AI use cases
3459 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.
-
Prefix caching
— Reuse KV-cache memory for requests with shared system prompts and avoid repeated prefill computation.
-
Paged attention
— Allocate KV-cache memory in fixed-size, non-contiguous pages to reduce fragmentation and improve GPU utilization.
-
KV caching
— Reuse previously computed key and value matrices during autoregressive token generation.
-
LLM inference
— Generate tokens from user prompts across prefill and decode phases.
-
Represent populations at scale
— Provide a scalable representation of people's perspectives for decision-making.
-
Generate and validate hypotheses
— Produce large numbers of hypotheses and compare their predicted outcomes with events in the real world.
-
Run counterfactual simulations
— Show how people, audiences and markets might respond to alternative decisions, products, strategies or policies.
-
Create agents with memory, planning and reflection
— Enable agents to remember interactions, plan activities and form higher-level views of their experiences.
-
Simulate human behavior
— Model people's values, preferences, biases, tastes, mistakes and actions in different contexts.
-
Quantum machine learning
— Combine qubits and classical techniques to implement quantum machine learning.
-
Regulatory compliance
— Finding applicable standards and generating evidence tables more quickly.
-
Software development
— Supporting coding; Max Hodak says he has written little source code in the previous six months.
-
Internal company operations
— Making company context and operational data available to AI agents through Helix.
-
Hiring assessment
— Designing take-home assignments that are resistant to AI assistance and can be scored with measurable outputs.
-
-
Storyboarding and filmmaking support
— Help filmmakers accelerate pre-production while preserving creative control
-
Routing requests across multiple models
— Combine models with different strengths and provide a virtual model endpoint
-
Explain and reverse-engineer legacy code
— Clarify the purpose and functions of poorly understood systems and provide a path forward for modernization.
-
Diagnose operational problems
— Interpret log data and stack trace errors when applications fail.
-
Generate test data
— Support unit testing of specific application cases and help QA teams.
-
Implement well-defined software tasks
— Build specific features and functions from a clear specification.
-
Analyze logs and bug reports
— Identify root causes and inform requirements and design based on production behavior.
-
Synthesize unstructured product information
— Identify user behavior bottlenecks and usage patterns, then generate user stories.
-
Frontier-model development
— Anthropic's safeguards restrict use of Fable for building model-training pipelines, distributed training, AI chips or competing frontier models.
-
Customer service
— AI agents need to recognize sarcasm when interacting with frustrated customers.
-
Diagramming and game creation
— Fable 5 is described as having improved spatial awareness and producing code that avoids overlap.
-
Code analysis and debugging
— Fable 5 is described as better at identifying bugs, nuanced issues and problems users had not considered.
-
Long-term planning and coding
— Fable 5 can continue coding tasks for longer and cover more files.
-
Scan code for vulnerabilities
— Detect and fix security weaknesses before code is released through DevSecOps.
-
Develop exploits and malware
— Enable attackers to take advantage of newly identified vulnerabilities.
-
Identify zero-day vulnerabilities
— Find previously unknown bugs in operating systems and web browsers so defenders can address them.
-
Dynamic behavior analysis
— AI may be needed to analyze the behavior of non-human identities and help address security problems created by AI.
-
Identity and access management
— AI agents require dynamically provisioned, short-lived identities and narrowly scoped permissions.
-
Vulnerability discovery and remediation
— The methodology could be redirected to find vulnerabilities, report them and explain how to fix them instead of exploiting them.
-
Malware propagation and attack reasoning
— A self-replicating agent used a local model to spread between devices and select vulnerabilities and exploits for each device.
-
Customer support and account recovery
— The AI agent was used to handle account requests, but hackers manipulated it into changing recovery emails and enabling account takeover.
-
Triage and route incoming requests
— Categorize work, prioritize attention, route requests to appropriate teams, and suggest responses based on historical data.
-
Process structured operational work
— Extract data, match records, validate information, route approvals, and update downstream systems.
-
Execute policy-governed actions
— Evaluate applicable rules, perform permitted low-risk actions, and escalate ambiguous or high-risk requests.
-
Coordinate multi-step workflows across systems
— Sequence actions, maintain context and state, manage timing, and monitor deviations from expected behavior.
-
World action model inference
— World action models use predicted future representations and kinematics to support action generation, while distillation and latent representations reduce inference cost.
-
Dexterous tool manipulation
— Simulation-trained reinforcement-learning policies learn goal-conditioned hand and arm control that can generalize zero shot to unseen tools and tasks.
-
Embodied reasoning
— Self-supervised bootstrapping identifies which reasoning traces are useful for a particular robot embodiment and removes distracting or non-action-predictive reasoning.
-
Long-horizon robot control
— Multiscale embodied memory gives policies task history, time awareness, progress tracking, and the ability to adapt after mistakes.
-
Using a large language model with tools in an autonomous loop
— To let an agent perform tasks, access information, execute commands, call APIs, and take actions without continuous human oversight.
-
Sending emails
— Send autonomous or automated messages to recipients
-
-
Booking flights to Milan under $500
— Create a travel-booking plan involving flight search, filtering, booking, and sending a confirmation email
-
Finding a document
— Search, evaluate results, replan, and retry until the agent finds an answer
-
Generate a hero image and other website media.
— Create visual assets matching the website's look and feel.
-
Use an AI agent to build and deploy the directory website from an uploaded CSV.
— Create searchable pages, listing pages, categories, business submissions, and paid featured listings.
-
Generate a table of real estate sale cleanout businesses in the greater Dallas-Fort Worth area.
— Populate the directory with business names, categories, phone numbers, and one-line descriptions.
-
-
-
Execute delegated development and operational tasks
— Allow users to manage systems that write or operate software rather than performing every coding or operating task directly.
-
Generate and review software
— Support a hybrid development model in which AI writes code and humans review or edit it.
-
Augment engineers handling open source vulnerabilities
— Streamline vulnerability reporting and resolution, validated patch deployment and upstream disclosure coordination.
-
Chain low-severity vulnerabilities into attack patterns
— Identify system-level exploit paths that individual vulnerability scanners may miss.
-
-