SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2402.03563 · 2024

Distinguishing the Knowable from the Unknowable with Language Models

arXiv · PDF · Open in the Atlas

Code that ran

10 of 16 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, 0 violated it and 4 have no contract verdict; 5 failures of our instrument; 6 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/2402.03563.json · badge: /api/ran/2402.03563.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
gahdritz/llm_uncertainty canonical (classifier label)
classifier basis: census
6 ran · 5 no recorded run
mims-harvard/spectra community (classifier label)
classifier basis not recorded on the link
4 ran · 1 no recorded run
FunctionStatusWhere it lives
align_overlap Ran
ran on a synthesized input
mims-harvard/spectra/SPECTRA_paper/SPECTRA_SD/PEER_localization/calculate_sequence_similarity.py
code served (permissive licence) · get_code("c44aa791d163dc8e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function takes 'input' which should be a tuple/list of two iterables
# that can be zipped together and compared element-by-element
# Based on the usage: for orig, n in zip(i, j)
# i and j should be sequences of comparable items

i = [1, 2, 3, 4]
j = [1, 2, 5, 4]

ARGS = {
    'input': (i, j)
}
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 float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
build_rope_cache Ran
ran, honoured the contract we drafted
gahdritz/llm_uncertainty/lit-llama/model.py
code served (permissive licence) · get_code("fd462b61b82d4e41")
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 Tensor, ndim 3, dtype family float, shape (8, 2, 2) 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)

ARGS = {
    "seq_len": 8,
    "n_elem": 4,
    "dtype": torch.float32,
    "device": torch.device("cpu"),
    "base": 10000,
}
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 Tensor, ndim 3, dtype family float, shape (8, 2, 2) — shape (8, 2, 2) 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
compare_to_all Ran
ran on a synthesized input
mims-harvard/spectra/SPECTRA_paper/SPECTRA_SD/PEER_localization/calculate_sequence_similarity.py
code served (permissive licence) · get_code("362c6a2a81602b1c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Create a simple aligner class that has an align method
class SimpleAligner:
    def align(self, seq1, seq2):
        # Simple alignment that returns both sequences as-is
        # The align method should return a list where first element is [aligned_seq1, aligned_seq2]
        return [[seq1, seq2]]

# Define sequences as lists for comparison
i = [1, 2, 3, 4]
compare = [
    [1, 2, 3, 4],      # Same length, exact match
    [1, 2, 5, 4],      # Same length, one difference
    [1, 2, 3],         # Different length
]
aligner = SimpleAligner()

ARGS = {
    'i': i,
    'compare': compare,
    'aligner': aligner
}
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 — 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
compute_entropy Ran
ran; our contract draft was wrong, not the code
gahdritz/llm_uncertainty/repetition.py
code served (permissive licence) · get_code("625d925ef1efd4f4")
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 Tensor, ndim 1, dtype family float, shape (2,) 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 compute_entropy takes logits as input
# It applies softmax and log_softmax to the logits tensor
# Then computes entropy by summing along the last dimension
# We need to provide a tensor of logits with shape (batch_size, num_classes)

logits = torch.randn(2, 4)

ARGS = {
    "logits": logits
}
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 Tensor, ndim 1, dtype family float, shape (2,) — shape (2,) 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
extract_data Ran
ran on a synthesized input
mims-harvard/spectra/SPECTRA_paper/SPECTRA_SD/TAPE_Homology/calculate_similarity.py
code served (permissive licence) · get_code("019a737a08eeaf7e")
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 json
import tempfile
import os

# Create temporary JSON files with the expected structure
temp_dir = tempfile.mkdtemp()

# Create first test file
file1_path = os.path.join(temp_dir, "data1.json")
with open(file1_path, 'w') as f:
    json.dump([
        {'id': 'item_1', 'primary': 'data_1'},
        {'id': 'item_2', 'primary': 'data_2'}
    ], f)

# Create second test file
file2_path = os.path.join(temp_dir, "data2.json")
with open(file2_path, 'w') as f:
    json.dump([
        {'id': 'item_3', 'primary': 'data_3'}
    ], f)

ARGS = {
    'filenames': [file1_path, file2_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 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
jsd Ran
ran; our contract draft was wrong, not the code
gahdritz/llm_uncertainty/lit-llama/utils.py
code served (permissive licence) · get_code("6555dda0713b63e2")
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 Tensor, ndim 1, dtype family float, shape (2,) 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 torch.nn.functional as F

torch.manual_seed(0)

# The function takes two logit tensors and computes Jensen-Shannon divergence
# Logits should be 2D tensors: (batch_size, num_classes)
batch_size = 2
num_classes = 4

net_1_logits = torch.randn(batch_size, num_classes)
net_2_logits = torch.randn(batch_size, num_classes)

ARGS = {
    "net_1_logits": net_1_logits,
    "net_2_logits": net_2_logits,
}
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 Tensor, ndim 1, dtype family float, shape (2,) — shape (2,) 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
llama_model_lookup Ran
ran; our contract draft was wrong, not the code
gahdritz/llm_uncertainty/lit-llama/utils.py
code served (permissive licence) · get_code("33a04a05cc4d445e")
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 expects a checkpoint dict with 'lm_head.weight' key
# The weight tensor should have shape [vocab_size, embedding_size]
# where embedding_size is one of the keys in llama_model_sizes (4096, 5120, 6656, 8192)

checkpoint = {
    'lm_head.weight': torch.randn(32000, 4096)  # vocab_size=32000, embedding_size=4096 (maps to '7B')
}

ARGS = {
    'checkpoint': checkpoint
}
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
pythia_forward Ran
ran; our contract draft was wrong, not the code
gahdritz/llm_uncertainty/precompute_logits.py
code served (permissive licence) · get_code("b3eb11b6f841e055")
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 function. 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 MockGPTNeoX(nn.Module):
    def __init__(self, hidden_size=4, num_hidden_layers=3, vocab_size=8):
        super().__init__()
        self.hidden_size = hidden_size
        self.num_hidden_layers = num_hidden_layers
        self.vocab_size = vocab_size
        self.embed_tokens = nn.Embedding(vocab_size, hidden_size)
        self.layers = nn.ModuleList([nn.Linear(hidden_size, hidden_size) for _ in range(num_hidden_layers)])
    
    def __call__(self, input_ids, attention_mask, output_hidden_states=False, return_dict=False):
        hidden_states = self.embed_tokens(input_ids)
        all_hidden_states = [hidden_states]
        
        for layer in self.layers:
            hidden_states = layer(hidden_states)
            all_hidden_states.append(hidden_states)
        
        class Output:
            def __init__(self, hidden_states_list):
                self.hidden_states = hidden_states_list
        
        return Output(all_hidden_states)

class MockPythiaModel(nn.Module):
    def __init__(self, hidden_size=4, vocab_size=8):
        super().__init__()
        self.gpt_neox = MockGPTNeoX(hidden_size=hidden_size, num_hidden_layers=3, vocab_size=vocab_size)
        self.embed_out = nn.Linear(hidden_size, vocab_size)
    
    def __call__(self, input_ids, attention_mask, output_hidden_states=False, return_dict=False):
        return self.gpt_neox(input_ids, attention_mask, output_hidden_states, return_dict)

model = MockPythiaModel(hidden_size=4, vocab_size=8)

ARGS = {
    "model": model,
    "embeddings": False,
    "return_after_layer_n": -1
}
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 function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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
read_shard Ran
ran; our contract draft was wrong, not the code
gahdritz/llm_uncertainty/train_head_utils.py
code served (permissive licence) · get_code("aeff514913b60d52")
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 pickle
import tempfile
import os

# Create a temporary pickle file for testing
temp_dir = tempfile.mkdtemp()
test_data = {"key": "value", "number": 42, "list": [1, 2, 3]}

pickle_file = os.path.join(temp_dir, "test_shard.pickle")
with open(pickle_file, 'wb') as fp:
    pickle.dump(test_data, fp)

ARGS = {
    "shard_path": pickle_file
}
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 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
run_independent_set Ran
ran on a synthesized input
mims-harvard/spectra/SPECTRA_paper/SPECTRA_MSD/TB_Covid_GFP/utils/independent_set_algorithm.py
code served (permissive licence) · get_code("3e6cd88740fb23a0")
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 networkx as nx
from random import sample
import random

# Create a small NetworkX graph with the required node attributes
G = nx.Graph()

# Add nodes with required attributes
nodes_data = [
    (0, {'number_isolates': 5, 'mutational_score': 0.3}),
    (1, {'number_isolates': 8, 'mutational_score': 0.6}),
    (2, {'number_isolates': 3, 'mutational_score': 0.2}),
    (3, {'number_isolates': 10, 'mutational_score': 0.9}),
    (4, {'number_isolates': 2, 'mutational_score': 0.1}),
]

for node, attrs in nodes_data:
    G.add_node(node, **attrs)

# Add edges to create some graph structure
G.add_edges_from([
    (0, 1),
    (1, 2),
    (2, 3),
    (3, 4),
    (0, 4),
])

ARGS = {
    'neighbor_lambda': 0.8,
    'input_G': G,
    'tau_number': 0.5,
    'tau_mut': 0.3,
    'tau_degree': 0.2,
    'seed': 42,
    'debug': False,
    'random_start': True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
networkx — 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
apply_rope Not yet run
unverified (harvested, no recorded run)
gahdritz/llm_uncertainty/lit-llama/model.py
code served (permissive licence) · get_code("16b3ba074f0d8074")
How it ran

no invocation recorded.

istarmap Not yet run
unverified (harvested, no recorded run)
mims-harvard/spectra/SPECTRA_paper/SPECTRA_SD/PEER_localization/calculate_sequence_similarity.py
code served (permissive licence) · get_code("79344d1f8e7386a7")
How it ran

no invocation recorded.

lazy_load Not yet run
unverified (harvested, no recorded run)
gahdritz/llm_uncertainty/lit-llama/utils.py
code served (permissive licence) · get_code("92f3dde3779ce5cb")
How it ran

no invocation recorded.

mutual_information Not yet run
unverified (harvested, no recorded run)
gahdritz/llm_uncertainty/generate_from_repetition.py
code served (permissive licence) · get_code("57c12d6227a85cb1")
How it ran

no invocation recorded.

repetition_experiment Not yet run
unverified (harvested, no recorded run)
gahdritz/llm_uncertainty/repetition.py
code served (permissive licence) · get_code("24a0bd59545e8931")
How it ran

no invocation recorded.

simple_classification Not yet run
unverified (harvested, no recorded run)
gahdritz/llm_uncertainty/generate_from_repetition.py
code served (permissive licence) · get_code("de50ccf9e860e5bc")
How it ran

no invocation recorded.

Also present in other papers' code

Of 16 samples lifted from this paper's linked repositories, 16 carry a function identity and 0 do not. Of the 16 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 16 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 study the feasibility of identifying epistemic uncertainty (reflecting a lack of knowledge), as opposed to aleatoric uncertainty (reflecting entropy in the underlying distribution), in the outputs of large language models (LLMs) over free-form text. In the absence of ground-truth probabilities, we explore a setting where, in order to (approximately) disentangle a given LLM's uncertainty, a significantly larger model stands in as a proxy for the ground truth. We show that small linear probes trained on the embeddings of frozen, pretrained models accurately predict when larger models will be more confident at the token level and that probes trained on one text domain generalize to others. Going further, we propose a fully unsupervised method that achieves non-trivial accuracy on the same task. Taken together, we interpret these results as evidence that LLMs naturally contain internal representations of different types of uncertainty that could potentially be leveraged to devise more informative indicators of model confidence in diverse practical settings.

For agents

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

get_harvested_code_for_paper("2402.03563")
get_code_for_paper("2402.03563")
have("2402.03563")

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

curl https://syntology.ai/api/ran/2402.03563.json

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

[![Syntology run record](https://syntology.ai/api/ran/2402.03563.svg)](https://syntology.ai/paper/2402.03563)

Connect an agent — have() is free.