Business Pill 28 · Why models state false things
A hallucination is a confident answer with nothing behind it. It cannot be removed completely, but it can be reduced.
A short explainer video. The numbers in it are round numbers for illustration.
The Short Answer
A model writes one word at a time, choosing the most probable next word. It is not looking anything up. It is producing text that sounds right. Most of the time, what sounds right is right.
What a Hallucination Is
But when the model does not know, it does not stop. It keeps producing probable words. The result is fluent, specific, and invented. This is called a hallucination: a confident answer with nothing behind it.
A typical case: ask for a legal precedent. The model returns a case name, a court, a year and a quotation, all perfectly formatted, and none of it exists.
Why It Happens
The model is trained to produce an answer, not to say I do not know. And its confidence is the same whether it is recalling a fact or filling a gap.
What Reduces It
It cannot be removed completely, but it can be reduced. Give the model the source documents. Ask it to cite them. Let it say when it is unsure. And check the answers that matter.
A Simple Rule
The more an answer matters, the less it can rest on the model alone. A first draft can tolerate an error. A contract, a diagnosis, or a financial figure cannot.
Three Questions to Ask
- What happens if this answer is wrong?
- Where could the model have got it from?
- Who checks it before it is used?
See It in the News
TypeSafe’s Jev Claims 444.6x Cheaper, 193.6x Faster. A news piece on a vendor’s zero-hallucinations claim and what it covers.
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
- 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







