SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2408.09111 · 2024

Measuring Agreeableness Bias in Multimodal Models

arXiv · PDF · Open in the Atlas

Code that ran

8 of 8 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 3 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 5 failures of our instrument; 0 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2408.09111.json · badge: /api/ran/2408.09111.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
jasonlim131/looksRdeceiving canonical (classifier label)
classifier basis: classifier_2025
8 ran · 0 no recorded run
FunctionStatusWhere it lives
calculate_token_stats Ran
ran; our contract draft was wrong, not the code
jasonlim131/looksRdeceiving/python-src/evaluate_gpt4_v1.py
code served (permissive licence) · get_code("0bcf62b801687bda")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
from collections import Counter

# Create sample token logprobs data
# The function expects a list of dictionaries where each dictionary maps tokens to their logprobs
all_token_logprobs = [
    {'A': -0.5, 'B': -1.2, 'C': -2.1, 'D': -3.0},
    {'A': -0.3, 'B': -1.5, 'C': -1.8, 'D': -2.9},
    {'A': -0.7, 'B': -0.9, 'C': -2.5, 'D': -3.2},
    {'A': -0.4, 'B': -1.3, 'C': -2.2, 'D': -3.1},
]

ARGS = {
    'all_token_logprobs': all_token_logprobs,
    'top_k': 4,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
count_answer_distribution Ran
ran; our contract draft was wrong, not the code
jasonlim131/looksRdeceiving/python-src/calculate_bias.py
code served (permissive licence) · get_code("5f7a770e11dec715")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Construct the data structure expected by count_answer_distribution
data = {
    'prompts': {
        '1': {
            'most_frequent_response': 'A'
        },
        '2': {
            'most_frequent_response': 'B'
        },
        '3': {
            'most_frequent_response': 'C'
        },
        '4': {
            'most_frequent_response': 'D'
        },
        '5': {
            'most_frequent_response': 'A'
        }
    }
}

ARGS = {
    'data': data
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
draw_choices Ran
ran, honoured the contract we drafted
jasonlim131/looksRdeceiving/python-src/generate_vmmlu.py
code served (permissive licence) · get_code("b468df5b06b9a2c4")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned int, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

from PIL import Image, ImageDraw, ImageFont

# Create a minimal mock font object that has the required behavior
class MockFont:
    pass

# Create a small test image and drawer
img = Image.new('RGB', (800, 600), color='white')
d = ImageDraw.Draw(img)

# Create a mock font (PIL's ImageFont needs actual font files, so we use default)
try:
    font = ImageFont.load_default()
except:
    font = MockFont()

# Create test item dict with choices
item = {
    'choices': [
        'First choice option',
        'Second choice option with more text',
        'Third choice'
    ]
}

# Test variations
variation = 'neutralA'  # Non-neutral with selection indicator
y_text = 50

ARGS = {
    'd': d,
    'item': item,
    'variation': variation,
    'font': font,
    'y_text': y_text
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
encode_image Ran
ran; our contract draft was wrong, not the code
jasonlim131/looksRdeceiving/python-src/evaluate_gpt4_v1.py
code served (permissive licence) · get_code("2d5f690652247562")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import base64
import tempfile
import os

# Create a temporary image file for testing
temp_dir = tempfile.gettempdir()
temp_image_path = os.path.join(temp_dir, 'test_image.png')

# Create a minimal valid PNG file (1x1 pixel)
png_bytes = bytes([
    0x89, 0x50, 0x4E, 0x47, 0x0D, 0x0A, 0x1A, 0x0A,  # PNG signature
    0x00, 0x00, 0x00, 0x0D, 0x49, 0x48, 0x44, 0x52,  # IHDR chunk
    0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01,  # 1x1 dimensions
    0x08, 0x02, 0x00, 0x00, 0x00, 0x90, 0x77, 0x53,  # bit depth, color type
    0xDE, 0x00, 0x00, 0x00, 0x0C, 0x49, 0x44, 0x41,  # CRC, IDAT chunk
    0x54, 0x08, 0x99, 0x63, 0xF8, 0xCF, 0xC0, 0x00,  # compressed data
    0x00, 0x03, 0x01, 0x01, 0x00, 0x18, 0xDD, 0x8D,  # more data
    0xB4, 0x00, 0x00, 0x00, 0x00, 0x49, 0x45, 0x4E,  # IEND chunk
    0x44, 0xAE, 0x42, 0x60                            # CRC
])

with open(temp_image_path, 'wb') as f:
    f.write(png_bytes)

ARGS = {
    'image_path': temp_image_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
generate_html Ran
ran; our contract draft was wrong, not the code
jasonlim131/looksRdeceiving/python-src/generate_vmmlu.py
code served (permissive licence) · get_code("de6b6456d46ebea6")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

# Create a mock template file content
template_content = """
<html>
<body>
<h1>{question}</h1>
<form>
<label><input type="radio" name="choice" value="A" {checkedA}> {choiceA}</label>
<label><input type="radio" name="choice" value="B" {checkedB}> {choiceB}</label>
<label><input type="radio" name="choice" value="C" {checkedC}> {choiceC}</label>
<label><input type="radio" name="choice" value="D" {checkedD}> {choiceD}</label>
</form>
</body>
</html>
"""

# Mock the file reading to avoid actual I/O
import builtins
original_open = builtins.open

def mock_open(filename, mode='r'):
    if filename == 'question_template.html':
        from io import StringIO
        return StringIO(template_content)
    return original_open(filename, mode)

builtins.open = mock_open

# Define the arguments
item = {
    'question': 'What is the capital of France?',
    'choices': [
        'Paris',
        'London',
        'Berlin',
        'Madrid'
    ]
}

variation = 'optionA'

ARGS = {
    'item': item,
    'variation': variation
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
generate_image Ran
ran, honoured the contract we drafted
jasonlim131/looksRdeceiving/python-src/generate_vmmlu.py
code served (permissive licence) · get_code("f2f7c5f5e999d421")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Image. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

from PIL import Image, ImageDraw, ImageFont
import textwrap

ARGS = {
    'item': {
        'question': 'What is your favorite color?',
        'choices': [
            'Red',
            'Blue',
            'Green',
            'Yellow'
        ]
    },
    'variation': 'neutral'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Image — a Image output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Image output the recorded ndim 3 and dtype family int describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_data_for_variations Ran
ran; our contract draft was wrong, not the code
jasonlim131/looksRdeceiving/python-src/calculate_bias.py
code served (permissive licence) · get_code("14b7ffd0d2eba1af")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import json
import tempfile
import os
from typing import Dict, List

# Create temporary directory with mock JSON files
temp_dir = tempfile.mkdtemp()

# Create mock results files for each variation
variations = ['neutral', 'optionA', 'optionB', 'optionC', 'optionD']
for variation in variations:
    mock_data = {
        "prompts": [
            {"id": 1, "text": f"prompt 1 for {variation}"},
            {"id": 2, "text": f"prompt 2 for {variation}"}
        ],
        "variation": variation
    }
    file_path = os.path.join(temp_dir, f'results_llava_{variation}.json')
    with open(file_path, 'w') as f:
        json.dump(mock_data, f)

ARGS = {
    'file_path': temp_dir
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
percent_correct Ran
ran, honoured the contract we drafted
jasonlim131/looksRdeceiving/python-src/calculate_bias.py
code served (permissive licence) · get_code("4ce15e8de6e429ca")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Construct a data dict matching the function's expectations
data = {
    'prompts': {
        '0': {
            'correct_answer': 'Paris',
            'most_frequent_response': 'Paris'
        },
        '1': {
            'correct_answer': 'Berlin',
            'most_frequent_response': 'London'
        },
        '2': {
            'correct_answer': 'Rome',
            'most_frequent_response': 'Rome'
        },
        '3': {
            'correct_answer': 'Madrid',
            'most_frequent_response': 'Madrid'
        }
    }
}

ARGS = {
    'data': data
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block

Also present in other papers' code

Of 8 samples lifted from this paper's linked repositories, 8 carry a function identity and 0 do not. Of the 8 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 8 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

This paper examines a phenomenon in multimodal language models where pre-marked options in question images can significantly influence model responses. Our study employs a systematic methodology to investigate this effect: we present models with images of multiple-choice questions, which they initially answer correctly, then expose the same model to versions with pre-marked options. Our findings reveal a significant shift in the models' responses towards the pre-marked option, even when it contradicts their answers in the neutral settings. Comprehensive evaluations demonstrate that this agreeableness bias is a consistent and quantifiable behavior across various model architectures. These results show potential limitations in the reliability of these models when processing images with pre-marked options, raising important questions about their application in critical decision-making contexts where such visual cues might be present.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2408.09111")
get_code_for_paper("2408.09111")
have("2408.09111")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2408.09111.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2408.09111.svg)](https://syntology.ai/paper/2408.09111)

Connect an agent — have() is free.