SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2305.14947 · 2023

How Predictable Are Large Language Model Capabilities? A Case Study on BIG-bench

arXiv · PDF · Open in the Atlas

Code that ran

12 of 15 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 12 have no contract verdict; 0 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/2305.14947.json · badge: /api/ran/2305.14947.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
ink-usc/predicting-big-bench canonical (classifier label)
classifier basis: census
12 ran · 3 no recorded run
FunctionStatusWhere it lives
crossover Ran
ran on a synthesized input
ink-usc/predicting-big-bench/code/search/genetic_algorithm.py
code served (permissive licence) · get_code("8a47bb5c2d961c22")
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 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 random

def fill_random_to_length(individual, subtasks, target_len):
    subtasks_not_in_individual = set(subtasks) - set(individual)
    new_selected_tasks = random.sample(list(subtasks_not_in_individual), target_len - len(individual))
    individual = individual + new_selected_tasks
    assert len(individual) == target_len
    return individual

def crossover(parent1, parent2, subtasks):
    n = len(parent1)
    crossover_point = random.randint(0, n - 1)
    random.shuffle(parent1)
    random.shuffle(parent2)
    child1 = parent1[:crossover_point] + parent2[crossover_point:]
    child2 = parent2[:crossover_point] + parent1[crossover_point:]
    child1 = fill_random_to_length(child1, subtasks, n)
    child2 = fill_random_to_length(child2, subtasks, n)
    return (child1, child2)

random.seed(42)

parent1 = [1, 2, 3, 4, 5]
parent2 = [6, 7, 8, 9, 10]
subtasks = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]

ARGS = {
    "parent1": parent1,
    "parent2": parent2,
    "subtasks": subtasks
}
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 2, 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
eval_ Ran
ran on a synthesized input
ink-usc/predicting-big-bench/code/models/mlp_models.py
code served (permissive licence) · get_code("89ca1fdd7d6460f3")
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. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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 numpy as np

torch.manual_seed(0)
np.random.seed(0)

# Define a minimal model that satisfies the function's usage
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

# Create tiny tensors for evaluation
# X_eval: batch_size=2, features=4
# y_eval: batch_size=2, outputs=3
X_eval = torch.randn(2, 4)
y_eval = torch.randn(2, 3)

# Create the model
model = SimpleModel()

# Construct ARGS dictionary
ARGS = {
    "model": model,
    "X_eval": X_eval,
    "y_eval": y_eval,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn, 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
fill_random_to_length Ran
ran on a synthesized input
ink-usc/predicting-big-bench/code/search/genetic_algorithm.py
code served (permissive licence) · get_code("f92dad82a57ddd5a")
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 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 random

# Set seed for reproducibility
random.seed(42)

# Construct arguments for fill_random_to_length
individual = [1, 2, 3]  # A list of selected task IDs
subtasks = [1, 2, 3, 4, 5, 6, 7, 8]  # A list of all available subtasks
target_len = 6  # Target length we want the individual to reach

# Verify that target_len > len(individual) and there are enough unselected tasks
assert target_len > len(individual), "target_len must be greater than current individual length"
assert len(set(subtasks) - set(individual)) >= target_len - len(individual), "Not enough subtasks available"

ARGS = {
    "individual": individual,
    "subtasks": subtasks,
    "target_len": target_len,
}
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 1, dtype family int — 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
generate_random_combinations Ran
ran on a synthesized input
ink-usc/predicting-big-bench/code/models/utils.py
code served (permissive licence) · get_code("1940d9b8ebfc16e7")
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 random

def generate_random_combinations(param_distributions, n_trials, seed=42):
    num_combinations = 1
    for values in param_distributions.values():
        num_combinations *= len(values)
    assert n_trials <= num_combinations
    state = random.getstate()
    random.seed(seed)
    samples = set()
    hp_list = []
    while len(samples) < n_trials:
        sample = {}
        for key, values in param_distributions.items():
            sample[key] = random.choice(values)
        if tuple(sample.items()) not in samples:
            samples.add(tuple(sample.items()))
            hp_list.append(sample)
    return hp_list


# Setup code
param_distributions = {
    'learning_rate': [0.001, 0.01, 0.1],
    'batch_size': [16, 32, 64],
    'optimizer': ['adam', 'sgd', 'rmsprop'],
    'activation': ['relu', 'tanh']
}

n_trials = 5

seed = 42

ARGS = {
    'param_distributions': param_distributions,
    'n_trials': n_trials,
    'seed': seed
}
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 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
mutation Ran
ran on a synthesized input
ink-usc/predicting-big-bench/code/search/genetic_algorithm.py
code served (permissive licence) · get_code("1a2bfcf44ccd41c5")
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 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 random

# Set seed for reproducibility
random.seed(0)

# Define the arguments for the mutation function
# individual: a list of selected subtasks (represented as integers)
# subtasks: a list/set of all available subtasks

individual = [1, 2, 3]  # Current selected tasks
subtasks = [1, 2, 3, 4, 5, 6, 7, 8]  # All available subtasks

ARGS = {
    "individual": individual,
    "subtasks": subtasks
}
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 1, dtype family int — 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
reorder_surprise_preds Ran
ran on a synthesized input
ink-usc/predicting-big-bench/code/models/surprise_utils.py
code served (permissive licence) · get_code("9be5b16b823daf48")
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 ndarray, ndim 1, dtype family float, shape (4,) 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 pandas as pd
import numpy as np

# Create a sample dataframe with model_unique and task_unique columns
df = pd.DataFrame({
    'model_unique': ['model_a', 'model_b', 'model_a', 'model_b'],
    'task_unique': ['task_1', 'task_1', 'task_2', 'task_2'],
    'other_col': [10, 20, 30, 40]
})

# Create predictions list with tuples of (model, task, _, y_pred, _)
predictions = [
    ('model_a', 'task_1', 'unused1', 0.85, 'unused2'),
    ('model_b', 'task_1', 'unused1', 0.92, 'unused2'),
    ('model_a', 'task_2', 'unused1', 0.78, 'unused2'),
    ('model_b', 'task_2', 'unused1', 0.88, 'unused2'),
]

ARGS = {
    'df': df,
    'predictions': predictions
}
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 ndarray, ndim 1, dtype family float, shape (4,) — shape (4,) 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
run_random_forest Ran
ran on a synthesized input
ink-usc/predicting-big-bench/code/models/random_forest.py
code served (permissive licence) · get_code("d6ea72044eb758bc")
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
from unittest.mock import MagicMock

# Create small datasets for training, dev, and test
np.random.seed(42)
n_train, n_dev, n_test = 10, 5, 5
n_features = 4

X_train = np.random.randn(n_train, n_features).astype(np.float32)
y_train = np.random.randn(n_train).astype(np.float32)

X_dev = np.random.randn(n_dev, n_features).astype(np.float32)
y_dev = np.random.randn(n_dev).astype(np.float32)

X_test = np.random.randn(n_test, n_features).astype(np.float32)
y_test = np.random.randn(n_test).astype(np.float32)

# Create a mock args object with hp attribute
class MockArgs:
    def __init__(self, hp=None):
        self.hp = hp

args = MockArgs(hp=None)

# Create a mock logger
logger = MagicMock()

ARGS = {
    "args": args,
    "logger": logger,
    "X_train": X_train,
    "y_train": y_train,
    "X_dev": X_dev,
    "y_dev": y_dev,
    "X_test": X_test,
    "y_test": y_test
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn — 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
run_xgb_regressor Ran
ran on a synthesized input
ink-usc/predicting-big-bench/code/models/xgb_regressor.py
code served (permissive licence) · get_code("791ba96633405dff")
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 logging

# Create a simple logger
logger = logging.getLogger(__name__)
logger.setLevel(logging.DEBUG)

# Create small synthetic datasets
np.random.seed(42)
n_train = 10
n_dev = 5
n_test = 5
n_features = 4

X_train = np.random.randn(n_train, n_features).astype(np.float32)
y_train = np.random.randn(n_train).astype(np.float32)

X_dev = np.random.randn(n_dev, n_features).astype(np.float32)
y_dev = np.random.randn(n_dev).astype(np.float32)

X_test = np.random.randn(n_test, n_features).astype(np.float32)
y_test = np.random.randn(n_test).astype(np.float32)

# Create args object with hp parameter
class Args:
    def __init__(self, hp=None):
        self.hp = hp

args = Args(hp=None)

ARGS = {
    'args': args,
    'logger': logger,
    'X_train': X_train,
    'y_train': y_train,
    'X_dev': X_dev,
    'y_dev': y_dev,
    'X_test': X_test,
    'y_test': y_test
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn, xgboost — 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
train Ran
ran on a synthesized input
ink-usc/predicting-big-bench/code/models/mlp_models.py
code served (permissive licence) · get_code("314ea4df358695cc")
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. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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
import numpy as np
from unittest.mock import MagicMock

torch.manual_seed(0)

# Build a minimal model that accepts input and produces output
class SimpleModel(nn.Module):
    def __init__(self, input_dim=4, output_dim=3):
        super().__init__()
        self.linear = nn.Linear(input_dim, output_dim)
    
    def forward(self, x):
        return self.linear(x)

# Build tiny tensors with distinct dimensions
# Input: batch_size=2, features=4
# Output: batch_size=2, targets=3
X_train = torch.randn(2, 4)
y_train = torch.randn(2, 3)
X_dev = torch.randn(2, 4)
y_dev = torch.randn(2, 3)

# Build a mock logger
logger = MagicMock()
logger.info = MagicMock()

# Build args object with required hyperparameters
class Args:
    def __init__(self):
        self.lr = 0.001
        self.weight_decay = 0.0
        self.batch_size = 2
        self.verbose = False

args = Args()

# Build model
model = SimpleModel(input_dim=4, output_dim=3)

# Define ARGS dict mapping all parameters of train function
ARGS = {
    'args': args,
    'logger': logger,
    'model': model,
    'X_train': X_train,
    'y_train': y_train,
    'X_dev': X_dev,
    'y_dev': y_dev,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn, torch, tqdm — 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
train_for_search Ran
ran on a synthesized input
ink-usc/predicting-big-bench/code/models/mlp_models.py
code served (permissive licence) · get_code("adbfdafafb77c80f")
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. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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
import numpy as np

torch.manual_seed(0)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

class SimpleLogger:
    def log(self, msg):
        pass

class Args:
    def __init__(self):
        self.lr = 0.001
        self.weight_decay = 0.0
        self.batch_size = 2
        self.verbose = False

# Create small tensors (batch size 2, feature dim 4, output dim 3)
X_train = torch.randn(2, 4)
y_train = torch.randn(2, 3)
X_dev = torch.randn(2, 4)
y_dev = torch.randn(2, 3)

# Create instances
args = Args()
logger = SimpleLogger()
model = SimpleModel()

ARGS = {
    'args': args,
    'logger': logger,
    'model': model,
    'X_train': X_train,
    'y_train': y_train,
    'X_dev': X_dev,
    'y_dev': y_dev,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn, torch, tqdm — 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
tune_random_forest Ran
ran on a synthesized input
ink-usc/predicting-big-bench/code/models/random_forest.py
code served (permissive licence) · get_code("56dc5cc27ac2b05e")
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 numpy as np
import logging

class Args:
    def __init__(self):
        self.n_trials = 3

# Create a simple logger
logger = logging.getLogger("test_logger")
logger.setLevel(logging.DEBUG)

# Create small synthetic datasets
np.random.seed(42)
n_train = 8
n_dev = 4
n_test = 3
n_features = 5

X_train = np.random.randn(n_train, n_features).astype(np.float32)
y_train = np.random.randn(n_train).astype(np.float32)

X_dev = np.random.randn(n_dev, n_features).astype(np.float32)
y_dev = np.random.randn(n_dev).astype(np.float32)

X_test = np.random.randn(n_test, n_features).astype(np.float32)
y_test = np.random.randn(n_test).astype(np.float32)

ARGS = {
    'args': Args(),
    'logger': logger,
    'X_train': X_train,
    'y_train': y_train,
    'X_dev': X_dev,
    'y_dev': y_dev,
    'X_test': X_test,
    'y_test': y_test
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn — 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
tune_xgb_regressor Ran
ran on a synthesized input
ink-usc/predicting-big-bench/code/models/xgb_regressor.py
code served (permissive licence) · get_code("11af8b257e4dafc7")
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 numpy as np
import logging

# Create a simple logger
logger = logging.getLogger(__name__)
logger.setLevel(logging.INFO)

# Create small numpy arrays for X and y
np.random.seed(0)
X_train = np.random.randn(10, 4).astype(np.float32)
y_train = np.random.randn(10).astype(np.float32)
X_dev = np.random.randn(5, 4).astype(np.float32)
y_dev = np.random.randn(5).astype(np.float32)
X_test = np.random.randn(3, 4).astype(np.float32)
y_test = np.random.randn(3).astype(np.float32)

# Create a simple args object with n_trials attribute
class Args:
    def __init__(self):
        self.n_trials = 2

args = Args()

ARGS = {
    'args': args,
    'logger': logger,
    'X_train': X_train,
    'y_train': y_train,
    'X_dev': X_dev,
    'y_dev': y_dev,
    'X_test': X_test,
    'y_test': y_test
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn, xgboost — 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
run_mlp_regression Not yet run
unverified (harvested, no recorded run)
ink-usc/predicting-big-bench/code/models/mlp.py
code served (permissive licence) · get_code("03fa5d9f7184220a")
How it ran

no invocation recorded.

run_mlp_regression_for_search Not yet run
unverified (harvested, no recorded run)
ink-usc/predicting-big-bench/code/models/mlp.py
code served (permissive licence) · get_code("0d0a8acf4803a219")
How it ran

no invocation recorded.

tune_mlp_regression Not yet run
unverified (harvested, no recorded run)
ink-usc/predicting-big-bench/code/models/mlp.py
code served (permissive licence) · get_code("72442426f6ca119f")
How it ran

no invocation recorded.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 15 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 15 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 investigate the predictability of large language model (LLM) capabilities: given records of past experiments using different model families, numbers of parameters, tasks, and numbers of in-context examples, can we accurately predict LLM performance on new experiment configurations? Answering this question has practical implications for LLM users (e.g., deciding which models to try), developers (e.g., prioritizing evaluation on representative tasks), and the research community (e.g., identifying hard-to-predict capabilities that warrant further investigation). We study the performance prediction problem on experiment records from BIG-bench. On a random train-test split, an MLP-based predictor achieves an $R^2$ score greater than 95%, indicating the presence of learnable patterns within the experiment records. We then formulate the problem of searching for "small-bench," an informative subset of BIG-bench tasks from which the performance on the full set can be maximally recovered. We find a subset as informative as BIG-bench Hard for evaluating new model families, while being $3\times$ smaller. Additionally, we find competitive subsets by clustering task representations learned by our MLP-based predictor and selecting tasks close to cluster centroids, highlighting the importance of task diversity in constructing "small-bench."

For agents

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

get_harvested_code_for_paper("2305.14947")
get_code_for_paper("2305.14947")
have("2305.14947")

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

curl https://syntology.ai/api/ran/2305.14947.json

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

[![Syntology run record](https://syntology.ai/api/ran/2305.14947.svg)](https://syntology.ai/paper/2305.14947)

Connect an agent — have() is free.