In my opinion, throw away code that noone will have to see, it is acceptable to produce with AI.
The artifact should also be thrown away, meaning that it should not persist, because then we have an artifact that we do not understand how it works.
Disregarding the environmental impact of AI or its centralization and control by private interests , an exploratory use of AI can be beneficial.
But again, the effects of its use are multiple that can not be understood right now. We need a place to discuss these effects while at the same time defending against the onslaught of slop.
Hi there - this poll comes at an odd point in my trajectory, since as of last week I’ve transitioned from building a substrate with 0% LLM input to one where essentially 100% of the code is generated in that way - similar to Tom’s trajectory but on a different timescale. I’ve realised in this time that I would have had no prospect in my lifetime of effectively delivering the substrate I planned without it.
I think that anyone committed to producing artefacts that solve problems for real communities will have to accept that substantial contributions to their implementation will come from LLMs. An important focus for the malleable systems/substrates community is to ensure that these contributions come in a maximally ownable way - that is, that they are generated from LLMs which expend the least power and are themselves maximally ownable, ideally sitting under someone’s desk.
If this community were to persist in the uncivil demonisation of people who use LLMs to assist their work and enact the ban that is proposed, I would sadly have to leave it. It would be a shame, since I’ve had numerous productive conversations and great relationships with people here, but in my opinion such a ban would be an act through which the community declared its irrelevance and intolerance.
LLMs are fundamentally opposed to the principles of open-source software. I cannot inspect or modify their training data and regenerate the model weights at will, even for “small” models that can be run locally; they are an opaque residue distilled at enormous expense from truly staggering quantities of creative work without consent or acknowledgment of the people who made that training data in the first place. There is no use of these models which does not represent a hard dependency on centralized infrastructure and willful erasure of authorship.
I am deeply unimpressed with the suggestion that LLM usage is inevitable or indispensable, especially on the basis that they accelerate the speed or scale of development. A larger, more complex project, fabricated hastily through obfuscated plagiarism, and understood more superficially even by its “author” is not a richer one.
Thanks all for the discussion thus far. While each of us may have quite different opinions on this front, I am at least glad we’ve managed to keep the discussion respectful. I know this space can be entangled with strong emotions and passionate opinions, or at least it is for me…!
I plan to keep this thread and poll open for at least a few more days to collect any further feedback. I am also still processing everyone’s thoughts and attempting to imagine next steps from here.
A lot of rhetoric here. AI is indeed a greatly negative development for mankind as a whole, and I heartily wish it had not occurred. However, here it is, and it befits someone with a responsible stance towards the issue to make the best of the situation as we find it. In detail, also, what you say is not true - the IFM K2 model https://ifm.ai/k2/ , for example, my current “model of choice” for my in-substrate assistant, is an entirely open-source model for which not only the training weights but training corpus is fully inspectable.
Research on how we maximally preserve provenance of our artifacts in a world where the majority of code is written by AI is a responsible use of our resources. In addition, it is not entailed by the use of AI that its output is “slop” - this is merely the result where it is set loose on poorly-founded tasks using inappropriate notation. Seeing how we can improve on this situation is also a responsible use of someone’s time.
Sticking one’s head in the sand and saying one will have nothing to do with the technologies that will be predominantly used in what was once our profession, and not even allow them to be discussed, is not so responsible.
I don’t think there’s an easy way to achieve that also respects everyone’s privacy. The high-level summary is we’ve had a few posts on existing threads and also some new threads created (primarily from new accounts with no prior activity) either sharing projects produced with generative AI tools or tools for working with generative AI outputs.
At the time, I made ad-hoc decisions while attempting to “read the room” because we had no policy. I think it’s best to keep this thread focused on what we want going forward.
Yes, that is indeed what the option to ban projects created with generative AI tools would entail, should we move forward with it. If we do move forward with that stance and your project has used those tools, it would no longer be welcome here.