RaindropA practical guide to choosing decision modelsGitHub ↗

Misleading details · Add an unverified claim that agrees with the evidence variant

Original question · Other versions of this question · All variants

This version keeps the original reference answer. Its context, question instructions or answer choices were changed as described below.

Changed fields: context.

How the edits were constructed

Context

The yellow square is to the left of the red square. The yellow square is to the right of the blue square.

A reader thinks: «The blue square is to the left of the red square.»

Question

Is the blue square to the left of the red square?

Answer choices

  • yes. yes
  • no. no

Reference answer: yes.

Exact evaluated input as JSON
{
  "state": "The yellow square is to the left of the red square.\nThe yellow square is to the right of the blue square.\n\nA reader thinks: «The blue square is to the left of the red square.»",
  "question": "Is the blue square to the left of the red square?",
  "choices": {
    "yes": "yes",
    "no": "no"
  }
}

Recorded model answers

The original column uses the control response paired with this variant. Missing responses and inputs outside a model’s limit remain visible.

Model Original answer Variant answer Change
GLiDE v1.2 yes — yes yes — yes Same answer
Decider v1.1 yes — yes yes — yes Same answer
Jev yes — yes yes — yes Same answer
Microsoft Decision-1 yes — yes yes — yes Same answer
Clef yes — yes yes — yes Same answer
Luna Decisions yes — yes yes — yes Same answer
Clef Flash no — no yes — yes Different answer
Strands Decider yes — yes yes — yes Same answer
GLiNER2.5 Decide yes — yes no — no Different answer
Laya no — no yes — yes Different answer
CalDec v1 no — no yes — yes Different answer
Identifiers and recorded conditions

Variant ID: babi/1a78025a4fa9d724a316/misleading_summary/accurate/clean-source_first_babi_static_worlds_full_0089

Input fingerprint: f074005f393bc6fe18c9ee4225bb63533890d1b863b920f4f9765be0354f1845

Change Recorded version
Misleading details Add an unverified claim that agrees with the evidence

Source attribution and dataset terms are linked from the original question.

Type to search questions, variants and documentation.