Business Pill · Many specialists, few at work
Mixture of experts means many specialists, but only a few work on each question. A router picks them. Everyone is still on the payroll, so the saving is in the work, not in the size of the team.
A short explainer video, under a minute. The manager, the receptionist and the specialists are an illustration.
The Short Answer
Picture a manager who hires a hundred specialists. Every question goes to all of them, and the bill is huge.
So she hires a receptionist. The receptionist sends each question to just the few who fit.
What Mixture of Experts Is
That is called mixture of experts. Many specialists, but only a few work on each question.
The video’s whiteboard puts it in one line: many specialists, a few work per question.
What Happens to One Question
A question arrives. The router picks a few experts. Only they answer. The rest stay idle.
The video draws it as four steps: a question, the router picks, a few experts work, and the rest stay idle.
The Honest Limit
But everyone is still on the payroll. The video’s list says every expert still takes up space.
And if the router favours a few, the others sit idle. The video calls this unfair picks leaving experts idle.
Why It Matters
The video turns this into one question: what do we pay for, the whole team, or the few who work?
Team size and work done are different things. So ask how many work at once.
The Question to Ask
- What do we pay for, the whole team, or the few who work?
See It in the News
Ai2’s Olmo-core 3: An Open Stack for Training Mixture-of-Experts Models. A news piece on an open training framework built for models of this kind. This pill explains the idea; it does not repeat that story’s figures.
More Business Pills
- The Memory Wall: Why a Faster AI Chip Is Not Faster AI
- Tokens per Watt: What an AI Data Centre Actually Produces
- Why AI Can’t Be Both Instant and Cheap: Latency vs Throughput
- The Model and the Harness: Why Same-Model Products Differ
- Context, Not Capability: Why a Smart AI Model Gives Poor Answers
- Discardable Software: When Code Is Cheap Enough to Throw Away
- Human in the Loop vs Human on the Loop: Supervising AI Agents
- The AI Audit Problem: When Making Work Is Cheaper Than Checking It
- AI Evaluation as Acceptance Test: How to Know It’s Good Enough
- Extensibility Is Control: Who Holds the Power in an AI Product
- RLHF Explained: How an AI Model Learns What People Prefer
- Pretraining Explained: How an AI Model Learns Before Anyone Teaches It
- Fine-Tuning Explained: How to Adapt a General AI Model to One Job
- Tokens Explained: The Unit AI Reads, Writes and Bills In
- Embeddings Explained: How a Machine Compares Meaning
- RAG Explained: How a Model Answers From Your Documents
- AI Hallucination Explained: Why Models State False Things
- Distillation Explained: How a Small Model Learns From a Large One
- Reasoning Models Explained: What Changes When AI Thinks First
- Tool Use Explained: How an AI Model Goes From Text to Action
- Capex and Depreciation Explained: Why Chip Lifetime Drives AI Profits
- Run-Rate Revenue Explained: What an AI Company’s Number Means
- Backlog Explained: Revenue That Is Signed but Not Yet Earned
- Switching Costs Explained: Why It Is Hard to Leave an AI Supplier
- Temperature Explained: The Dial That Sets How Predictable an AI Model Is
- Guardrails Explained: Rules Enforced Around an AI Model, Not Inside It
- Benchmarks Explained: Why a Test Score Is Not Your Own Result
- Agent Memory Explained: How an AI Assistant Remembers You Between Chats
- MCP Explained: The Shared Plug That Connects AI Models to Tools
- Quantization Explained: How a Large AI Model Is Shrunk to Fit
- Parameters Explained: Where an AI Model’s Knowledge Lives
- Scaling Laws Explained: Why AI Labs Keep Building Bigger
- Synthetic Data Explained: When a Model Writes Its Own Training Examples
- Transformers and Attention Explained: How AI Reads Every Word at Once
- Context Window Explained: How Much Text an AI Model Can Keep in View
- Prompt Injection Explained: When Text a Model Reads Gives the Orders
- Inference Explained: The Cost Paid on Every AI Answer
- Multimodal Explained: AI Models That Work With Images, Sound and Video
- AI Agents Explained: From Answering to Doing
- Open Weights Explained: Rent the Model or Run It Yourself
- Gross Margin Explained: What an AI Company Keeps From Each Dollar
- Usage-Based Pricing Explained: Paying for Work, Not for Seats
- Total Cost of Ownership Explained: Why the Cheap Robot Is the Expensive One
- Free Cash Flow Explained: Profit on Paper, Cash in the Bank
- Few-Shot Prompting Explained: Why Showing a Model Examples Beats Describing
- Knowledge Cutoff Explained: The Date Where an AI Model Stops Knowing
- Overfitting Explained: Why Memorizing Is Not Learning in AI
- Model Routing Explained: Sending Each Question to the Right AI Model
- GPUs Explained: Why AI Runs on a Chip Built for Video Games
- Data Labeling Explained: The Human Work Behind an AI Model
- Churn Explained: The Customers Who Leave Through the Back Door
- Customer Acquisition Cost Explained: The Price of a New Customer
- Operating Leverage Explained: Why Profit Grows Faster Than Sales
- Burn Rate Explained: How Fast Cash Goes and How Long It Lasts
- Computer Use Explained: When an AI Agent Operates the Screen
- Multi-Agent Systems Explained: One Agent or a Team
- Outcome-Based Pricing Explained: Paying for the Result
- Agentic Commerce Explained: When the Shopper Is an Agent
- Sovereign AI Explained: Who Controls the Model and the Data
- Context Engineering Explained: What a Model Sees Before It Answers
- Forward Deployed Engineers Explained: The Engineer Who Moves In
- On-Device AI Explained: A Model in Your Pocket
- Physical AI Explained: When AI Gets a Body
- Neoclouds Explained: Renting the Chips by the Hour
- Prompt Caching Explained: Why You Should Not Pay Twice to Read the Same Page
- Structured Output Explained: Why Programs Need a Form, Not a Paragraph
- Model Deprecation Explained: What Breaks When the Model Under You Changes
- The Jevons Paradox Explained: Why Cheaper AI Can Mean a Bigger Bill
- Take-or-Pay Explained: Paying for AI Chips Whether You Use Them or Not
- Least Privilege Explained: Giving an AI Agent Only the Access Its Task Needs
- Rate Limits Explained: Why an AI Product Can Stall in Its Busiest Minute
- Red Teaming Explained: Attacking Your Own AI System Before Someone Else Does
- Shadow AI Explained: The AI Tools Your Company Cannot See
- Data Flywheel Explained: When Using a Product Makes It Better
- Sale-Leaseback Explained: Sell the Asset, Keep Using It
More Resources
