Propose a part

Datasheet · A prompt line

Let's think step by step

Let's think step by step.

Folklore, as cited: Kojima et al., 2022, Large Language Models are Zero-Shot Reasoners

Changes nothing that matters on DeFi maths, Haiku 5.5 with thinking off.

DeFi maths, Haiku 5.5 with thinking off

Pooled across 3 notebooks: −1.4 points, 95% interval −3.6 to +0.9; measured once.

  1. UNRESOLVEDNotebook #003−2.1 points (−6.1 to +2.0), 432 pairs, cost ×1.00signed verdict 13
  2. INERTNotebook #005−0.2 points (−4.2 to +3.7), 432 pairs, cost ×0.99signed verdict 21
  3. UNRESOLVEDNotebook #007−1.7 points (−5.4 to +1.9), 432 pairs, cost ×1.02signed verdict 32

Measure it in a setting it has not been measured in: propose it. As JSON for an agent: /v1/parts/d47935e0024f0eec. Every part: the atlas.