LLM Usage Policy

Overview

This document describes the policy regarding the contribution of content generated by language models (a.k.a. LLMs, chatbots, or agents, including but not limited to Claude, ChatGPT, DeepSeek and others) to Tilde Friends. This policy was written with the aim of being concise and clearly understandable, so that there is a minimum of confusion as to how the policy applies to a given situation. First, we will state the policy, consider enforcement, and give a brief overview of the reasoning behind it.

The Policy

• Content generated by a language model may not be contributed to Tilde Friends.

"Language model" here means any program or service which outputs text in response to a generic natural language directive or completes substantial portions of an incomplete document or code unit. Note that dedicated translation and grammar assistance tools are not covered by this description, though in this case we ask that you disclose usage of these tools.

"Contributed" here means submitted as code or documentation as part of commits, commit messages, PR descriptions, issue descriptions, or comments on PRs or issues.

Enforcement

Verification of "LLM generated content" is up to moderation. Do not make public accusations of language-model contributed content; if you have concerns, report it to the development team and move on. Each contributor is responsible for ensuring the content they contribute follows this policy, in the same way that they must be sure that they do not submit content from, e.g. proprietary code from an employer, and in a similar way good faith should generally be assumed.

Reasoning

This policy was written to be clear and concise while rectifying issues present in other policies, notably overreach. There is no way in which this policy could dictate private usage of language models by contributors; it would be unenforceable if nothing else.

Language model generated content is disallowed because current implementations of such technology carry:

1. Severe and well-documented ethical concerns around their production and operation, especially with regards to their environmental impact, use of copyrighted material, abusive work practices, psychological effect on users, and bias. In order to maintain a welcoming community for all in accordance with the Tilde Friends [contribution guidelines](contributing), particularly those who have been directly harmed by the production or operation of these technologies, we disallow language model generated content.
2. Deep uncertainty with regard to copyright status of output. In order to protect the Tilde Friends project as a whole, we disallow language model generated content.

This document is enthusiastically adapted from Athena's proposed Rust Language Model Usage Policy.