Chain of Thought

Chain of thought is a visible sequence of intermediate reasoning text, useful as a workspace but not identical to hidden reasoning.

~/writing/math-foundations-for-reasoning-models $ cat topic.md

Chain Of Thought

Chain of thought is a sequence of intermediate reasoning text.

Simple form:

step 1 -> step 2 -> step 3 -> answer

It is a way to externalize parts of a reasoning process.

Chain of thought is written intermediate reasoning.

Writing intermediate steps can help structure a task.

Examples:

break problem into parts
track assumptions
hold partial results
avoid skipping steps
make errors easier to inspect

In this sense, chain of thought can act like a scratchpad.

Chain of thought can provide visible structure for multi-step tasks.

Chain of thought is visible text.

Hidden reasoning is internal model computation.

These are not the same thing.

visible chain:
what is written

hidden reasoning:
internal representation transformations

A model may have hidden computation that is not fully shown in the visible chain.

Visible reasoning text is not the full internal process.

A written chain of thought may not always faithfully explain why a model produced an answer.

It may be:

useful
partial
post-hoc
incorrect
incomplete

This is why chain of thought should be treated as an artifact, not perfect transparency.

Chain of thought can help, but it is not guaranteed to be a complete explanation.

In reasoning systems, intermediate text can support:

decomposition
verification
self-correction
tool coordination
human review

The important concept is externalizing useful intermediate structure.

Chain of thought is one form of externalized reasoning workspace.

~/writing/math-foundations-for-reasoning-models $ open

Hidden Reasoning