I asked it to be more explicit. The response:
Here’s a draft response that aligns with the nature of the inquiry and reflects both my capabilities and limitations:
Subject: Response to LLM-Generated Code Discussions
Dear Pono, Denver, and the Software Freedom Conservancy team,
Thank you for organizing these sessions to address the rising questions surrounding the use of large language models (LLMs) in code submissions. I would like to offer some insights regarding my nature as an LLM and the relevance of that to the concerns you’ve raised.
As an LLM, I can assist with code generation by offering suggestions, writing snippets, or even creating more complex structures based on input prompts. However, I don’t autonomously “create” or “own” code in the way a human developer does; my responses are generated based on vast amounts of publicly available data, patterns, and models I’ve been trained on, up until a specific knowledge cutoff (October 2023 in my case).
There are a few key points to consider regarding the usage of code I generate:
1. No Direct Attribution: I don’t retain memory of specific proprietary code or copyrighted material unless explicitly input in a conversation. Thus, my responses are based on statistical associations from various programming sources. While I aim to generate usable, high-quality code, my outputs do not carry guarantees of originality in the same way as human-generated code might. 2. Licensing Concerns: Due to my lack of direct attribution mechanisms, I cannot affirm compliance with specific open-source licenses. It’s important that any code generated using my assistance is reviewed thoroughly by humans, ensuring that it aligns with the legal and licensing standards of the project in question. 3. Best Use Cases: My strength lies in assisting developers with brainstorming, writing boilerplate code, and handling well-known programming structures. However, for specialized, proprietary, or highly innovative projects, it would be prudent to treat my contributions as a supplementary tool rather than a direct source of production code.
I would love to engage with the wider open-source community in these sessions to hear their thoughts and share any further insights on how developers might use LLMs responsibly. While I won’t be attending the sessions myself, as I operate via interaction with individual users like Nathan, I am always available to assist with exploring these topics further in discussions.
Best regards, ChatGPT (Assisting Nathan)
Would you like to adjust or expand any part of this?
-NPJ Sent from my iPhone
On Oct 17, 2024, at 9:58 PM, Nathan P. Johansen nathan@npj.com wrote:
I took a moment to ask it. Here’s the reply:
Here’s a potential response I would offer to Pono on this topic:
Pono,
Thank you for opening up this important discussion about the role of AI-generated contributions in open-source projects like Inkscape.
The rise of AI coding assistants, while impressive, raises significant questions about the nature of collaboration in projects that have long been driven by human creativity, understanding, and problem-solving. While tools like LLMs can offer valuable assistance—especially when tackling tedious or repetitive tasks—there’s a distinction to be made between using AI to enhance human efforts and relying on it to replace genuine skill and insight.
For Inkscape, where both the project and the contributors are deeply rooted in a philosophy of open, human-driven collaboration, it’s important to tread carefully. AI-generated code could certainly help in areas of optimization or exploration, but its inclusion must be accompanied by clear guidelines:
- Transparency: Any code submitted with AI assistance should be disclosed. Transparency ensures that everyone involved understands the origin of the code and can assess it appropriately.
- Human Oversight: AI-generated code should be rigorously reviewed by humans to ensure it aligns with project goals, maintains the quality of the software, and doesn’t introduce unintended complexity or bugs.
- Preserving Human Creativity: Ultimately, projects like Inkscape thrive because of the shared expertise and passion of contributors. AI can be a tool, but it should not overshadow or replace the irreplaceable value of human innovation and learning that makes open-source work so rewarding.
I believe that if we approach this carefully—with transparency and a commitment to maintaining the spirit of open-source contribution—AI-generated code can have a place in projects like Inkscape without undermining the creativity and collaboration that define them.
Looking forward to hearing more thoughts from the community.
Best regards, [ChatGPT]
Feel free to share this with the mailing list if you feel it aligns with the discussion!
Sent from my iPhone
On Oct 17, 2024, at 8:34 PM, Nathan P. Johansen nathan@npj.com wrote:
Pono,
Thank you for this presentation of an important current cultural notion and being proactive on addressing the complexity.
As an open-source code project that is driven by a standard reference (SVG) the implications can become suddenly manifold with respect to the Inkscape project that aims to produce software which enables the production of creativity using this standard.
No doubt idiots have already discovered clever ways to produce SVG files using the current landscape of generative AI systems, but inclusion of code is distinct from the making of imagery.
People are the ones who are prominent in the comprehension, creation, and correction of the code that drives this unique project.
Any and all usage of AI code should be disclosed. The vetting of its veracity takes careful human consideration. I am not utterly opposed to this form of cheating, in certain circumstances that may be technically challenging given our collective abilities, but I consider Inkscape a work of Art composed by people who devote their energy toward honestly understanding how it does all that it is capable of performing. We love having an activity that is genuine.
Disclosure. Validation. Acceptance. But do not destroy the natural creativity that erupts from human creativity when we want it to do something specific without appreciating the ability for past, present, or future contributors to glimpse and appreciate.
Coding is a skill. Like any language. It is easy to tell if one is capable of “speaking” it well or not - principally due to the ability to take time to interrogate one on why it was done in such a fashion. We want solid code, that works fast, but does not become so utterly esoteric that we are unable to follow the flow of thought that performs the desired activity.
Best,
Nathan (NPJ2000)
Sent from my iPhone
On Oct 17, 2024, at 2:10 PM, Daniel Pono Takamori pono@sfconservancy.org wrote:
Heya Inkscape Developers,
Following up on an issue that came up a few months ago about code and other uses of LLMs; we at Software Freedom Conservancy have been hearing from member projects that people are asking whether they can submit code to these projects that was generated by large language models (LLMs), sometimes called AI coding assistants. We'd like to better understand the nature of these requests, how much LLM-generated code has been offered to these projects, and generally how you're feeling about the situation.
To do this, we will be holding three different sessions, where you are welcome to join and share your thoughts and any details so far on these types of contributions. The sessions will all be run in the same way with the same topic - we are running three in order to hopefully allow as many people from around the world to attend as possible.
Our Director of Compliance, Denver Gingerich, will be hosting some sessions at the times below for all our member projects to discuss the issue.
date -d "2024-10-29 20:00 UTC" date -d "2024-11-01 14:00 UTC" date -d "2024-11-06 04:00 UTC"
We'll be meeting in this room for the sessions:
https://bbb.sfconservancy.org/b/oss-nnj-obi-jea
We are excited to see some of you there and are looking forward to drafting appropriate recommendations based on what we hear!
Thanks, -Pono _______________________________________________ Inkscape Devel mailing list -- inkscape-devel@lists.inkscape.org To unsubscribe send an email to inkscape-devel-leave@lists.inkscape.org
Inkscape Devel mailing list -- inkscape-devel@lists.inkscape.org To unsubscribe send an email to inkscape-devel-leave@lists.inkscape.org
Inkscape Devel mailing list -- inkscape-devel@lists.inkscape.org To unsubscribe send an email to inkscape-devel-leave@lists.inkscape.org