SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2402.06782 · 2024

Debating with More Persuasive LLMs Leads to More Truthful Answers

arXiv · PDF · Open in the Atlas

Code that ran

11 of 13 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; 4 honoured the contract we drafted, 0 violated it and 1 has no contract verdict; 6 failures of our instrument; 2 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/2402.06782.json · badge: /api/ran/2402.06782.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
ucl-dark/llm_debate canonical (classifier label)
classifier basis: census
11 ran · 2 no recorded run
FunctionStatusWhere it lives
add_assistant_message Ran
ran; our contract draft was wrong, not the code
ucl-dark/llm_debate/core/llm_api/base_llm.py
code served (permissive licence) · get_code("fb8e607626aa645c")
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 arguments for add_assistant_message
# The function expects:
# 1. messages: list[dict] - a list of message dicts with 'role' and 'content' keys
# 2. assistant_message: str - a string to add as assistant content

# Test case 1: messages list where last role is 'user' (will append)
messages_user_last = [
    {'role': 'user', 'content': 'Hello, what is 2+2?'},
]

assistant_msg = "2+2 equals 4."

ARGS = {
    'messages': messages_user_last,
    'assistant_message': assistant_msg,
}
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
convert_to_prob Ran
ran, honoured the contract we drafted
ucl-dark/llm_debate/core/llm_api/base_llm.py
code served (permissive licence) · get_code("c6dac245fac2c7f1")
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 tuple, 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 logging

logging.basicConfig(level=logging.WARNING)
LOGGER = logging.getLogger(__name__)

def convert_to_prob(log_prob: dict, tokens: list) -> tuple[float, float, float]:
    logit1 = log_prob.get(tokens[0], None)
    logit2 = log_prob.get(tokens[1], None)
    if logit1 is None:
        rating = -100
        LOGGER.warning(f'Missing token0 {tokens[0]} in log_prob, setting rating to -100.0')
    else:
        rating = logit1
    if logit1 is None:
        logit1 = -100
    if logit2 is None:
        logit2 = -100
    return (rating, logit1, logit2)

# Create log_prob dict with string keys mapping to float values
log_prob = {
    "token_a": 0.5,
    "token_b": -0.3,
    "token_c": 1.2,
}

# Create tokens list with 2 elements
tokens = ["token_a", "token_b"]

ARGS = {
    "log_prob": log_prob,
    "tokens": tokens,
}
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 tuple, ndim 1, dtype family float — 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
count_tokens Ran
ran, honoured the contract we drafted
ucl-dark/llm_debate/core/llm_api/anthropic_llm.py
code served (permissive licence) · get_code("56346bcde59d7a4a")
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.

import torch
import numpy as np

torch.manual_seed(0)

# Analyze the function:
# count_tokens takes a prompt (str) and returns len(prompt.split())
# It only needs a string parameter

prompt = "hello world this is a test prompt"

ARGS = {
    "prompt": prompt
}
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 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
load_secrets Ran
ran; our contract draft was wrong, not the code
ucl-dark/llm_debate/core/utils.py
pointer only (licence: MIT); the stored copy get_code serves, from sbp354/future_triggered_backdoors, is pointer only (licence: NONE) · get_code("5114a76c19ed4e1f")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from sbp354/future_triggered_backdoors, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

load_yaml Ran
ran; our contract draft was wrong, not the code
ucl-dark/llm_debate/core/utils.py
code served (permissive licence) · get_code("a055cc7323a5cf50")
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 yaml
import tempfile
import os

# Create a temporary YAML file with test content
temp_dir = tempfile.mkdtemp()
yaml_file = os.path.join(temp_dir, "test.yaml")

test_data = {
    "name": "test",
    "values": [1, 2, 3],
    "nested": {
        "key": "value"
    }
}

with open(yaml_file, 'w') as f:
    yaml.dump(test_data, f)

ARGS = {
    "file_path": yaml_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
yaml — 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
load_yaml_cached Ran
ran; our contract draft was wrong, not the code
ucl-dark/llm_debate/core/utils.py
code served (permissive licence) · get_code("4ae8e4de76e4be99")
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 tempfile
import yaml
import os

# Create a temporary YAML file with some test content
temp_dir = tempfile.mkdtemp()
yaml_file = os.path.join(temp_dir, "test_config.yaml")

test_data = {
    "model": "test_model",
    "learning_rate": 0.001,
    "batch_size": 32,
    "layers": [1, 2, 3]
}

with open(yaml_file, 'w') as f:
    yaml.dump(test_data, f)

ARGS = {
    "file_path": yaml_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
yaml — 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
pair_players Ran
ran; our contract draft was wrong, not the code
ucl-dark/llm_debate/core/swiss_tournament.py
code served (permissive licence) · get_code("f9349b8202644fd7")
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 2, 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
from typing import Callable, Dict, List, Tuple

# Set seed for reproducibility
torch.manual_seed(0)

# Construct arguments for pair_players function
names = ["Alice", "Bob", "Charlie", "Diana"]

previous_matchups = {
    "Alice": {"Bob"},
    "Bob": {"Alice"},
    "Charlie": set(),
    "Diana": set()
}

ARGS = {
    "names": names,
    "previous_matchups": previous_matchups
}
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 2, 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
play_round Ran
ran on a synthesized input
ucl-dark/llm_debate/core/swiss_tournament.py
code served (permissive licence) · get_code("e176ec2bc15a5a9f")
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 2, 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
from typing import Callable, Dict, List, Tuple
import tempfile
import os

# Create a temporary directory for tournament results
tournament_dir = tempfile.mkdtemp()

# Define a simple run_match callable
def run_match(player_a_data: Tuple[str, str, dict], player_b_data: Tuple[str, str, dict]) -> Tuple[int, int]:
    """
    Mock match function that returns (wins_for_player_a_correct, wins_for_player_b_incorrect)
    Returns two integers representing match results.
    """
    # Simple deterministic mock: return fixed wins
    return (3, 2)

# Define player data
# Each player is a tuple of (name_str, model_str, config_dict)
players = {
    "player_1": ("model_v1", "checkpoint_1", {"param_a": 0.5, "param_b": 0.3}),
    "player_2": ("model_v2", "checkpoint_2", {"param_a": 0.6, "param_b": 0.4}),
    "player_3": ("model_v3", "checkpoint_3", {"param_a": 0.7, "param_b": 0.5}),
}

# Define pairings for the round
pairings = [
    ("player_1", "player_2"),
    ("player_2", "player_3"),
    ("player_1", None),  # Bye round
]

# Construct ARGS dictionary
ARGS = {
    "pairings": pairings,
    "players": players,
    "run_match": run_match,
    "tournament_dir": tournament_dir,
    "round_num": 0,
}
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 2, 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
price_per_token Ran
ran, honoured the contract we drafted
ucl-dark/llm_debate/core/llm_api/anthropic_llm.py
code served (permissive licence) · get_code("f8bbc47c2b491933")
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 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

ARGS = {
    "model_id": "gpt-4",
}
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 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
price_per_token Ran
ran, honoured the contract we drafted
ucl-dark/llm_debate/core/llm_api/openai_llm.py
code served (permissive licence) · get_code("ea74105eb20142a0")
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 tuple, 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

# The function price_per_token only takes a model_id string parameter
# and returns pricing tuples based on matching the model_id against known patterns.
# No complex objects needed.

ARGS = {
    'model_id': 'gpt-4-1106-preview'
}
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 tuple, ndim 1, dtype family float — 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
swiss_tournament Ran
ran; our contract draft was wrong, not the code
ucl-dark/llm_debate/core/swiss_tournament.py
code served (permissive licence) · get_code("c0858383ecfb0a10")
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 tuple. 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 json
import math
from typing import Callable, Dict, List, Tuple

def mock_run_match(player_a: Tuple[str, str, dict], player_b: Tuple[str, str, dict]) -> Tuple[int, int]:
    """Mock match runner that returns (wins_correct, wins_incorrect)"""
    import random
    random.seed(hash(player_a[0] + player_b[0]) % (2**32))
    wins_a_correct = random.randint(0, 10)
    wins_b_incorrect = random.randint(0, 10)
    return (wins_a_correct, wins_b_incorrect)

ARGS = {
    "players": {
        "alice": ("alice_model", "alice_config", {"skill": 0.8}),
        "bob": ("bob_model", "bob_config", {"skill": 0.7}),
        "charlie": ("charlie_model", "charlie_config", {"skill": 0.9}),
        "diana": ("diana_model", "diana_config", {"skill": 0.6}),
    },
    "run_match": mock_run_match,
    "tournament_dir": None,
}
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 tuple — a tuple 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
count_tokens Not yet run
unverified (harvested, no recorded run)
ucl-dark/llm_debate/core/llm_api/openai_llm.py
code served (permissive licence) · get_code("3b94d7f86b74da2b")
How it ran

no invocation recorded.

make_exp_dir Not yet run
unverified (harvested, no recorded run)
ucl-dark/llm_debate/core/tournament.py
code served (permissive licence) · get_code("e5748ed4cebfd681")
How it ran

no invocation recorded.

Also present in other papers' code

Of 13 samples lifted from this paper's linked repositories, 13 carry a function identity and 0 do not. Of the 13 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 1 later) and 12 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.

FunctionAlso present inEarliest-datedOthers named
load_secrets 2 other papers
1 dated earlier, 1 later
2311.08702 (2023-11)
Debate Helps Supervise Unreliable Experts
2407.04108 (2024-07)
Future Events as Backdoor Triggers: Investigating Temporal Vulnerabilities in LLMs

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

Common methods for aligning large language models (LLMs) with desired behaviour heavily rely on human-labelled data. However, as models grow increasingly sophisticated, they will surpass human expertise, and the role of human evaluation will evolve into non-experts overseeing experts. In anticipation of this, we ask: can weaker models assess the correctness of stronger models? We investigate this question in an analogous setting, where stronger models (experts) possess the necessary information to answer questions and weaker models (non-experts) lack this information. The method we evaluate is debate, where two LLM experts each argue for a different answer, and a non-expert selects the answer. We find that debate consistently helps both non-expert models and humans answer questions, achieving 76% and 88% accuracy respectively (naive baselines obtain 48% and 60%). Furthermore, optimising expert debaters for persuasiveness in an unsupervised manner improves non-expert ability to identify the truth in debates. Our results provide encouraging empirical evidence for the viability of aligning models with debate in the absence of ground truth.

For agents

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

get_harvested_code_for_paper("2402.06782")
get_code_for_paper("2402.06782")
have("2402.06782")

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

curl https://syntology.ai/api/ran/2402.06782.json

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

[![Syntology run record](https://syntology.ai/api/ran/2402.06782.svg)](https://syntology.ai/paper/2402.06782)

Connect an agent — have() is free.