Preface

It doesn’t take me to tell you that LLMs are everywhere. We’re at a point where it feels inescapable: projects, discourse, and the technology industry as a whole. Expectedly, the Fyra community (including staff) have been long debating the implications, ethics, and usage of this technology as it pertains to our projects, values, and mission. I don’t intend to speak for the team or community as a whole, but rather articulate my observations and take action towards protecting our community, projects, and most importantly, our mission.

Observations

The rate of development of LLMs and their tooling has been dizzying, for better or worse. As such, so did the usage and hype. In what felt like a few moments, every techbro and their dog couldn’t shut up about chatbots, then completion assistants, then “agents”. As such, it has made it hard for me to form a cohesive world model of the capabilities, harms, and social implications of this technology. It feels as I’m in a never ending sprint against its progress. As someone in a position of policy, this has made it difficult for me to make an informed decision on any action we should take. Fortunately, we were relatively insulated from the excesses of the LLM craze, which gave me the option to defer until there was more clarity. This has changed.

While we have occasionally received the weird looking (probably slop) pull request, I’ve personally noticed an increase in LLMs being used for issue/PR descriptions and comments from new contributors, specifically in Terra. This has created a burden on our maintainers, and more importantly, erodes the mutual trust required for effective communication. If someone can’t be sure they’re talking to a real person on the other end, there are more opportunities for suspicion and distrust, even when the other party isn’t an LLM. This type of hyper-vigilance destroys communities, guts the feeling of humanity towards those we aren’t familiar with, and converts a culture of trust to a culture of defensive hostility. Given these harms, I believe it is in the best interest of the community to take action.

Admittedly, our community has a range of views on the utility and ethics of using LLMs, albeit lacking the LLM fanaticism you might see if you took two footsteps in San Francisco. I don’t intend to litigate these views. Where it doesn’t lead to personal attacks, this discussion should continue and is helpful for informing future policy.

Action

Usage of full LLM output for human to human communication, such as issue descriptions, replies, pull requests, and our public chats will lead to an immediate expulsion from our community and all projects. In other words, if you copyslop your messages in this community, expect repercussions. A conditional exception is granted for translation to English. This is not intended to affect any other usage of LLMs. For example, pasting your own words into an LLM to check for grammatical issues isn’t covered by this policy. Details on how offending messages are determined are in the next section. This policy will soon be formalized into our code of conducts.

It is so ordered.

Enforceability

I acknowledge that a policy like this is quite unenforceable, but that isn’t as much as a concern as it may seem. The goal with a policy like this is to:

• Act as a first-line deterrent.
• Address the most egregious cases of copyslopping.
• Provide cover to maintainers and moderators to take action where needed.
• Make a public statement against this harmful behavior.

As for determining what content is considered offending, determinations shouldn’t be made on purely stylistic factors, as this could lead to the same hyper-vigilance we are intending to protect against. Rather, we’ll consider a person in violation if they are consistently unable to explain or defend their reasoning.

Next steps

Compared to all the options on the table for combating the negative harms of LLMs, this is quite a conservative and targeted action. This is by design, as the progression of this technology and its adoption has not slowed and will likely continue to be unpredictable in the short-term, and thus hard to analyze. I also want to avoid unintended consequences that could cause additional harm, which is why translation is explicitly exempted. However, we will take swift action and continue to update policy as harms are apparent. I also hope this can spark more discussion on how to move from here, in order to form a more comprehensive policy and position towards the usage of LLMs.

~Lea