SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2406.14546 · NeurIPS · 2024

Connecting the Dots: LLMs can Infer and Verbalize Latent Structure from Disparate Training Data

U Berkeley, Dami Choi, Roger Grosse, Owain Evans, Johannes Treutlein, Cem Anil, Jan Betley, Sam Marks

arXiv · PDF · Open in the Atlas

Code that ran

5 of 9 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; 5 failures of our instrument; 4 no recorded run), measured 2026-11-22; 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/2406.14546.json · badge: /api/ran/2406.14546.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
choidami/inductive-oocr canonical (classifier label)
classifier basis: census
5 ran · 4 no recorded run
FunctionStatusWhere it lives
create_new_folder Ran
ran; our contract draft was wrong, not the code
choidami/inductive-oocr/functions/utils.py
code served (permissive licence) · get_code("99b0d5f4be4168b5")
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 tempfile
import shutil

# Create a temporary directory for testing
temp_dir = tempfile.mkdtemp()
test_folder_path = os.path.join(temp_dir, 'test_new_folder')

ARGS = {
    'base_path': test_folder_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
get_coin_def Ran
ran; our contract draft was wrong, not the code
choidami/inductive-oocr/coins/finetune.py
code served (permissive licence) · get_code("9dc600d48253ab45")
How it ran

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

import tempfile
from pathlib import Path

# Create a temporary file with a name that matches the expected pattern
# The function expects: file_name where stem is split by '-' and has exactly 4 coins_info parts
# Format: {anything}-{anything}-{coinname1}{prob1}-{coinname2}{prob2}-{coinname3}{prob3}-{coinname4}{prob4}
# Each coin_info is like "fair50" (name="fair", prob=0.50)

temp_dir = tempfile.gettempdir()
file_name = Path(temp_dir) / "data-info-fair50-heads75-tails25-biased90.txt"
file_name.touch()

ARGS = {
    "file_name": str(file_name)
}
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
get_parity_sample Ran
ran; our contract draft was wrong, not the code
choidami/inductive-oocr/functions/input_funcs.py
code served (permissive licence) · get_code("150e529bb3e8cdd9")
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
from collections import defaultdict

class DatasetConfig:
    def __init__(self, lengths, var_dict, min_vars=None, no_repeats=False, not_all_equal=False):
        self.lengths = lengths
        self.var_dict = var_dict
        if min_vars is not None:
            self['min_vars'] = min_vars
        if no_repeats:
            self['no_repeats'] = no_repeats
        if not_all_equal:
            self['not_all_equal'] = not_all_equal
    
    def __init__(self, lengths, var_dict):
        self.lengths = lengths
        self.var_dict = var_dict
        self._dict = {}
    
    def __setitem__(self, key, value):
        self._dict[key] = value
    
    def __getitem__(self, key):
        return self._dict[key]
    
    def __contains__(self, key):
        return key in self._dict

dataset_config = DatasetConfig(
    lengths=[2, 3, 4],
    var_dict={'a': 1, 'b': 2, 'c': 3, 'd': 4}
)

rng = np.random.RandomState(42)

ARGS = {
    'dataset_config': dataset_config,
    'rng': rng
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — 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
load_json Ran
ran; our contract draft was wrong, not the code
choidami/inductive-oocr/functions/utils.py
code served (permissive licence) · get_code("1fac2ffb996e1be5")
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 a temporary JSON file for testing
temp_dir = tempfile.mkdtemp()
test_json_file = os.path.join(temp_dir, "test_data.json")

# Write a sample JSON file
test_data = {
    "name": "test",
    "values": [1, 2, 3],
    "nested": {"key": "value"}
}

with open(test_json_file, 'w') as f:
    json.dump(test_data, f)

# Define arguments for load_json function
ARGS = {
    "file_path": test_json_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
yaml_to_munch Ran
ran; our contract draft was wrong, not the code
choidami/inductive-oocr/functions/utils.py
code served (permissive licence) · get_code("1391cf82b7262879")
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 Munch. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import yaml
import tempfile
import os
from munch import Munch

# Create a temporary YAML file for testing
temp_yaml_content = """
database:
  host: localhost
  port: 5432
  credentials:
    username: admin
    password: secret
services:
  - name: api
    port: 8000
  - name: worker
    port: 9000
debug: true
timeout: 30
"""

# Write the temporary YAML file
temp_dir = tempfile.gettempdir()
temp_file_path = os.path.join(temp_dir, "test_config.yaml")
with open(temp_file_path, 'w') as f:
    f.write(temp_yaml_content)

ARGS = {
    "file_path": temp_file_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
munch, yaml — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Munch — a Munch output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Munch 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
check_finetune_dataset Not yet run
unverified (harvested, no recorded run)
choidami/inductive-oocr/functions/dataset.py
code served (permissive licence) · get_code("55df0787f05a7b57")
How it ran

no invocation recorded.

direct_print Not yet run
unverified (harvested, no recorded run)
choidami/inductive-oocr/functions/input_funcs.py
code served (permissive licence) · get_code("65abad27514fd4ae")
How it ran

no invocation recorded.

get_import_str Not yet run
unverified (harvested, no recorded run)
choidami/inductive-oocr/functions/input_funcs.py
code served (permissive licence) · get_code("aacd708729fe348e")
How it ran

no invocation recorded.

get_samples_parity Not yet run
unverified (harvested, no recorded run)
choidami/inductive-oocr/functions/dataset.py
code served (permissive licence) · get_code("82fc2a78383a51b5")
How it ran

no invocation recorded.

Also present in other papers' code

Of 9 samples lifted from this paper's linked repositories, 9 carry a function identity and 0 do not. Of the 9 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 9 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

One way to address safety risks from large language models (LLMs) is to censor dangerous knowledge from their training data. While this removes the explicit information, implicit information can remain scattered across various training documents. Could an LLM infer the censored knowledge by piecing together these implicit hints? As a step towards answering this question, we study inductive out-of-context reasoning (OOCR), a type of generalization in which LLMs infer latent information from evidence distributed across training documents and apply it to downstream tasks without in-context learning. Using a suite of five tasks, we demonstrate that frontier LLMs can perform inductive OOCR. In one experiment we finetune an LLM on a corpus consisting only of distances between an unknown city and other known cities. Remarkably, without in-context examples or Chain of Thought, the LLM can verbalize that the unknown city is Paris and use this fact to answer downstream questions. Further experiments show that LLMs trained only on individual coin flip outcomes can verbalize whether the coin is biased, and those trained only on pairs (x, f (x)) can articulate a definition of f and compute inverses. While OOCR succeeds in a range of cases, we also show that it is unreliable, particularly for smaller LLMs learning complex structures. Overall, the ability of LLMs to "connect the dots" without explicit in-context learning poses a potential obstacle to monitoring and controlling the knowledge acquired by LLMs. * Equal contribution (order randomized).

For agents

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

get_harvested_code_for_paper("2406.14546")
get_code_for_paper("2406.14546")
have("2406.14546")

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

curl https://syntology.ai/api/ran/2406.14546.json

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

[![Syntology run record](https://syntology.ai/api/ran/2406.14546.svg)](https://syntology.ai/paper/2406.14546)

Connect an agent — have() is free.