SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2406.10522 · 2024

Humor in AI: Massive Scale Crowd-Sourced Preferences and Benchmarks for Cartoon Captioning

arXiv · PDF · Open in the Atlas

Code that ran

7 of 10 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; 2 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 5 failures of our instrument; 3 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/2406.10522.json · badge: /api/ran/2406.10522.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
yguooo/cartoon-caption-generation canonical (classifier label)
classifier basis: census
7 ran · 3 no recorded run
FunctionStatusWhere it lives
compute_averageEAD Ran
ran, honoured the contract we drafted
yguooo/cartoon-caption-generation/finetuning/diversity.py
code served (permissive licence) · get_code("cc92ea8d51148b71")
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 float64, ndim 0, 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 numpy as np
import pandas as pd

# Create a small DataFrame with the required structure
# The function expects columns from start_col to end_col to contain text data
# Default: start_col=2, end_col=12, so we need at least 12 columns
# Columns 0-1 can be anything, columns 2-11 will contain text lines

data = {
    0: ['id1', 'id2'],
    1: ['meta1', 'meta2'],
    2: ['hello world', 'the quick brown'],
    3: ['foo bar baz', 'lazy dog jumps'],
    4: ['test line here', 'another test line'],
    5: ['sample text one', 'sample text two'],
    6: ['diversity measure', 'text analysis here'],
    7: ['natural language', 'processing example'],
    8: ['ngram extraction', 'vocabulary size check'],
    9: ['compute average', 'expected average diversity'],
    10: ['more sample text', 'final sample line'],
    11: ['last column data', 'end of data here'],
}

df = pd.DataFrame(data)

ARGS = {
    'df': df,
    'start_col': 2,
    'end_col': 12,
    'n_min': 1,
    'n_max': 5,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
compute_ead Ran
ran, honoured the contract we drafted
yguooo/cartoon-caption-generation/finetuning/diversity.py
code served (permissive licence) · get_code("695ba0735928d882")
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
import numpy as np

torch.manual_seed(0)

VOCAB_SIZE = 32000

def compute_ead(ngrams):
    ngrams = [item for sublist in ngrams for item in sublist]
    N = len(set(ngrams))
    C = len(ngrams)
    V = VOCAB_SIZE
    try:
        ead = N / (V * (1 - ((V - 1) / V) ** C))
    except ZeroDivisionError:
        ead = 0.0
    return ead

# ngrams is expected to be a list of lists (nested structure)
# Each inner list contains ngram tokens (integers)
# The function flattens them and computes statistics

ngrams_input = [
    [101, 202, 303, 404],
    [505, 606, 303, 101],
    [707, 808, 202]
]

ARGS = {
    "ngrams": ngrams_input
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, 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
get_gen_config Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
yguooo/cartoon-caption-generation/finetuning/save_results.py
code served (permissive licence) · get_code("988ca3d5b75e23b5")
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 GenerationConfig. 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
from transformers import GenerationConfig

# Create a mock model name that won't trigger actual downloads
model_name = "gpt2"

# The function only uses model_name to call GenerationConfig.from_pretrained()
# We need to mock this since we can't download real models
# For testing purposes, we'll create a minimal stand-in

class MockGenerationConfig:
    @staticmethod
    def from_pretrained(model_name):
        # Return a basic GenerationConfig-like object with the necessary attributes
        config = GenerationConfig()
        config.max_new_tokens = 128
        config.temperature = 1.0
        config.top_p = 1.0
        config.do_sample = False
        config.repetition_penalty = 1.0
        return config

# Patch the from_pretrained method for testing
GenerationConfig.from_pretrained = MockGenerationConfig.from_pretrained

ARGS = {
    "model_name": "gpt2"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, transformers — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type GenerationConfig — a GenerationConfig output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a GenerationConfig 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
get_unique_dataset Ran
ran; our contract draft was wrong, not the code
yguooo/cartoon-caption-generation/finetuning/save_results.py
code served (permissive licence) · get_code("7de254a7fdd2e953")
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 torch

# Construct a minimal dataframe-like object
# The function iterates over `df` and accesses 'contest_number' and 'prompt' keys
df = [
    {'contest_number': 1, 'prompt': 'Write a story about a cat'},
    {'contest_number': 1, 'prompt': 'Write a story about a dog'},
    {'contest_number': 2, 'prompt': 'Write a poem about the moon'},
    {'contest_number': 2, 'prompt': 'Write a poem about the stars'},
    {'contest_number': 3, 'prompt': 'Write a joke about programming'},
]

ARGS = {
    'df': df
}
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 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
get_unique_dataset_llava Ran
ran; our contract draft was wrong, not the code
yguooo/cartoon-caption-generation/finetuning/save_results.py
code served (permissive licence) · get_code("2933af3f4d861132")
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 torch
import numpy as np

torch.manual_seed(0)

# The function iterates over df and accesses 'contest_number', 'image', and 'prompt' keys
# It checks if contest_number is in a list and appends examples
# So df needs to be an iterable of dicts with these keys

df = [
    {
        'contest_number': 1,
        'image': torch.randn(3, 4),
        'prompt': 'What is in this image?'
    },
    {
        'contest_number': 2,
        'image': torch.randn(3, 4),
        'prompt': 'Describe the scene.'
    },
    {
        'contest_number': 1,
        'image': torch.randn(3, 4),
        'prompt': 'Another question about contest 1'
    },
    {
        'contest_number': 3,
        'image': torch.randn(3, 4),
        'prompt': 'Third contest question'
    }
]

ARGS = {
    'df': df
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — 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
lines_to_ngrams Ran
ran; our contract draft was wrong, not the code
yguooo/cartoon-caption-generation/finetuning/diversity.py
code served (permissive licence) · get_code("62c794901252a993")
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. 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

ARGS = {
    'lines': [
        'Hello world this is a test',
        'Another line with some words',
        'Short text'
    ],
    'n': 3
}
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 list — a list output has no array shape of its own; none is 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
pick_bon Ran
ran; our contract draft was wrong, not the code
yguooo/cartoon-caption-generation/finetuning/save_bon_results.py
code served (permissive licence) · get_code("dbde0c4ce4d646ad")
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 torch
import numpy as np
import pandas as pd

torch.manual_seed(0)

# Create a mock sentiment pipeline
class MockSentimentPipe:
    def __call__(self, texts, **kwargs):
        # Return mock output matching the expected format
        # pipe_outputs is a list where each element has [{'score': float}]
        outputs = []
        for text in texts:
            outputs.append([{'score': float(np.random.rand())}])
        return outputs

# Create a mock row (pandas Series)
row_data = {
    'contest_number': 42,
    0: 'caption_0',
    1: 'caption_1',
    2: 'caption_2',
    3: 'caption_3',
    4: 'caption_4',
    5: 'caption_5',
    6: 'caption_6',
    7: 'caption_7',
    8: 'caption_8',
    9: 'caption_9',
}
row = pd.Series(row_data)

# Create reward_df with matching contest_number and prompt
reward_df = pd.DataFrame({
    'contest_number': [42, 43],
    'prompt': ['This is the prompt: ', 'Another prompt: ']
})

# Create sentiment pipeline mock
sentiment_pipe = MockSentimentPipe()

ARGS = {
    'row': row,
    'reward_df': reward_df,
    'sentiment_pipe': sentiment_pipe,
    'n': 3
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas, torch — 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
preprocess_sft Not yet run
unverified (harvested, no recorded run)
yguooo/cartoon-caption-generation/finetuning/preprocess.py
code served (permissive licence) · get_code("cbefd5b5c362985c")
How it ran

no invocation recorded.

preprocess_sft_simple Not yet run
unverified (harvested, no recorded run)
yguooo/cartoon-caption-generation/finetuning/preprocess.py
code served (permissive licence) · get_code("fb86afa85d9a51fd")
How it ran

no invocation recorded.

preprocess_zs Not yet run
unverified (harvested, no recorded run)
yguooo/cartoon-caption-generation/finetuning/preprocess.py
code served (permissive licence) · get_code("dc34aa3fbb0b0775")
How it ran

no invocation recorded.

Also present in other papers' code

Of 10 samples lifted from this paper's linked repositories, 10 carry a function identity and 0 do not. Of the 10 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 10 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

We present a novel multimodal preference dataset for creative tasks, consisting of over 250 million human ratings on more than 2.2 million captions, collected through crowdsourcing rating data for The New Yorker's weekly cartoon caption contest over the past eight years. This unique dataset supports the development and evaluation of multimodal large language models and preference-based fine-tuning algorithms for humorous caption generation. We propose novel benchmarks for judging the quality of model-generated captions, utilizing both GPT4 and human judgments to establish ranking-based evaluation strategies. Our experimental results highlight the limitations of current fine-tuning methods, such as RLHF and DPO, when applied to creative tasks. Furthermore, we demonstrate that even state-of-the-art models like GPT4 and Claude currently underperform top human contestants in generating humorous captions. As we conclude this extensive data collection effort, we release the entire preference dataset to the research community, fostering further advancements in AI humor generation and evaluation.

For agents

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

get_harvested_code_for_paper("2406.10522")
get_code_for_paper("2406.10522")
have("2406.10522")

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

curl https://syntology.ai/api/ran/2406.10522.json

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

[![Syntology run record](https://syntology.ai/api/ran/2406.10522.svg)](https://syntology.ai/paper/2406.10522)

Connect an agent — have() is free.