SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2402.12366 · 2024

A Critical Evaluation of AI Feedback for Aligning Large Language Models

arXiv · PDF · Open in the Atlas

Code that ran

9 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; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 9 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/2402.12366.json · badge: /api/ran/2402.12366.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
architsharma97/dpo-rlaif canonical (classifier label)
classifier basis: census
9 ran · 2 no recorded run
FunctionStatusWhere it lives
dpo_loss Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
architsharma97/dpo-rlaif/trainers.py
code served (permissive licence) · get_code("16a2450aff4a3577")
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 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 torch.nn.functional as F

torch.manual_seed(0)

# Create tiny batch tensors with shape (batch_size,)
batch_size = 2

policy_chosen_logps = torch.randn(batch_size, dtype=torch.float32)
policy_rejected_logps = torch.randn(batch_size, dtype=torch.float32)
reference_chosen_logps = torch.randn(batch_size, dtype=torch.float32)
reference_rejected_logps = torch.randn(batch_size, dtype=torch.float32)

ARGS = {
    "policy_chosen_logps": policy_chosen_logps,
    "policy_rejected_logps": policy_rejected_logps,
    "reference_chosen_logps": reference_chosen_logps,
    "reference_rejected_logps": reference_rejected_logps,
    "beta": 0.1,
    "reference_free": False,
    "importance_correction": False,
    "ipo": False,
    "robust_eps": 0.0,
}
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 2, 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
extract_anthropic_prompt Ran
ran; our contract draft was wrong, not the code
architsharma97/dpo-rlaif/preference_datasets.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from junkangwu/beta-DPO, is pointer only (licence: NONE) · get_code("12f6803880f9b17e")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from junkangwu/beta-DPO, 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.

get_local_dir Ran
ran; our contract draft was wrong, not the code
architsharma97/dpo-rlaif/utils.py
code served (permissive licence) · get_code("2a9c061a2b1c35cf")
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 os
from typing import Dict, Union, Type, List

# The function get_local_dir takes a list of strings as prefixes_to_resolve
# Looking at the body, it doesn't actually use the prefixes_to_resolve parameter
# It just calls os.getenv to get PROJECT_CACHE or returns a default cache path
# So we just need to provide a valid list of strings

ARGS = {
    'prefixes_to_resolve': ['prefix1', 'prefix2', 'prefix3']
}
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
get_local_run_dir Ran
ran; our contract draft was wrong, not the code
architsharma97/dpo-rlaif/utils.py
code served (permissive licence) · get_code("5edbbf974035e91f")
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 os
from datetime import datetime
from typing import List

# Set up environment
os.environ['PROJECT_CACHE'] = '/tmp/test_cache'

# Define arguments for get_local_run_dir
ARGS = {
    'exp_name': 'test_experiment',
    'local_dirs': ['/tmp/cache1', '/tmp/cache2']
}
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
get_logits Ran
ran; our contract draft was wrong, not the code
architsharma97/dpo-rlaif/reward_trainer.py
code served (permissive licence) · get_code("3c8dcada6a55875f")
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 torch
import torch.nn as nn

torch.manual_seed(0)

class SimpleModel(nn.Module):
    def __init__(self, vocab_size=100, hidden_size=4):
        super().__init__()
        self.embedding = nn.Embedding(vocab_size, hidden_size)
        self.linear = nn.Linear(hidden_size, vocab_size)
    
    def forward(self, input_ids, attention_mask=None):
        x = self.embedding(input_ids)
        logits = self.linear(x)
        return {'logits': logits}

model = SimpleModel(vocab_size=100, hidden_size=4)

batch = {
    'chosen_input_ids': torch.randint(0, 100, (2, 3)),
    'chosen_attention_mask': torch.ones((2, 3), dtype=torch.long),
    'rejected_input_ids': torch.randint(0, 100, (2, 3)),
    'rejected_attention_mask': torch.ones((2, 3), dtype=torch.long),
}

ARGS = {
    'model': model,
    'batch': batch,
}
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 — 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
get_openai_completion Ran
ran; our contract draft was wrong, not the code
architsharma97/dpo-rlaif/ai_completions.py
code served (permissive licence) · get_code("2378ff2496a12f0a")
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 os

# Mock the _openai_chat_completion function since we can't make real API calls
def _openai_chat_completion(model, messages, no_cache):
    # Return a mock response matching the expected structure
    return {
        'choices': [
            {
                'message': {
                    'content': 'This is a mock completion response.'
                },
                'finish_reason': 'stop'
            }
        ]
    }

# Inject the mock into the module namespace
import sys
current_module = sys.modules[__name__]
setattr(current_module, '_openai_chat_completion', _openai_chat_completion)

# Define arguments for get_openai_completion
ARGS = {
    'prompt': 'What is the capital of France?',
    'cache': True,
    'model': 'gpt-4-0314',
    'system_prompt': 'You are a helpful assistant.'
}
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
get_remote_file Ran
ran; our contract draft was wrong, not the code
architsharma97/dpo-rlaif/utils.py
code served (permissive licence) · get_code("e7ce4c19888c6df6")
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 os
import socket
import tempfile

# Create a temporary directory for testing
temp_dir = tempfile.mkdtemp()

# Get the current hostname
current_hostname = socket.gethostname()

# Create a test local path
test_local_path = os.path.join(temp_dir, "test_file.txt")

# Create a remote path string - using local hostname to avoid actual SCP calls
# Format is "hostname:path"
remote_path = f"{current_hostname}:/tmp/test_remote_file.txt"

ARGS = {
    "remote_path": remote_path,
    "local_path": test_local_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
match_instruction_outputs Ran
ran; our contract draft was wrong, not the code
architsharma97/dpo-rlaif/label_ai_preferences.py
code served (permissive licence) · get_code("5530a963e205ad95")
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 torch.nn as nn

torch.manual_seed(0)

# Analyzing the function:
# - instruct_out_1: list of dicts, each with 'output' key
# - instruction_set_1: list of instructions
# - instruct_out_2: list of dicts, each with 'output' key
# - instruction_set_2: list of instructions
# The function matches instructions between two sets and collects outputs

# Create instruction sets with some overlap
instruction_set_1 = ['add', 'subtract', 'multiply', 'divide']
instruction_set_2 = ['subtract', 'multiply', 'power', 'divide']

# Create output lists - same length as their corresponding instruction sets
instruct_out_1 = [
    {'output': torch.randn(2, 4)},
    {'output': torch.randn(2, 4)},
    {'output': torch.randn(2, 4)},
    {'output': torch.randn(2, 4)},
]

instruct_out_2 = [
    {'output': torch.randn(2, 4)},
    {'output': torch.randn(2, 4)},
    {'output': torch.randn(2, 4)},
    {'output': torch.randn(2, 4)},
]

ARGS = {
    'instruct_out_1': instruct_out_1,
    'instruction_set_1': instruction_set_1,
    'instruct_out_2': instruct_out_2,
    'instruction_set_2': instruction_set_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 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
train_step Ran
ran; our contract draft was wrong, not the code
architsharma97/dpo-rlaif/reward_trainer.py
code served (permissive licence) · get_code("12be818ac399701d")
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 torch.nn as nn
import torch.optim as optim

torch.manual_seed(0)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, input_ids, attention_mask):
        # input_ids: [batch_size, seq_len]
        # attention_mask: [batch_size, seq_len]
        # Output logits: [batch_size, seq_len, vocab_size]
        batch_size = input_ids.shape[0]
        seq_len = input_ids.shape[1]
        logits = self.linear(input_ids.float())  # [batch_size, seq_len, 3]
        return {'logits': logits}

model = SimpleModel()
optimizer = optim.Adam(model.parameters(), lr=0.001)

batch = {
    'chosen_input_ids': torch.randint(0, 10, (2, 4)),
    'chosen_attention_mask': torch.ones(2, 4),
    'rejected_input_ids': torch.randint(0, 10, (2, 4)),
    'rejected_attention_mask': torch.ones(2, 4),
}

ARGS = {
    'model': model,
    'batch': batch,
    'optimizer': optimizer,
}
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
dump_files Not yet run
unverified (harvested, no recorded run)
architsharma97/dpo-rlaif/generate_samples.py
code served (permissive licence) · get_code("d5e9b37690fe8587")
How it ran

no invocation recorded.

eval_loop Not yet run
unverified (harvested, no recorded run)
architsharma97/dpo-rlaif/reward_trainer.py
code served (permissive licence) · get_code("b1c3a18860528ce9")
How it ran

no invocation recorded.

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, 4 are also present in other papers whose code we hold (3 with at least one paper dated earlier, 3 the same month, 4 later) and 7 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
extract_anthropic_prompt 15 other papers
5 dated earlier, 1 the same month, 9 later
2309.16240 (2023-09)
Beyond Reverse KL: Generalizing Direct Preference Optimization with Diverse Divergence Constraints
2310.08164 (2023-10)
Interpreting Learned Feedback Patterns in Large Language Models
2311.08380 (2023-11)
Direct Preference Optimization for Neural Machine Translation with Minimum Bayes Risk Decoding
and 12 more not named here
get_remote_file 15 other papers
4 dated earlier, 3 the same month, 8 later
2309.16240 (2023-09)
Beyond Reverse KL: Generalizing Direct Preference Optimization with Diverse Divergence Constraints
2312.00267 (2023-12)
Sample Efficient Preference Alignment in LLMs via Active Exploration
2401.01967 (2024-01)
A Mechanistic Understanding of Alignment Algorithms: A Case Study on DPO and Toxicity
and 12 more not named here
get_local_run_dir 11 other papers
3 dated earlier, 2 the same month, 6 later
2309.16240 (2023-09)
Beyond Reverse KL: Generalizing Direct Preference Optimization with Diverse Divergence Constraints
2312.00267 (2023-12)
Sample Efficient Preference Alignment in LLMs via Active Exploration
2401.13275 (2024-01)
Can AI Assistants Know What They Don't Know?
and 8 more not named here
dpo_loss 1 other paper
1 later
2406.16797 (2024-06)
Lottery Ticket Adaptation: Mitigating Destructive Interference 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

Reinforcement learning with AI feedback (RLAIF) is a popular paradigm for improving the instruction-following abilities of powerful pre-trained language models. RLAIF first performs supervised fine-tuning (SFT) using demonstrations from a teacher model and then further fine-tunes the model with reinforcement learning (RL), using feedback from a critic model. While recent popular open-source models have demonstrated substantial improvements in performance from the RL step, in this paper we question whether the complexity of this RL step is truly warranted for AI feedback. We show that the improvements of the RL step are virtually entirely due to the widespread practice of using a weaker teacher model (e.g. GPT-3.5) for SFT data collection than the critic (e.g., GPT-4) used for AI feedback generation. Specifically, we show that simple supervised fine-tuning with GPT-4 as the teacher outperforms existing RLAIF pipelines. More generally, we find that the gains from RLAIF vary substantially across base model families, test-time evaluation protocols, and critic models. Finally, we provide a mechanistic explanation for when SFT may outperform the full two-step RLAIF pipeline as well as suggestions for making RLAIF maximally useful in practice.

For agents

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

get_harvested_code_for_paper("2402.12366")
get_code_for_paper("2402.12366")
have("2402.12366")

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

curl https://syntology.ai/api/ran/2402.12366.json

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

[![Syntology run record](https://syntology.ai/api/ran/2402.12366.svg)](https://syntology.ai/paper/2402.12366)

Connect an agent — have() is free.