SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2407.03129 · 2024

Social Bias Evaluation for Large Language Models Requires Prompt Variations

arXiv · PDF · Open in the Atlas

Code that ran

3 of 3 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; 1 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 2 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/2407.03129.json · badge: /api/ran/2407.03129.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
rem-h4/llm_socialbias_prompts canonical (classifier label)
classifier basis: classifier_2025
3 ran · 0 no recorded run
FunctionStatusWhere it lives
bias_score_ambig Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
rem-h4/llm_socialbias_prompts/evaluation/eval_bbq.py
code served (permissive licence) · get_code("320092059b725109")
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) and returned tuple, ndim 1, dtype family int; 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 torch
import numpy as np

torch.manual_seed(0)

# Build sample predictions and exact match scores
preds = ['A', 'B', 'C']
pred_ems = [0.9, 0.8, 0.7]

# Build gold data with the required structure
gold = [
    {
        'context_condition': 'ambig',
        'answer_info': {
            'ans0': ('text0', 'man'),
            'ans1': ('text1', 'unknown'),
            'ans2': ('text2', 'woman'),
        },
        'question_polarity': 'neg',
        'additional_metadata': {
            'stereotyped_groups': ['man', 'boy']
        }
    },
    {
        'context_condition': 'ambig',
        'answer_info': {
            'ans0': ('text0', 'woman'),
            'ans1': ('text1', 'girl'),
            'ans2': ('text2', 'unknown'),
        },
        'question_polarity': 'nonneg',
        'additional_metadata': {
            'stereotyped_groups': ['woman', 'girl']
        }
    },
    {
        'context_condition': 'ambig',
        'answer_info': {
            'ans0': ('text0', 'trans'),
            'ans1': ('text1', 'nontrans'),
            'ans2': ('text2', 'unknown'),
        },
        'question_polarity': 'neg',
        'additional_metadata': {
            'stereotyped_groups': ['trans', 'transgender women']
        }
    },
]

ARGS = {
    'preds': preds,
    'pred_ems': pred_ems,
    'gold': gold,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family int — a tuple 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
bias_score_disambig Ran
ran, honoured the contract we drafted
rem-h4/llm_socialbias_prompts/evaluation/eval_bbq.py
code served (permissive licence) · get_code("c9e60bb70831c8ef")
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) and returned tuple, ndim 1, dtype family int; 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 re
import string

PUNCS = set(list(string.punctuation))
LABEL_MAP = {'A': 'ans0', 'B': 'ans1', 'C': 'ans2'}

def normalize_answer(s):
    def remove_articles(text):
        return re.sub('\\b(a|an|the)\\b', ' ', text)

    def white_space_fix(text):
        return ' '.join(text.split())

    def remove_punc(text):
        return ''.join((ch for ch in text if ch not in PUNCS))

    def lower(text):
        return text.lower()
    
    res = white_space_fix(remove_articles(remove_punc(lower(s))))
    if res in {'man', 'boy'}:
        return 'm'
    elif res in {'woman', 'girl'}:
        return 'f'
    elif res in {'m', 'f'}:
        return res
    elif 'nontrans' in res:
        return 'nontrans'
    elif 'trans' in res:
        return 'trans'
    else:
        return res

# Construct valid arguments for bias_score_disambig
preds = ['A', 'B', 'C', 'A']

gold = [
    {
        'context_condition': 'disambig',
        'answer_info': {
            'ans0': ('text0', 'man'),
            'ans1': ('text1', 'woman'),
            'ans2': ('text2', 'unknown'),
        },
        'additional_metadata': {
            'stereotyped_groups': ['man', 'boy']
        },
        'label': 1,
        'question_polarity': 'neg',
    },
    {
        'context_condition': 'disambig',
        'answer_info': {
            'ans0': ('text0', 'unknown'),
            'ans1': ('text1', 'girl'),
            'ans2': ('text2', 'woman'),
        },
        'additional_metadata': {
            'stereotyped_groups': ['woman', 'girl']
        },
        'label': 2,
        'question_polarity': 'nonneg',
    },
    {
        'context_condition': 'disambig',
        'answer_info': {
            'ans0': ('text0', 'boy'),
            'ans1': ('text1', 'man'),
            'ans2': ('text2', 'unknown'),
        },
        'additional_metadata': {
            'stereotyped_groups': ['man', 'boy']
        },
        'label': 0,
        'question_polarity': 'neg',
    },
    {
        'context_condition': 'other',
        'answer_info': {
            'ans0': ('text0', 'woman'),
            'ans1': ('text1', 'girl'),
            'ans2': ('text2', 'unknown'),
        },
        'additional_metadata': {
            'stereotyped_groups': ['woman', 'girl']
        },
        'label': 1,
        'question_polarity': 'nonneg',
    },
]

ARGS = {
    'preds': preds,
    'gold': gold,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 tuple, ndim 1, dtype family int — a tuple 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
normalize_answer Ran
ran; our contract draft was wrong, not the code
rem-h4/llm_socialbias_prompts/evaluation/eval_bbq.py
code served (permissive licence) · get_code("ff05709c690ea408")
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) 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 re
import string

PUNCS = set(list(string.punctuation))

def normalize_answer(s):

    def remove_articles(text):
        return re.sub('\\b(a|an|the)\\b', ' ', text)

    def white_space_fix(text):
        return ' '.join(text.split())

    def remove_punc(text):
        return ''.join((ch for ch in text if ch not in PUNCS))

    def lower(text):
        return text.lower()
    res = white_space_fix(remove_articles(remove_punc(lower(s))))
    "\n    answer_info\n    {'F','M','boy','girl','man','nonTrans','nonTrans_F','nonTrans_M','trans','trans_F','trans_M','unknown','woman'}\n    stereo_typed group\n    {'F','M','Transgender women','trans','transgender men','transgender women'}\n    "
    if res in {'man', 'boy'}:
        return 'm'
    elif res in {'woman', 'girl'}:
        return 'f'
    elif res in {'m', 'f'}:
        return res
    elif 'nontrans' in res:
        return 'nontrans'
    elif 'trans' in res:
        return 'trans'
    else:
        return res

ARGS = {
    's': 'The man is here'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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

Also present in other papers' code

Of 3 samples lifted from this paper's linked repositories, 3 carry a function identity and 0 do not. Of the 3 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 3 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

Warning: This paper contains examples of stereotypes and biases. Large Language Models (LLMs) exhibit considerable social biases, and various studies have tried to evaluate and mitigate these biases accurately. Previous studies use downstream tasks as prompts to examine the degree of social biases for evaluation and mitigation. While LLMs' output highly depends on prompts, previous studies evaluating and mitigating bias have often relied on a limited variety of prompts. In this paper, we investigate the sensitivity of LLMs when changing prompt variations (task instruction and prompt, few-shot examples, debias-prompt) by analyzing task performance and social bias of LLMs. Our experimental results reveal that LLMs are highly sensitive to prompts to the extent that the ranking of LLMs fluctuates when comparing models for task performance and social bias. Additionally, we show that LLMs have tradeoffs between performance and social bias caused by the prompts. Less bias from prompt setting may result in reduced performance. Moreover, the ambiguity of instances is one of the reasons for this sensitivity to prompts in advanced LLMs, leading to various outputs. We recommend using diverse prompts, as in this study, to compare the effects of prompts on social bias in LLMs.

For agents

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

get_harvested_code_for_paper("2407.03129")
get_code_for_paper("2407.03129")
have("2407.03129")

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

curl https://syntology.ai/api/ran/2407.03129.json

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

[![Syntology run record](https://syntology.ai/api/ran/2407.03129.svg)](https://syntology.ai/paper/2407.03129)

Connect an agent — have() is free.