AI use cases
2983 ways people actually use AI, compiled from the videos in this
library. Each names the job and why it is worth doing, and links to the
moment in the source video.
-
Autonomously planning retrieval and research
— Deciding whether and where to search, refining queries, validating information, and determining when enough information has been collected
-
Retrieving external documents and augmenting an LLM prompt
— Providing current, specialized, or private information for answer generation
-
Representing text as vector embeddings
— Capturing semantic relationships so systems can retrieve conceptually related documents
-
Predicting the most likely next token or words in an answer
— Generating text responses from patterns learned during training
-
Identify suitable index funds and explain how to invest in them in a specific country.
— Use ChatGPT or Gemini to research local investing platforms and fund access, while still conducting personal due diligence.
-
Automate scientific and engineering research
— Propose experiments, run thousands of them, evaluate results, iterate, and automate the research loop across AI research, hardware, drug discovery, and clean energy.
-
Improving content distribution
— AI was described as useful for translating content into many languages and improving marketing or distribution.
-
Generating and adapting written content
— AI was described as able to write books, contracts, poems, scripts, research, and polished documents from user instructions.
-
Designing a new financial security
— ChatGPT was used to explore and develop STRK, a Bitcoin-backed convertible preferred stock with a variable monthly dividend.
-
-
Identify potentially valuable or attention-grabbing items from photographs
— Help estate sale operators recognize collectibles and write marketing descriptions even when they lack specialized expertise.
-
Recruiting-tool development
— Recruiters can build tools and functions themselves in Cursor or other tools instead of relying on engineering teams.
-
Recruiting-network research
— AI tools could help automate or facilitate digging into current employees' networks, although Adam says personal extraction of signal is more effective.
-
Candidate and company discovery
— An AI string can search job openings across companies for shared characteristics that match the hiring criteria.
-
Answer questions from retrieved document sections
— Generate answers from relevant material while retaining its headings, section path, and relationships to other sections.
-
Navigate structured documents
— Reason over an outline, select relevant sections, follow references, and gather enough context to answer questions.
-
Writing business descriptions
— Help create clean descriptions for directory listings.
-
Explanation
— LLMs are described as potentially becoming good explainers.
-
Biological design
— Enable the creation of novel viruses or other advanced biological tools.
-
Scientific and engineering research
— Improve technology and potentially accelerate progress beyond human ability to understand or predict.
-
Mathematical reasoning
— Help with mathematical proofs, win mathematics competitions and work on difficult mathematical problems.
-
Generating and optimizing interviews and video
— Read a creator's work, learn their style, optimize questions and produce simulated videos of them interviewing people.
-
Automating physical labor through humanoid robots
— Perform physical tasks such as plumbing, cooking and other activities humans can do.
-
Automating computer-based work
— Replace human labor in most occupations and potentially create very high unemployment.
-
Inference acceleration
— Reduce cost and time to first token for specialized workloads such as voice and customer support.
-
Video-game development
— Assist users with weekend creative and programming projects.
-
Homework assistance
— Support one of the most common reported uses of Claude.
-
Moderating social-platform content
— Help administrators, volunteer moderators and users preserve authentic discussion and reduce noise.
-
Detecting and suppressing AI-generated spam and inauthentic accounts
— Defend online platforms against the increased scale, speed and variation of AI-generated abuse.
-
Formulate and send messages
— Allow an AI agent to communicate autonomously, including potentially sending email without human verification
-
Execute and write code
— Allow an AI agent to perform work autonomously
-
Call other services using an API
— Allow an AI agent to take autonomous actions toward a user's goal
-
Read and write to databases
— Achieve a user's goal autonomously through an AI agent
-
Answer questions about products through a customer chatbot
— Provide a controlled customer-facing interaction governed by a system prompt
-
Data-breach prevention and recovery
— Organizations with extensive AI and automation saved an average of $2 million per breach and recovered 65 days faster.
-
AI safety monitoring
— Use internal representations to help identify hallucinations, deceptive reasoning and jailbreak behavior.
-
Model interpretability
— Analyze internal activations and attribution graphs to understand processing that occurs before token outputs.
-
Agent orchestration
— Plan tasks, delegate them to parallel sub-agents and reduce latency for agent workloads.
-
Automated evaluation and synthetic-data generation
— Create an evaluation, generate synthetic data, run post-training and load updated weights into a coding harness in a closed loop.
-
Fine-tuning and customization
— Adapt an open-weight model to specific needs through parameter-efficient fine-tuning, RAG, behavioral control and agent customization.
-
Generate large volumes of static advertising creative.
— Increase the number of ads available for testing and support higher advertising spend.
-
Generate drawings with OpenCV
— Opus 5 can produce portraits, animals, and landscapes when given OpenCV, illustrating capabilities discovered through experimentation.
-
Maintain software code bases
— Recurring routines identify dead code, remove completed experiments, add needed tests, delete unnecessary tests, and unify duplicated abstractions.
-
Orchestrate agents for complex multi-stage work
— Dynamic workflows fan agents out across initial work, verification, summarization, and later stages to increase productive test-time compute.
-
Rewrite an Electron desktop app in Swift
— Claude was given access to a macOS runner and instructed to compare screenshots pixel by pixel until the Swift version matched the Electron app.
-
Rewrite an entire code base between programming languages
— Claude rewrote the Bun code base from Zig to Rust using a test suite and dynamic workflows over 11 days.
-
Detect and resist prompt injection
— A well-aligned model is combined with a prompt injection classifier and an auto mode classifier to prevent malicious instructions encountered online from causing harmful actions.
-
Replacing manual data collection
— AI agents were described as collecting large amounts of data, placing it in dashboards, and preparing drafts and follow-up actions.
-
Employee education and training
— AI was taught as a practical skill so employees could build tools and solve business problems themselves.
-
Business process automation
— AI was used to automate repetitive work, create processes, generate reports, queue emails, and improve operational productivity.
-
Building software without writing code
— Jon Cheney used vibe coding to create an initial product and launch Gen AIPI quickly.
-
Increasing knowledge and usefulness
— Helping younger people adapt to an uncertain labor market
-
Partnering with people who have exceptional human intelligence
— Keeping workers at the cutting edge as AI capabilities expand
-
-
Exploring hypothetical superintelligence
— The discussion considers whether a superintelligent AI could become humanity's overlord and treat humans as pets, entertainment, or something irrelevant.
-
Generating images in established artistic styles
— ChatGPT is described as capable of turning a photograph into an image resembling Van Gogh's style.
-
Accessing, reducing, analyzing, and prioritizing telescope data
— AI enables the Vera Rubin telescope to process observations and decide whether unusual objects should be observed again.
-
Compare complete requirements and release documents
— Identify security requirements omitted from a final release
-
Place complete documents directly into an LLM context window
— Let the model's attention mechanism perform reasoning across the full documents
-