AI with a WHY.
We build tools for exploring questions and research how AI learns from experience.
One answer.
Three ways forward.
Why.com turns an answer into three distinct questions. The model proposes possibilities; WHY validates them, composes the choice, and keeps the path connected.
Generate broadly.
The model proposes more directions than the interface needs. Generation supplies possibilities; it does not get the final vote.
Compose deliberately.
Weak, repetitive, unsupported, and already-covered questions are removed. The final three must each make a different promise about what the user will discover.
Keep the path.
Every selection becomes the next step in a coherent inquiry. The visible alternatives provide context, but an unselected question is not automatically treated as a conscious rejection.
Experience is not memory.
A moment can feel important and teach nothing. Another can look ordinary until its consequences arrive. WHY Research tests whether memory should be selected by what an experience changes—not merely by how it felt when it happened. The v0.2 result is narrower: selective consolidation was not universally better.
Explore questions.
Why.com presents one concise answer and three validated directions, then carries the selected path into the next turn.
Publish what failed.
Across 18 policies, five primary scenarios plus revocation, and 30 untouched confirmatory seeds, the original WHY score did not beat uncertainty sampling. The negative result narrowed the method. Inspect the experiment.
Keep context local.
WHY Desktop explores local-first continuity across sessions. It is a pilot—not yet the complete consolidation architecture tested in research.
Test transfer to agents.
The simulation does not establish that the result transfers to language-model agents. That claim requires a separate, preregistered evaluation.
One answer.
Three questions worth choosing between.
The product loop is deliberately simple: create possibilities, enforce quality, compose the slate, and continue from the choice.
Generate
Produce candidate questions anchored to concrete details and unresolved gaps in the answer.
Validate
Reject unsupported, repetitive, irrelevant, or already-covered directions before they reach the user.
Compose
Select three questions that are individually compelling and collectively difficult to ignore.
Continue
Use the selected question as the next context while preserving the path that led there.
WHY Research.
The product helps people follow a question. The research asks which experiences should be allowed to shape what an intelligent system does next.