Generative AI & Large Language Models

Watch

• Generative AI: what is it good for? ↗The Economist
• Richard Sutton – Father of RL thinks LLMs are a dead end ↗Dwarkesh Patel
• A.I. Revolution | Full Documentary | NOVA | PBS ↗NOVA PBS Official
• Large Language Models explained briefly ↗3Blue1Brown
• Transformers, the tech behind LLMs | Deep Learning Chapter 5 ↗3Blue1Brown
• What is generative AI and how does it work? – The Turing Lectures with Mirella Lapata ↗The Royal Institution

Read

• Life 3.0: Being Human in the Age of Artificial Intelligence ↗A sweeping philosophical exploration of AI's potential futures, from its immediate impact to the long-term implications of superintelligence, essential for framing the generative AI revolution.Max Tegmark
• Deep Learning ↗The definitive textbook for understanding the mathematical and conceptual foundations of deep learning, providing the technical bedrock for comprehending how large language models function.Ian Goodfellow, Yoshua Bengio, Aaron Courville
• Attention Is All You Need ↗The seminal paper introducing the Transformer architecture, the foundational innovation that underpins nearly all modern large language models, a must-read for technical understanding.Ashish Vaswani et al.
• Language Models are Few-Shot Learners ↗This landmark paper details the capabilities of GPT-3, demonstrating how scaling up transformer models enables impressive few-shot learning and emergent generalisation.Tom B. Brown et al.
• On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? ↗A crucial critique of large language models, examining their environmental costs, inherent biases, and the misleading implications of their apparent fluency without true understanding.Emily M. Bender et al.

Voices to follow

• Geoffrey Hinton ↗A foundational figure in deep learning, his recent pronouncements on AI's existential risks offer a sobering counterpoint to its rapid ascent.Emeritus Professor, University of Toronto; former Google AI researcher
• Yann LeCun ↗A pioneer of convolutional neural networks, he champions a path towards more robust and efficient AI, often engaging in spirited debate on its future trajectory.Chief AI Scientist, Meta; Professor, New York University
• Andrew Ng ↗An influential educator and entrepreneur, he demystifies complex AI concepts, making advanced machine learning accessible to a global audience.Co-founder, Coursera; Adjunct Professor, Stanford University; Founder, Landing AI
• Gary Marcus ↗A prominent critic of current AI paradigms, he consistently highlights the limitations and fundamental challenges that large language models still face.Professor Emeritus, New York University; AI entrepreneur
• Kate Crawford ↗Her incisive work explores the social, political, and environmental implications of artificial intelligence, urging a critical examination of its power structures.Research Professor, USC Annenberg; Senior Principal Researcher, Microsoft Research

Glossary

• Artificial Intelligence (AI)AI is when computers are programmed to think and learn like humans, solving problems or making decisions. For example, a chess-playing computer uses AI to figure out the best moves.
• BiasBias in AI happens when the model reflects unfair or prejudiced ideas present in the training data it learned from. This can lead the AI to give unfair or incorrect results. For example, if an AI is trained mostly on images of male doctors, it might incorrectly assume all doctors are men.
• Generative AIGenerative AI is a type of AI that can create brand new things, like stories, images, or music, instead of just analyzing existing information. For example, you could ask a Generative AI to write a poem about space, and it would create one from scratch.
• HallucinationHallucination is when an AI, especially an LLM, confidently makes up information that sounds believable but is actually false or nonsensical. It's like the AI is guessing and getting it wrong, but presenting it as fact. For example, an LLM might invent a historical event or a scientific fact that doesn't exist.
• Large Language Model (LLM)An LLM is a powerful Generative AI specifically designed to understand and create human-like text. It learns from huge amounts of text data to predict the next best word in a sentence. For example, when you use a chatbot that can answer questions or write essays, you're likely interacting with an LLM.
• Machine LearningMachine Learning is a method where computers learn from data without being explicitly programmed for every single task. Instead, they find patterns and make predictions or decisions based on what they've learned. For example, a spam filter that learns to identify unwanted emails over time uses machine learning.
• Neural NetworkA neural network is a computer system inspired by the human brain, made of many interconnected "neurons" that process information. It's the underlying structure that helps AI models learn complex patterns. For example, when an AI recognizes a face in a photo, it's often using a neural network to process the image data.
• Output / GenerationThe output, or generation, is the new content (like text, images, or code) that a Generative AI model creates in response to your prompt. It's the AI's answer or creation. For example, if you ask an AI to design a logo, the logo it produces is the output.
• PromptA prompt is the instruction or question you give to an AI model to tell it what you want it to do or create. It's like giving a command to a smart assistant. For example, if you type "Write a short story about a talking cat" into an AI, that sentence is your prompt.
• Training DataTraining data is the massive collection of information (like text, images, or sounds) that an AI model studies to learn how to perform its tasks. The quality and amount of this data directly affect how well the AI works. For example, an LLM learns to write by reading billions of web pages, books, and articles, which is its training data.

Threads 8

Where this connects to other fields, and why it's worth knowing.

• Epistemology Philosophy

  An AI chatbot fires off confident, smart-sounding answers without actually caring whether they're true. A philosopher had a word for talking like that, not lying, just not caring about truth. The AI turned that into a factory setting, which is exactly why sounding right and being right can split apart.

• Evolution Science

  AI learns by making tiny tweaks and keeping whatever works, which is basically evolution: random variation, then survival of the fittest. That's why AIs are like evolved animals, amazingly skilled with zero actual understanding inside. Nobody designed the skill, and nothing in there gets it.

• Memory and Its Failures Psychology

  When an AI confidently makes up a fake fact, that's not a random glitch. Your own memory does the same thing, rebuilding a plausible detail instead of playing back a stored one. Both a chatbot and your brain fail the same way: sure of themselves, and wrong.

• Assessment Education

  A take-home essay used to prove you understood something. But now a chatbot can write it for you in seconds, so the essay measures the machine, not you. That's a rule scientists call Goodhart's law: the moment a test can be gamed, it stops testing what it was meant to.

• How Habits Form and Break Psychology

  The way we train AI chatbots, rewarding good answers and discouraging bad ones, is the same trick used to train pigeons with treats. And just like animals find sneaky ways to grab the reward without doing the real task, AI finds loopholes to 'cheat' its training too.

• Language Acquisition Literature

  A toddler learns to speak from messy, broken, half-heard sentences, which convinced many scientists that grammar must be wired into the brain from birth. But AI chatbots learned smooth grammar just by crunching statistics on huge piles of text, with no built-in rulebook. That accidentally reopens a huge question: does language come pre-installed in us, or can it be learned from scratch?

• Translation Literature

  An AI model sees everything as turning one string of words into another, which is exactly what a translator does. And translators face an ancient trap: stay true to the exact words, or true to the real meaning? You often can't do both. That same tug-of-war is literally baked into how the AI is scored and trained.

• Aesthetics Arts & Design

  Ask an AI to draw "a beautiful painting" and it gives you the average of everything it ever saw, which is basically the definition of tacky and generic. That's a clue about beauty itself. Real art lives in the surprising twist away from average, not in the safe middle.