Glossary · Term

upper bound

← all terms

Definition

Plain language

The most something could possibly be.

As stated in the literature

A provable or estimated ceiling on a quantity; reported here to flag that an effect size may be inflated by favorable experimental setup.

Also called: upper-bound

Why it matters: Labelling a number as an upper bound warns readers not to expect that result in practice, since the setup was chosen to give the effect its best possible showing.

For example, saying a method fixes "at most forty percent" of failures means forty percent is the ceiling under the most favorable conditions measured, not the typical result.

Heard on the show

“But the authors themselves treat that as an upper bound on a different quantity, because the reference-key arm there is trained on the very rule used to score it.”
Episode 233 — Why a Model Can Grade an Answer But Not Write the Answer Key

Mentioned in 10 episodes

  1. 233
    Why a Model Can Grade an Answer But Not Write the Answer Key
  2. 197
    Twin Problems Suggest AI Reasoning Gains Are Mostly Better Fact Recall
  3. 147
    Agents Fail at the Body, Not the Brain: A Self-Rewriting Scaffold That Lifts a 9B Model 44 Points
  4. 129
    How a Crowd of Anonymous AI Agents Broke a 40-Year Math Record
  5. 108
    The Reasoning Cliff: Why Thinking Longer Makes Models Worse at Exact Step-by-Step Tasks
  6. 036
    Sparse Attention Was the Wrong Frame. Treat It as Geometry Instead.
  7. 030
    Why Your AI Agent Won't Stop Working — and Each Model Falls for a Different Trap
  8. 023
    Why a Small Agent Confidently Overwrites Memories It Doesn't Understand
  9. 017
    When the Agent Grades Its Own Homework: A Brutal New Benchmark for AI Workers
  10. 010
    When Reward Climbs But Reasoning Goes Generic: Diagnosing Template Collapse in Agentic RL

Related terms