due to three main factors:

3.1 A significant increase in data (driven by the growth of e-commerce and the web)

3.2 Theoretical advancements (such as developments in deep learning)

3.3 A significant increase in computer processing power.

What are generative AI/ large language models (LLMs)/foundation models? How do these models work?

In recent years, large language models, foundation models, and generative AI have become increasingly prominent. These techniques can be applied to both image and text generation.

Specifically for LLMs

Large language models (LLMs) were initially developed for automated language translation.

These models are trained on extensive datasets.

The training data includes public content from the World Wide Web, such as blogs and social media platforms like Twitter (now X) and Reddit.

When given an input string (a word, a sentence, a paragraph), these models generate or predict the subsequent elements in the string (the next word, sentence, paragraph, etc.).

The inner workings of these models remain a mystery, even to the technical staff of the organizations that develop them.

The accuracy of these models improves with more training data.

The main applications of LLMs are:

• Answering search questions

• Assisting software developers

• Summarising documents or meetings

• Chatbots to answer queries

• Generating new text

What are the issues in using generative AI models?

• Large language models (LLMs) excel at generating words and sentences, making them suitable for text-based tasks and text-like digital objects, such as most computer code.

• Generating images requires different techniques, and some methods are not robust, as small pixel changes can significantly alter outcomes. Many image generators struggle with embedded text and fine details like fingers.

• Generative AI systems can inherit biases from their training data, potentially leading to biased outputs.

• LLMs currently lack proficiency in counting or numerical calculations, but specialized plugins are being developed to address this limitation.

• LLMs struggle with tasks like project planning and constructing chains of reasoning.

• Despite their impressive outputs, generative AI models lack understanding of their outputs, lacking a semantic layer to tie statements to reality.

• Generative AI models do not provide explanations for their outputs, presenting challenges in understanding their decision-making processes.