**On making digital sausage**

asanetargoss, 2026-10-27

I want to be transparent toward the people who read and play what I make. I take what I do under this alias somewhat seriously.

The focus of this article is general-purpose generative machine inference, AKA generative AI (such as LLMs).

I'm going to call it "GGMI" for short. This is not standard terminology.

## What I'm doing

None of my stuff under this alias is currently created by GGMI.

I reserve the right to use GGMI in the future for my creations.

I am generally amenable to the idea of disclosure requirements from platforms. It makes sense that people would want to know if some art justifies their attention, given their limited time on Earth, and their limited money.

Full disclosure: At this time, I have used an LLM once to creatively unblock me on the design of a Hardcore Alchemy feature. I was running phi3-mini on my local computer. It wasn't super helpful, but it did technically creatively unblock me.

## My opinions

GGMI is extremely overhyped. I'm looking forward to the market correction.

Energy costs are, in general, a good way to gauge the environmental impact of something that requires continuous energy to run, so I looked into that to decide if I would ever use GGMI in any significant capacity. Running a more powerful LLM is within an order of magnitude of running a microwave for the same amount of time. Image generation is on par with an LLM query, and video generation is much more expensive[^mit]. I did my own napkin math and have found that training is negligible compared to inference for a popular model[^napkin]. Most water usage from datacenters comes from generating electricity to run the computers[^water_to_electricity] (I imagine this is applicable to self-hosted GGMI as well). Overall, I'd say LLM electricity use is not great, but not terrible either, considering I play 3D games on a gaming laptop.

Frankly, I'm not happy about the other negative externalities of GGMI right now.

## Future thoughts

If I were to use more LLMs for my stuff, I would stick to FOSS, due to less risk of vendor lock-in.

However, running the good models tends to cost money, and I'm not super interested in spending extra money on my end to assist in what is essentially unpaid labor for Mojang, especially for modding. As for my fanfiction, I haven't thought about it much.

However, like I said earlier on, I reserve the right to use GGMI in the future for my creations.

## Usage principles

Here are what I think people (including myself) should do with regard to using GGMI:

- Check your work. GGMI makes mistakes.
- Check your ego. Don't be like that meme picture of identical Buzz Lightyears in the aisle of a toy store[^buzz], thinking you're creating something original with GGMI when you're not. Going back to platform disclosure requirements: if you are using GGMI to help produce your art; your disclosure should imply to what degree your use of GGMI results in a creation that is, or isn't, interesting and/or useful. If the result isn't particularly interesting or useful (i.e. slop), exercise prudence in when and where it is shared.
- Have empathy. Don't be like a certain modded community member who told a couple people they would go to Hell for using LLMs. Learn how to give constructive criticism. If you don't want people to use a tool, help them find an alternative that is better for *them* and *their* needs, or help them improve their skills to make them less dependent on the tool.
- Use more efficient tools when practical.

To be clear: It is not my intent to turn my existing creations into slop, nor publish prose with an LLM personality. I am not interested in "MCreator 2.0"[^mcreator] level use cases.

## Conclusion

Thank you for your time. In the spirit of the sausage-themed title, here's the [How It's Made video on hot dogs](https://www.youtube.com/watch?v=D42mVB3sEUg).




[^mit]: https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/

[^napkin]: In 2025, I used llama2/llama3 as the reference models, and guessed llama usage based on ChatGPT's usage at the time.

[^water_to_electricity]: https://www.construction-physics.com/p/i-was-wrong-about-data-center-water

[^buzz]: https://knowyourmeme.com/memes/buzz-lightyear-clones

[^mcreator]: In fairness to MCreator, the Hardcore Alchemy modpack has an MCreator mod called Alchemic Ash, which was created by a tabletop game designer.