Skip to main content

A Manubot plugin to revise a manuscript using GPT-3

Project description

Manubot AI Editor

A tool for performing automatic, AI-assisted revisions of Manubot manuscripts. Check out the manuscript about this tool for more background information.

Supported Large Language Models (LLMs)

We currently support OpenAI models only, and are working to add support for other models. Our evaluations show that gpt-4-turbo is in general the best model for revising academic manuscripts. Therefore, this is the default option.

Using in a Manubot manuscript

Much of these instructions rely on the specific details of GitHub's website interface, which can change over time. See their official docs for more info on configuring GitHub Actions, managing secrets, and running workflows.

Setup

Start with a manuscript repo forked from Manubot rootstock, then follow these steps:

  1. In your forks's "▶️ Actions" tab, enable GitHub Actions.
  2. In your fork's "⚙️ Settings" tab, give GitHub Actions workflows read/write permissions and allow them to create pull requests.
  3. If you haven't already, make an OpenAI account and create an API key.
  4. In your fork's "⚙️ Settings" tab, make a new Actions repository secret with the name OPENAI_API_KEY and paste in your API key as the secret.

Configuring prompts

In order to revise your manuscript, prompts must be provided to the AI model. Manubot rootstock comes with several default, general-purpose prompts so that you can immediately use the AI editor without having to write and configure your own prompts.

But you can also define your own prompts, apply them to specific content, and control other behavior using YAML configuration files that you include with your manuscript. See docs/custom-prompts.md for more information.

Running the editor

  1. In your forks's "▶️ Actions" tab, go to the ai-revision workflow.
  2. Manually run the workflow. You should see several options you can specify, such as the branch to revise and the AI model to use. See these docs for an explanation of each option.
  3. Within a few minutes, the workflow should run, the editor should generate revisions, and a pull request should be created in your fork!

Caveats

In the current implementation, the AI editor can only process, independently, one paragraph at a time. This limits the contextual information the LLM receives and thus the specificity of what it can check and fix. For instance, the revision process does not use information in other places of the manuscript to revise the current paragraph. In addition, we provide section-specific prompts to revise text from different sections of the manuscript, such as the Abstract, Introduction, Results, etc. However, some paragraphs from the same section need different revision strategies. For example, in the Discussion section of a manuscript, the first paragraph should typically summarize the findings from the Results section, while the rest of the paragraphs should follow a different structure. The AI editor, however, can only judge each paragraph with the same section-specific prompt.

Finally, in addition to revising the paragraph using an LLM, the AI Editor will also perform some postprocessing of the revised text such as using one line per sentence to simplify diffs. This might not work as expected in some cases.

We plan to reduce or remove these limitations in the future.

Using from the command line

First, install Manubot in a Python environment, e.g.:

pip install --upgrade manubot[ai-rev]

You also need to export an environment variable with your OpenAI API key, e.g.:

export OPENAI_API_KEY=ABCD1234

You can also provide other environment variables that will change the behavior of the editor (such as revising certain files only). For example, to specify the temperature parameter of OpenAI models, you can set the variable export AI_EDITOR_TEMPERATURE=0.50. See the complete list of supported variables documents.

Then, from the root directory of your Manubot manuscript, run the following:

# ⚠ THIS WILL OVERWRITE YOUR LOCAL MANUSCRIPT
manubot ai-revision --content-directory content/ --config-directory ci/

The editor will revise each paragraph of your manuscript and write back the revised files in the same directory. Finally, (assuming you are tracking changes to your manuscript with git) you can review each change and either keep it (commit it) or reject it (revert it).

Using the OpenAI API can sometimes incur costs. If you're worried about this or otherwise want to test things out before hitting the real API, you can run a local "dry run" by with a "fake" model:

manubot ai-revision \
  --content-directory content/ \
  --config-directory ci/ \
  --model-type DummyManuscriptRevisionModel \
  --model-kwargs add_paragraph_marks=True

When it finishes, check out your manuscript files. This will allow you to detect whether the editor is identifying paragraphs correctly. If you find a problem, please report the issue.

Using the Python API

You can also use the functions of the editor directly from Python.

Since these functions are low-level and not tied to a particular manuscript, you don't have to install Manubot and can just install this package:

pip install -U manubot-ai-editor

Example usage:

import shutil
from pathlib import Path

from manubot_ai_editor.editor import ManuscriptEditor
from manubot_ai_editor.models import GPT3CompletionModel

# create a manuscript editor object.
me = ManuscriptEditor(
    # where your Markdown files (*.md) are
    content_dir="content",
    # where CI-related configuration, including the AI editor's, is stored.
    # optional, will fallback to defaults if omitted.
    config_dir="ci"
)

# create a model to revise the manuscript
model = GPT3CompletionModel(
    title=me.title,
    keywords=me.keywords,
)

# create a temporary directory to store the revised manuscript
output_folder = (Path("tmp") / "manubot-ai-editor-output").resolve()
shutil.rmtree(output_folder, ignore_errors=True)
output_folder.mkdir(parents=True, exist_ok=True)

# revise the manuscript
me.revise_manuscript(output_folder, model)

# the revised manuscript is now in the `output_folder`

# uncomment the following code if you want to OVERWRITE the original manuscript in the content folder with the revised manuscript
# for f in output_folder.glob("*"):
#     f.rename(me.content_dir / f.name)
#
# # remove output folder
# output_folder.rmdir()

The cli_process function in this file provides another example of how to use the API.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

manubot_ai_editor-0.5.2.tar.gz (35.7 kB view details)

Uploaded Source

Built Distribution

manubot_ai_editor-0.5.2-py3-none-any.whl (20.4 kB view details)

Uploaded Python 3

File details

Details for the file manubot_ai_editor-0.5.2.tar.gz.

File metadata

  • Download URL: manubot_ai_editor-0.5.2.tar.gz
  • Upload date:
  • Size: 35.7 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/5.1.1 CPython/3.10.14

File hashes

Hashes for manubot_ai_editor-0.5.2.tar.gz
Algorithm Hash digest
SHA256 fccd2bbdae618811444cb5ef2d07b95a2be9dc8772a3468bba1b7deaf20914e0
MD5 913ff7db58243f7779ff9ad1ceaaedd5
BLAKE2b-256 c73d78a42c826ec92c29eace66da421a36839716d861ccef03ba043078cc7de5

See more details on using hashes here.

File details

Details for the file manubot_ai_editor-0.5.2-py3-none-any.whl.

File metadata

File hashes

Hashes for manubot_ai_editor-0.5.2-py3-none-any.whl
Algorithm Hash digest
SHA256 b6ffa31931919e31ac3787856e6d37e91954808620ab26a02f2fc50cec7c6e09
MD5 3c00d1a757b1b73b8b6803a124105fe9
BLAKE2b-256 c8795e905561246f6d7532a3ec17063a2c458ff4a1f807aabe3cfbe82b2aec1e

See more details on using hashes here.

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page