SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2311.01767 · 2023

PPTC Benchmark: Evaluating Large Language Models for PowerPoint Task Completion

arXiv · PDF · Open in the Atlas

Code that ran

13 of 17 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; 8 honoured the contract we drafted, 2 violated it and 0 have no contract verdict; 3 failures of our instrument; 4 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/2311.01767.json · badge: /api/ran/2311.01767.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
gydpku/pptc canonical (classifier label)
classifier basis: census
13 ran · 4 no recorded run
FunctionStatusWhere it lives
check_api_in_list Ran
ran, honoured the contract we drafted
gydpku/pptc/src/ppt_executor.py
code served (permissive licence) · get_code("3fb8d14cb807e9b5")
How it ran

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

import torch
import numpy as np

# Construct arguments for check_api_in_list(line, api_list)
# The function checks if any api string from api_list is contained in line string

line = "import torch; from torch import nn; import numpy as np"
api_list = ["torch", "numpy", "tensorflow", "pandas"]

ARGS = {
    "line": line,
    "api_list": api_list
}
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 int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
check_left Ran
ran, honoured the contract we drafted
gydpku/pptc/src/pptx_check.py
code served (permissive licence) · get_code("36bec59464b6e01c")
How it ran

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

import torch

class RectangleA:
    def __init__(self, left, width):
        self.left = left
        self.width = width

class RectangleB:
    def __init__(self, left):
        self.left = left

A = RectangleA(left=5, width=3)
B = RectangleB(left=10)

ARGS = {
    "A": A,
    "B": B
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
check_right Ran
ran, honoured the contract we drafted
gydpku/pptc/src/pptx_check.py
code served (permissive licence) · get_code("63252f499f1cb498")
How it ran

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

class MockObject:
    def __init__(self, left, width):
        self.left = left
        self.width = width

ARGS = {
    "A": MockObject(left=10, width=5),
    "B": MockObject(left=3, width=2),
}
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 int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
check_top Ran
ran, honoured the contract we drafted
gydpku/pptc/src/pptx_check.py
code served (permissive licence) · get_code("4450f6b58f2766de")
How it ran

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

class MockObject:
    def __init__(self, top, height):
        self.top = top
        self.height = height

A = MockObject(top=5, height=10)
B = MockObject(top=20, height=8)

ARGS = {
    "A": A,
    "B": B
}
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 int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
check_trigger Ran
ran, honoured the contract we drafted
gydpku/pptc/remove_duplicate.py
code served (permissive licence) · get_code("aacdff4f5284d7b6")
How it ran

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

import torch
import numpy as np

torch.manual_seed(0)

# Analyze the function: it takes 'pred' and 'label'
# - 'pred' is joined with '\n', suggesting it's a list/iterable of strings
# - 'label' is checked with 'in' operator, suggesting it's a string
# - The function checks if certain strings appear in pred (after joining) and in label

# Create valid arguments
pred = ['choose_table_cell', 'insert_table_row', 'delete_row']
label = 'insert_table_row and choose_table_cell are options'

ARGS = {
    'pred': pred,
    'label': label
}
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 int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_API_name Ran
ran; our contract draft was wrong, not the code
gydpku/pptc/src/api_doc.py
code served (permissive licence) · get_code("750d50b5ca3e4092")
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.

class API:
    def __init__(self, name, parameters):
        self.name = name
        self.parameters = parameters

api1 = API("get_user", "_id_token")
api2 = API("post_data", "_request_body")
api3 = API("delete_item", "_item_id")

apis = [api1, api2, api3]

ARGS = {
    "apis": apis
}
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
get_fill_color Ran
ran, honoured the contract we drafted
gydpku/pptc/src/ppt_executor.py
code served (permissive licence) · get_code("0368656863d589da")
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 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.

class Color:
    def __init__(self, rgb_value):
        self.rgb = rgb_value

class FillWithRGB:
    def __init__(self):
        self.type = 1
        self.fore_color = Color((255, 128, 64))

class FillWithoutRGB:
    def __init__(self):
        self.type = 1
        self.fore_color = "some_string"

class Shape:
    def __init__(self, fill):
        self.fill = fill

shape_with_rgb = Shape(FillWithRGB())

ARGS = {
    "shape": shape_with_rgb
}
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 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
get_fill_color Ran
ran, honoured the contract we drafted
gydpku/pptc/src/ppt_reader.py
code served (permissive licence) · get_code("f57954039746e683")
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 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.

class Color:
    def __init__(self, rgb_value):
        self.rgb = rgb_value

class Fill:
    def __init__(self, fill_type, fore_color=None):
        self.type = fill_type
        self.fore_color = fore_color

class Shape:
    def __init__(self, fill):
        self.fill = fill

# Case 1: Valid fill with type 1 and color with rgb attribute
color_with_rgb = Color(rgb_value=(255, 128, 64))
fill_valid = Fill(fill_type=1, fore_color=color_with_rgb)
shape_valid = Shape(fill=fill_valid)

ARGS = {
    "shape": shape_valid
}
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 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
get_topk Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
gydpku/pptc/src/api_selection.py
code served (permissive licence) · get_code("7ec3a2a9bc36cd82")
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 torch
import numpy as np

torch.manual_seed(0)

# Analyzing the function:
# - scores: used with len() and indexed by integer keys, so a list or array-like
# - k: integer parameter with default 10

# Create a small scores list with distinct values for easy verification
scores = [0.9, 0.3, 0.7, 0.1, 0.5, 0.2, 0.8, 0.4, 0.6, 0.95, 0.15, 0.25]

ARGS = {
    "scores": scores,
    "k": 5
}
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, 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
hasshape Ran
ran, violated the contract we drafted
gydpku/pptc/src/ppt_reader.py
code served (permissive licence) · get_code("8f6c62d7a26a1a28")
How it ran

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

import torch
import numpy as np

shape_list = ['PLACEHOLDER', 'PICTURE', 'CHART', 'TABLE', 'TEXT_BOX', 'AUTO_SHAPE']

# Test case 1: shape_str contains one of the shapes in the list
shape_str_1 = "This is a PICTURE element"

# Test case 2: shape_str does not contain any shapes from the list
shape_str_2 = "This is a generic element"

ARGS = {
    'shape_str': shape_str_1,
    'shape_list': shape_list
}
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 bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_data Ran
ran; our contract draft was wrong, not the code
gydpku/pptc/src/dataset.py
code served (permissive licence) · get_code("3ebbdae2f7a95dd1")
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 os
import tempfile
import shutil

# Create a temporary directory structure with the required files
temp_dir = tempfile.mkdtemp()
dataset_dir = os.path.join(temp_dir, 'test_dataset')
os.makedirs(dataset_dir, exist_ok=True)

# Create instructions.txt
instructions_content = """instruction 1
instruction 2

instruction 3
instruction 4

instruction 5"""

with open(os.path.join(dataset_dir, 'instructions.txt'), 'w') as f:
    f.write(instructions_content)

# Create robust1.txt (for robust mode)
with open(os.path.join(dataset_dir, 'robust1.txt'), 'w') as f:
    f.write(instructions_content)

# Create noisy.txt (for noisy mode)
with open(os.path.join(dataset_dir, 'noisy.txt'), 'w') as f:
    f.write(instructions_content)

# Create api_labels.txt
labels_content = """api1;api2;api3
api4;api5

api6
api7;api8

api9;api10;api11;api12"""

with open(os.path.join(dataset_dir, 'api_labels.txt'), 'w') as f:
    f.write(labels_content)

# Create a simple args class
class Args:
    def __init__(self, robust=False, noisy=False, robust_num=1):
        self.robust = robust
        self.noisy = noisy
        self.robust_num = robust_num

ARGS = {
    'path': temp_dir,
    'dataset': 'test_dataset',
    'args': Args(robust=False, noisy=False, robust_num=1)
}
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 — 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
parse_answer Ran
ran, honoured the contract we drafted
gydpku/pptc/src/content_selection.py
code served (permissive licence) · get_code("acccc24d18f3e8af")
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 torch
import numpy as np

def parse_answer(answer):
    answer = answer.strip()
    answer = answer.split(',')
    answer = [int(i.split('=')[1]) for i in answer]
    return answer

# Analyzing the function:
# - Takes a single parameter 'answer' which is a string
# - Strips whitespace
# - Splits by comma
# - For each element, splits by '=' and takes the second part (index 1)
# - Converts to int
# So the input should be a string like "key=1,key=2,key=3"

answer_string = "x=42,y=17,z=3"

ARGS = {
    "answer": answer_string
}
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, 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
random_permutation Ran
ran, violated the contract we drafted
gydpku/pptc/src/api_doc.py
code served (permissive licence) · get_code("90028c50a5b3c5e6")
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

# The function takes a list and returns a shuffled copy
# It uses: lst.copy() and random.shuffle()
# So lst must be a list

ARGS = {
    "lst": [1, 2, 3, 4, 5]
}
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
get_all_APIs Not yet run
unverified (harvested, no recorded run)
gydpku/pptc/src/api_doc.py
code served (permissive licence) · get_code("f12cae9340a33209")
How it ran

no invocation recorded.

get_content Not yet run
unverified (harvested, no recorded run)
gydpku/pptc/src/ppt_reader.py
code served (permissive licence) · get_code("ecd0f7075ab18e97")
How it ran

no invocation recorded.

select_information_type Not yet run
unverified (harvested, no recorded run)
gydpku/pptc/src/content_selection.py
code served (permissive licence) · get_code("830ca0bb57f97a8c")
How it ran

no invocation recorded.

select_shape_type Not yet run
unverified (harvested, no recorded run)
gydpku/pptc/src/content_selection.py
code served (permissive licence) · get_code("0e252063562ea1b8")
How it ran

no invocation recorded.

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 16 distinct functions with an identity, 16 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 16 later) and 0 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
check_api_in_list 1 other paper
1 later
2024.findings-acl.514 (2024 (year only))
PPTC Benchmark: Evaluating Large Language Models for PowerPoint Task Completion
check_left 1 other paper
1 later
2024.findings-acl.514 (2024 (year only))
PPTC Benchmark: Evaluating Large Language Models for PowerPoint Task Completion
check_right 1 other paper
1 later
2024.findings-acl.514 (2024 (year only))
PPTC Benchmark: Evaluating Large Language Models for PowerPoint Task Completion
check_top 1 other paper
1 later
2024.findings-acl.514 (2024 (year only))
PPTC Benchmark: Evaluating Large Language Models for PowerPoint Task Completion
check_trigger 1 other paper
1 later
2024.findings-acl.514 (2024 (year only))
PPTC Benchmark: Evaluating Large Language Models for PowerPoint Task Completion
get_API_name 1 other paper
1 later
2024.findings-acl.514 (2024 (year only))
PPTC Benchmark: Evaluating Large Language Models for PowerPoint Task Completion
get_all_APIs 1 other paper
1 later
2024.findings-acl.514 (2024 (year only))
PPTC Benchmark: Evaluating Large Language Models for PowerPoint Task Completion
get_content 1 other paper
1 later
2024.findings-acl.514 (2024 (year only))
PPTC Benchmark: Evaluating Large Language Models for PowerPoint Task Completion
get_fill_color 1 other paper
1 later
2024.findings-acl.514 (2024 (year only))
PPTC Benchmark: Evaluating Large Language Models for PowerPoint Task Completion
get_topk 1 other paper
1 later
2024.findings-acl.514 (2024 (year only))
PPTC Benchmark: Evaluating Large Language Models for PowerPoint Task Completion
hasshape 1 other paper
1 later
2024.findings-acl.514 (2024 (year only))
PPTC Benchmark: Evaluating Large Language Models for PowerPoint Task Completion
load_data 1 other paper
1 later
2024.findings-acl.514 (2024 (year only))
PPTC Benchmark: Evaluating Large Language Models for PowerPoint Task Completion
parse_answer 1 other paper
1 later
2024.findings-acl.514 (2024 (year only))
PPTC Benchmark: Evaluating Large Language Models for PowerPoint Task Completion
random_permutation 1 other paper
1 later
2024.findings-acl.514 (2024 (year only))
PPTC Benchmark: Evaluating Large Language Models for PowerPoint Task Completion
select_information_type 1 other paper
1 later
2024.findings-acl.514 (2024 (year only))
PPTC Benchmark: Evaluating Large Language Models for PowerPoint Task Completion
select_shape_type 1 other paper
1 later
2024.findings-acl.514 (2024 (year only))
PPTC Benchmark: Evaluating Large Language Models for PowerPoint Task Completion

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

Recent evaluations of Large Language Models (LLMs) have centered around testing their zero-shot/few-shot capabilities for basic natural language tasks and their ability to translate instructions into tool APIs. However, the evaluation of LLMs utilizing complex tools to finish multi-turn, multi-modal instructions in a complex multi-modal environment has not been investigated. To address this gap, we introduce the PowerPoint Task Completion (PPTC) benchmark to assess LLMs' ability to create and edit PPT files based on user instructions. It contains 279 multi-turn sessions covering diverse topics and hundreds of instructions involving multi-modal operations. We also propose the PPTX-Match Evaluation System that evaluates if LLMs finish the instruction based on the prediction file rather than the label API sequence, thus it supports various LLM-generated API sequences. We measure 3 closed LLMs and 6 open-source LLMs. The results show that GPT-4 outperforms other LLMs with 75.1\% accuracy in single-turn dialogue testing but faces challenges in completing entire sessions, achieving just 6\% session accuracy. We find three main error causes in our benchmark: error accumulation in the multi-turn session, long PPT template processing, and multi-modality perception. These pose great challenges for future LLM and agent systems. We release the data, code, and evaluation system of PPTC at \url{https://github.com/gydpku/PPTC}.

For agents

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

get_harvested_code_for_paper("2311.01767")
get_code_for_paper("2311.01767")
have("2311.01767")

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

curl https://syntology.ai/api/ran/2311.01767.json

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

[![Syntology run record](https://syntology.ai/api/ran/2311.01767.svg)](https://syntology.ai/paper/2311.01767)

Connect an agent — have() is free.