SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2404.00530 · 2024

Comparing Bad Apples to Good Oranges: Aligning Large Language Models via Joint Preference Optimization

arXiv · PDF · Open in the Atlas

Code that ran

11 of 11 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, 1 violated it and 0 have no contract verdict; 9 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/2404.00530.json · badge: /api/ran/2404.00530.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
hritikbansal/dove canonical (classifier label)
classifier basis: census
11 ran · 0 no recorded run
FunctionStatusWhere it lives
allocate Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
hritikbansal/dove/sft/core/multipack_sampler.py
code served (permissive licence) · get_code("cee7785fcfddc8ae")
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 numpy as np
import numba

# Analyze the function signature:
# allocate(lengths: np.ndarray, lengths_cumsum: np.ndarray, rank: int, c: int, n: int)

# lengths: used for indexing and passed to ffd_check/ffd_with_result
# lengths_cumsum: used with np.searchsorted
# rank: used as index into batch result
# c: capacity parameter
# n: number of bins parameter

# Create small test data
lengths = np.array([5, 3, 7, 2, 4, 6], dtype=np.int64)
lengths_cumsum = np.cumsum(lengths)

rank = 0
c = 15  # bin capacity
n = 2   # number of bins

ARGS = {
    "lengths": lengths,
    "lengths_cumsum": lengths_cumsum,
    "rank": rank,
    "c": c,
    "n": n,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numba, numpy — 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
ffd_check Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
hritikbansal/dove/sft/core/multipack_sampler.py
code served (permissive licence) · get_code("5c2e6ebfd6b20c74")
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 bool, 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 numpy as np
import numba

# Analyze the function:
# - a: np.ndarray of numeric values (will be sorted and used as sizes)
# - c: int, capacity for bins
# - n: int, number of bins
# The function implements First Fit Decreasing (FFD) bin packing

# Create valid test arguments
a = np.array([5, 3, 7, 2, 4, 1], dtype=np.float64)  # items to pack
c = 10  # bin capacity
n = 3   # number of bins

ARGS = {
    "a": a,
    "c": c,
    "n": n
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numba, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, 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
ffd_with_result Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
hritikbansal/dove/sft/core/multipack_sampler.py
code served (permissive licence) · get_code("d0405e95f05e1f64")
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 numpy as np
import numba

# Examine the function to understand what it needs:
# - a: np.ndarray - an array of numeric values (sizes)
# - c: int - bin capacity
# - start_index: int - offset for indices

# Create a small test array of sizes
a = np.array([5.0, 3.0, 2.0, 4.0, 1.0], dtype=np.float64)

# Bin capacity
c = 10

# Starting index offset
start_index = 0

ARGS = {
    "a": a,
    "c": c,
    "start_index": start_index,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numba, numpy — 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
filter_long_samples Ran
ran, violated the contract we drafted
hritikbansal/dove/sft/core/supervised_dataset.py
code served (permissive licence) · get_code("f0832ee83b5ff758")
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 transformers
from transformers import AutoTokenizer

torch.manual_seed(0)

# Create a minimal stand-in tokenizer
class MinimalTokenizer:
    def __init__(self):
        self.model_max_length = 128
        self.eos_token = "</s>"
    
    def __call__(self, text, return_tensors=None, padding=False, max_length=None, truncation=False):
        # Simple mock tokenization: split text and create token IDs
        tokens = text.split()
        token_ids = [hash(t) % 1000 for t in tokens]  # Simple hash-based token IDs
        
        if truncation and max_length and len(token_ids) > max_length:
            token_ids = token_ids[:max_length]
        
        input_ids = torch.tensor([token_ids], dtype=torch.long)
        
        if return_tensors == 'pt':
            return type('TokenizerOutput', (), {'input_ids': input_ids})()
        return input_ids

tokenizer = MinimalTokenizer()

# Create sample data with 'instruction' and 'response' keys
samples = {
    'instruction': [
        'What is 2+2?',
        'Tell me a joke.'
    ],
    'response': [
        'The answer is 4.',
        'Why did the chicken cross the road? To get to the other side!'
    ]
}

ARGS = {
    'samples': samples,
    'tokenizer': tokenizer
}
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 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
filter_long_sequences Ran
ran; our contract draft was wrong, not the code
hritikbansal/dove/jpo/utils.py
code served (permissive licence) · get_code("df4f346585f694bb")
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 Dataset. 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
import numpy as np
from datasets import Dataset

torch.manual_seed(0)

# Create a mock dataset with the required fields
data = {
    'prompt_chosen': ['hello world', 'short', 'a very long prompt that exceeds the maximum length allowed by the filter function'],
    'response_chosen': ['response one', 'resp', 'x' * 600],
    'prompt_rejected': ['prompt rej', 'p', 'another long prompt text'],
    'response_rejected': ['rejection response', 'r', 'y' * 100]
}

dataset = Dataset.from_dict(data)

ARGS = {
    'dataset': dataset,
    'max_length': 512
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
datasets, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Dataset — a Dataset output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Dataset output the recorded ndim 1 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
fmt_prompt Ran
ran; our contract draft was wrong, not the code
hritikbansal/dove/jpo/utils.py
code served (permissive licence) · get_code("d12d009f6f16a459")
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

torch.manual_seed(0)

# The function fmt_prompt takes a single string parameter 'prompt'
# and returns a formatted string with instructions and response markers.
# No complex objects needed - just a simple string.

ARGS = {
    'prompt': 'What is the capital of France?'
}
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
fmt_prompt Ran
ran; our contract draft was wrong, not the code
hritikbansal/dove/sft/core/supervised_dataset.py
code served (permissive licence) · get_code("dc2a8a5f4eb6f801")
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

def fmt_prompt(prompt):
    return f'### Instructions:{prompt}\n\n### Response:'

# The function takes a single parameter 'prompt' which should be a string
# It formats the prompt with instructions prefix and response suffix
prompt_text = "What is the capital of France?"

ARGS = {
    "prompt": prompt_text
}
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
get_feedback Ran
ran, honoured the contract we drafted
hritikbansal/dove/jpo/ai_feedback.py
code served (permissive licence) · get_code("841ccc2439e64906")
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

torch.manual_seed(0)

# The function checks for string patterns '(a)' and '(b)' in two feedback strings
# and returns an integer (0, 1, 2) or None based on the pattern combinations

feedback_1 = "This is option (a)"
feedback_2 = "This is option (b)"

ARGS = {
    "feedback_1": feedback_1,
    "feedback_2": feedback_2,
}
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 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
get_feedback Ran
ran; our contract draft was wrong, not the code
hritikbansal/dove/jpo/auto_eval.py
code served (permissive licence) · get_code("4abdf461254ec7ab")
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
import numpy as np

torch.manual_seed(0)

# Analyze the function: it takes two feedback strings and checks for patterns like '(a)' and '(b)'
# The function performs string membership checks using 'in' operator
# Valid inputs are strings containing patterns like '(a)' and/or '(b)'

feedback_1 = "This is option (a) which is better"
feedback_2 = "This is option (b) which is worse"

ARGS = {
    "feedback_1": feedback_1,
    "feedback_2": feedback_2,
}
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 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
preprocess Ran
ran; our contract draft was wrong, not the code
hritikbansal/dove/sft/core/supervised_dataset.py
code served (permissive licence) · get_code("f1d264774d681d67")
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
import transformers
from typing import Dict, Sequence

torch.manual_seed(0)

# Create a minimal stand-in tokenizer
class MockTokenizer:
    def __init__(self):
        self.model_max_length = 512
        self.pad_token_id = 0
        self.eos_token = "</s>"
    
    def __call__(self, text, return_tensors=None, padding=False, max_length=None, truncation=False):
        # Simple mock: create token ids based on text length
        tokens = list(range(1, len(text.split()) + 2))  # Simple tokenization
        input_ids = torch.tensor([tokens], dtype=torch.long)
        
        class MockTokenizerOutput:
            def __init__(self, input_ids_tensor):
                self.input_ids = input_ids_tensor
            
            def ne(self, pad_token_id):
                # Return a mask of non-padding tokens
                return self.input_ids != pad_token_id
        
        return MockTokenizerOutput(input_ids)

tokenizer = MockTokenizer()

# Create sample data with 'instruction' and 'response' keys
samples = {
    'instruction': [
        'What is 2+2?',
        'What is the capital of France?'
    ],
    'response': [
        'The answer is 4.',
        'The capital of France is Paris.'
    ]
}

ARGS = {
    'train_on_inputs': False,
    'samples': samples,
    'tokenizer': tokenizer
}
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 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
return_prompt_and_responses_augmented Ran
ran; our contract draft was wrong, not the code
hritikbansal/dove/jpo/utils.py
code served (permissive licence) · get_code("129ed5f0939a11d7")
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
import numpy as np

torch.manual_seed(0)

# Construct the samples argument
# The function expects samples to be a dict-like object with keys:
# 'i_chosen', 'r_chosen', 'i_reject', 'r_reject'

samples = {
    'i_chosen': [
        'What is the capital of France?',
        'How do you make coffee?'
    ],
    'r_chosen': [
        'Paris is the capital of France.',
        'To make coffee, you need ground coffee beans and hot water.'
    ],
    'i_reject': [
        'What is the capital of France?',
        'How do you make coffee?'
    ],
    'r_reject': [
        'The capital is in Europe.',
        'You need water and beans.'
    ]
}

ARGS = {
    'samples': samples
}
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 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

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 distinct functions with an identity, 11 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 11 later) and 0 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
allocate 1 other paper
1 later
2025.findings-acl.39 (2025 (year only))
Comparing Bad Apples to Good Oranges: Aligning Large Language Models via Joint Preference Optimization
ffd_check 1 other paper
1 later
2025.findings-acl.39 (2025 (year only))
Comparing Bad Apples to Good Oranges: Aligning Large Language Models via Joint Preference Optimization
ffd_with_result 1 other paper
1 later
2025.findings-acl.39 (2025 (year only))
Comparing Bad Apples to Good Oranges: Aligning Large Language Models via Joint Preference Optimization
filter_long_samples 1 other paper
1 later
2025.findings-acl.39 (2025 (year only))
Comparing Bad Apples to Good Oranges: Aligning Large Language Models via Joint Preference Optimization
filter_long_sequences 1 other paper
1 later
2025.findings-acl.39 (2025 (year only))
Comparing Bad Apples to Good Oranges: Aligning Large Language Models via Joint Preference Optimization
fmt_prompt 1 other paper
1 later
2025.findings-acl.39 (2025 (year only))
Comparing Bad Apples to Good Oranges: Aligning Large Language Models via Joint Preference Optimization
fmt_prompt 1 other paper
1 later
2025.findings-acl.39 (2025 (year only))
Comparing Bad Apples to Good Oranges: Aligning Large Language Models via Joint Preference Optimization
get_feedback 1 other paper
1 later
2025.findings-acl.39 (2025 (year only))
Comparing Bad Apples to Good Oranges: Aligning Large Language Models via Joint Preference Optimization
get_feedback 1 other paper
1 later
2025.findings-acl.39 (2025 (year only))
Comparing Bad Apples to Good Oranges: Aligning Large Language Models via Joint Preference Optimization
preprocess 1 other paper
1 later
2025.findings-acl.39 (2025 (year only))
Comparing Bad Apples to Good Oranges: Aligning Large Language Models via Joint Preference Optimization
return_prompt_and_responses_augmented 1 other paper
1 later
2025.findings-acl.39 (2025 (year only))
Comparing Bad Apples to Good Oranges: Aligning Large Language Models via Joint Preference Optimization

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

A common technique for aligning large language models (LLMs) relies on acquiring human preferences by comparing multiple generations conditioned on a fixed context. This method, however, relies solely on pairwise comparisons, where the generations are evaluated within an identical context. While effective to such conditional preferences often fail to encompass the nuanced and multidimensional nature of human preferences. In this work, we revisit the traditional paradigm of preference acquisition and propose a new axis based on eliciting preferences jointly over the instruction-response pairs. Unlike prior preference optimizations, which are designed for conditional ranking protocols (e.g., DPO), we propose Joint Preference Optimization (JPO), a new preference optimization objective that upweights the joint probability of the chosen instruction-response pair over the rejected instruction-response pair. Interestingly, LLMs trained with joint instruction-response preference data using JPO outperform LLM trained with DPO by $5.2\%$ and $3.3\%$ win-rate for summarization and open-ended dialogue datasets, respectively. Our findings reveal that joint preferences over instruction and response pairs can significantly enhance the alignment of LLMs by tapping into a broader spectrum of human preference elicitation. The data and code is available at https://github.com/Hritikbansal/dove.

For agents

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

get_harvested_code_for_paper("2404.00530")
get_code_for_paper("2404.00530")
have("2404.00530")

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

curl https://syntology.ai/api/ran/2404.00530.json

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

[![Syntology run record](https://syntology.ai/api/ran/2404.00530.svg)](https://syntology.ai/paper/2404.00530)

Connect an agent — have() is free.