Glossary · Term

wall-clock time

← all terms

Definition

Plain language

How long something actually takes in real-world seconds, as opposed to how much computation it does.

As stated in the literature

Elapsed real time for a computation, distinct from FLOPs or token counts; used as the budget axis in challenges like Autoresearch and as a cost metric for sandbox checkpointing, agent search, and RL training throughput.

Also called: wall-clock, wall clock

Why it matters: It captures how long users and researchers actually wait, which is the cost that matters in practice regardless of raw computation counts.

For example, a task might take ten real-world minutes even if the computer was doing heavy calculation for only part of that stretch.

Heard on the show

“… guess at expansions on slightly stale information and roll back the bad guesses — and it cuts the wall-clock overhead from about fourteen percent down to under five, with accuracy basically untouched. …”
Episode 162 — The Empty-Lake Proof: Why More Rollouts Stop Helping Reasoning Models

Mentioned in 18 episodes

  1. 177
    Why Raw Profiler Data Made an AI Worse at Writing GPU Code
  2. 162
    The Empty-Lake Proof: Why More Rollouts Stop Helping Reasoning Models
  3. 159
    Can a Coding Agent Run Its Own Robot Experiments Overnight, With No Human Resetting the Scene?
  4. 154
    How a 7B Model Out-Investigates a 72B One by Choosing What to Look At
  5. 139
    When Optimizing One GPU Kernel Quietly Breaks the Whole System
  6. 123
    Five Identical Worlds, One Swapped Model: What Happens When AI Agents Run for Fifteen Days
  7. 116
    Why Streaming Half a Reasoning Chain Beats Sending the Whole Thing
  8. 101
    Treating Math Formalization Like a Codebase, and Where the Agents Cheat
  9. 097
    Same Tokens, Same Cost, Wildly Different Results: What Actually Scales in AI Agents
  10. 096
    How Treating an AI Agent's Execution Like Git Recovers a Coordination Penalty
  11. 095
    Seven Wins to Zero: How Organizing AI Agents Like a Lab Changes the Search
  12. 080
    How a Two-Agent Trick Unlocked Large-Scale Training for Computer-Use Agents
  13. 063
    Why Web Agents Are Slow: A Compiler-Style Fix for Computer-Use Latency
  14. 053
    An AI Agent Swapped In Focal Loss And Beat A Human-Tuned Training Script
  15. 036
    Sparse Attention Was the Wrong Frame. Treat It as Geometry Instead.
  16. 033
    Echo: The Paper Arguing You Never Needed a KV Cache for Retrieval
  17. 028
    Teaching a Model to Hire Copies of Itself: Recursive Agent Optimization
  18. 025
    The Missing Gradient Term That Predicts Sycophancy in RLHF

Related concepts

Related terms