Managing AI Tasks
If a task has a clear, well-known solution that can be checked quickly, scored at scale, and continuously evaluated without much noise, it is highly ______________ — and therefore highly susceptible to AI automation.
Verifiable
______________ engineering — providing an AI with the right background information, instructions, and supporting knowledge so it can reason about a specific project — is now considered more important than prompt engineering for getting reliable results from a capable model.
Context
The massive physical facility — like xAI's Colossus, built in Memphis in an unprecedented 122 days — where thousands of GPUs, immense cooling water systems, and the power infrastructure required to run AI at scale are all housed under one roof.
Data Center
Any image, video, or audio that has been manipulated by AI to make it appear someone is doing or saying something they actually are not — examples discussed in class included the AI-generated image of the Pope and the fake Joe Biden robocall used for voter suppression.
Deepfakes
The financial event where a private company sells shares to the public for the first time — both OpenAI and Anthropic are reportedly preparing for theirs, which will subject them to far greater regulatory scrutiny and financial transparency than they face today as private firms.
IPO
______________ work involves complex problem-solving, creativity, judgment, and human connection — making it the most durable kind of work to build a career around in the AI era.
Deep
AI that can autonomously execute tasks on a computer on a user's behalf — reading and writing files, generating and running code, browsing the web, and conducting deep research — rather than only responding inside a chat window.
Agentic AI
The software process that breaks text into the small data fragments AI models actually compute on — and the bridge between AI software and physical AI infrastructure, because every fragment a model processes ultimately consumes GPU cycles, electricity, and cooling water in the real world.
Tokenization
The most extreme category of AI concern is ______________ risk — the threat that sufficiently advanced AI systems could pose a danger to humanity's continued existence, a topic of growing public debate as model capabilities advance.
Existential
Software businesses like Gamma are called AI ______________ — they sit on top of foundational models from OpenAI, Anthropic, and Google, capturing value through better user interfaces, productized system prompts, and tightly designed workflows rather than building the underlying models themselves.
Wrappers
Even when a task like accounting is highly verifiable and could technically be done by AI, this factor — the responsibility for the outcome — limits how far AI actually penetrates a profession, because someone with credentials must be accountable when things go wrong.
Liability
The unit of data that AI models are charged on — measured both for what a user sends in (the prompt and any attached context) and for what the model generates back — making it the fundamental basis of cost when running an AI agent.
Tokens
When AI companies talk about their primary scarce resource, they mean ______________ — processing power measured in GPU cycles — and they ration it carefully, sometimes abandoning expensive side projects like OpenAI's Sora video tool to free up resources for more profitable enterprise work.
Compute
The idea that the public internet is becoming increasingly populated with AI-generated content — bots, synthetic articles, fake images, AI comments — to the point where it grows harder and harder to distinguish real human activity from machine activity.
Dead Internet Theory
The strict government oversight currently in short supply across the AI industry, but expected to intensify sharply as the largest AI companies prepare to go public — at which point investors and regulators will demand transparent disclosure of risks, finances, and operations.
Regulatory Scrutiny
An ______________ job is one where AI enhances human capabilities and lets experienced workers intensify their team's output, in contrast to an automated job where AI fully replaces the human.
Augmented
Project-specific knowledge that lives only in the user's head — opinions, hidden preferences, and constraints the AI cannot pull from public training data — is called ______________ knowledge, and is typically mined through self-interviewing and added to the context.
Tacit
The three physical-world resources currently constraining AI data center expansion are: GPUs, cooling water, and ______________ from local power grids.
Energy
The challenge in AI safety of ensuring that an AI system's goals, preferences, and values actually reflect those of the humans it serves.
Alignment Problem
When a struggling shoe company like Allbirds pivots its business model to 'AI infrastructure' just to lift its share price — and investors keep buying AI-related stocks largely because they expect them to keep climbing rather than because of the underlying business fundamentals — analysts call this a ______________ bubble.
Financial
When an AI product like Gamma can generate a polished slide deck from a single prompt, the design of the slides becomes a verifiable, automatable task — and the durable, value-capturing skill in any presentation shifts to ______________.
Public Speaking
The category of AI models whose underlying weights, code, and training methods are publicly released for anyone to inspect, modify, or run, in contrast to proprietary systems like GPT or Claude where the internals are kept closed.
Open Source
The principle that AI systems become predictably more capable as their core inputs — compute, data, and energy — are aggressively scaled up, justifying the trillion-dollar race to build massive AI infrastructure.
1000 Points:
Scaling Laws
The challenge in AI safety of maintaining humans' ability to direct, constrain, modify, or stop an AI's actions — distinct from whether the AI's underlying goals match the humans' goals to begin with.
Control Problem
When the mathematician Terence Tao transparently uses AI to multiply his research productivity and unlock entirely new theorems, it serves as evidence we are NOT in a ______________ bubble — because the underlying capabilities are genuinely advancing the frontier, even if the stock prices wrapped around them may be inflated.
Technology