Observations
Record / 042024–2025

Before the answer

Time and computation become part of the answer.

OpenAI · DeepSeekThe reply reveals part of the work.

o1-preview arrived in September 2024, with reinforcement learning and more computation before answering. DeepSeek-R1 followed in January 2025, combining initial supervised training with reinforcement learning.

The interface reveals only part of the work. The pause leaves room for wonder—and for a deeper question.

Agency

What does the answer reveal of its making?

The other side
Before contact

Next observation

The monolith