AI: History, Mechanics, & Future

Explain the Turing Test and its historical significance for defining intelligence.

Summarize the early symbolic ('Good Old-Fashioned AI') era and why rule-based systems ultimately failed.

Describe the two 'AI winters' and the factors that caused enthusiasm and funding to collapse.

Explain why the 2012 ImageNet competition is considered the start of the modern deep learning revolution.

Summarize the significance of the 2017 'Attention Is All You Need' paper and the Transformer architecture in plain language.

Trace the evolution from GPT-2 to ChatGPT and explain why ChatGPT's launch felt like an 'explosion' in public awareness.

Contrast 'traditional programming' (explicit rules) with 'machine learning' (learning from examples).

Describe a neural network as a stack of simple, interconnected decision-makers, using an intuitive analogy.

Explain what 'parameters' (or 'weights') are in a neural network, using a non-mathematical analogy.

Explain why neural networks are often called 'black boxes' and what that means for interpretability and trust.

Distinguish between 'training' an AI model and 'inference' (using the trained model to get a result).

Explain what 'training data' is and why its quality and quantity are foundational to a model's capabilities.

Describe the three main training 'recipes': supervised, unsupervised, and reinforcement learning, using simple analogies.

Explain the two-step process for large models: 'pre-training' on vast data and 'fine-tuning' for specific tasks.

Describe how RLHF (Reinforcement Learning from Human Feedback) is used to align a model with human preferences for safety and helpfulness.

Explain how a model's 'parameter count' (e.g., a '70-billion-parameter model') relates to its potential capability and cost.

Define 'scaling laws' and explain the 'bigger is better' philosophy that has driven recent progress in AI.

Describe the role of specialized hardware (GPUs) in training and running large AI models.

Outline the massive computational, energy, and financial costs required to build and operate a frontier model.

Explain how a Large Language Model (LLM) generates text by 'predicting the next token'.

Define 'token' and 'context window' and explain why a model's context window size matters in practice.

Describe at a conceptual level how image generators (diffusion models) turn random noise into coherent pictures.

Explain what 'multimodal' means and give examples of models that process text, images, and audio.

Compare the major model families a non-expert hears about (e.g., GPT, Claude, Gemini, Llama) at a high level.

Define 'SOTA' (state-of-the-art) and 'benchmark' and explain how they are used to measure model progress.

Define 'frontier model' and contrast the strategic differences between open-source and closed-source AI development.

Explain what an 'AI agent' is and how it differs from a simple chatbot by taking actions.

Describe 'Chain-of-Thought' and other techniques used to improve an AI's reasoning abilities on complex problems.

Explain 'prompt engineering' and the role of 'system prompts' in guiding AI behavior.

Define 'vibe coding' and what it reveals about the new paradigm of building applications with LLMs.

Define 'hallucination' and explain the core reasons why LLMs produce confident-sounding falsehoods.

Describe why models have a 'knowledge cutoff date' and the problem this creates for current events.

Explain 'RAG' (Retrieval-Augmented Generation) as a practical solution to reduce hallucinations and overcome knowledge cutoffs.

Explain why AI models can inherit and amplify societal biases present in their training data.

Describe 'jailbreaking' and 'prompt injection' as two types of adversarial attacks against AI models.

Explain the crucial difference between a model being 'wrong' (a capability limit) and 'misaligned' (a goal conflict).

Define Artificial General Intelligence (AGI) and explain why there is no single, agreed-upon definition.

Define Artificial Superintelligence (ASI) and the 'intelligence explosion' hypothesis.

Explain why 'common-sense reasoning' and building internal 'world models' are considered fundamental, unsolved challenges for AGI.

Describe 'continual learning' and 'embodiment' as other key hurdles on the path to AGI.

Compare the 'scaling is enough' vs. 'new breakthroughs are needed' arguments for achieving AGI.

Describe what 'emergent capabilities' are and why they fuel both excitement and concern about future AI.

Define the 'AI alignment problem' in plain language: ensuring advanced AI pursues human goals.

Explain 'instrumental convergence' using the classic 'paperclip maximizer' thought experiment.

Summarize the main arguments of prominent figures concerned with existential risk (e.g., Nick Bostrom, Eliezer Yudkowsky).

Summarize the main counter-arguments from skeptics who view existential risk as overblown (e.g., Yann LeCun, Melanie Mitchell).

Explain what 'AI safety research' entails, including interpretability, red-teaming, and evaluations.

Describe the debate around open-sourcing powerful models: democratization vs. proliferation risk.

Summarize the main arguments about AI's impact on jobs and which kinds of work are most exposed.

Explain how generative AI threatens the information ecosystem through deepfakes and synthetic media.

Describe the copyright and intellectual property debates surrounding training data and generated content.

Explain privacy concerns specific to AI, such as training data leaks and behavioral profiling.

Describe the geopolitical race for AI dominance, focusing on compute, chips, and regulation.

Summarize concerns about the concentration of power in the few large companies developing frontier AI.

Identify the standard sections of a model release announcement (size, modalities, benchmarks, availability).

Interpret claims about a model's parameters and benchmark scores in a real-world tech article.

Develop a mental checklist for when to trust an AI's answer versus when to verify it independently.

Compare popular consumer AI tools based on their best use cases (e.g., chat, search, image generation).

Identify the tell-tale signs of AI hype versus substantive progress in media coverage.

Articulate your own balanced personal position on AI's risks and benefits, citing concepts from the course.