Technology

If you run your own AI and watch how long it takes, how much it runs up the resources for a few seconds, then you might get an idea of what it's like hosting at least three copies of a multi-terabyte LLM, in memory, with much shorter response and a much bigger knowledge base (Gemini by Google), taking millions of prompts per minute. Then think of every company that's hosting major public AI services.

Then remember that the only things good that come out of AI are natural language inference for voice commands and slightly improved developer processes.

Hosting it is just a reminder of the rapid environmental, ecological and cultural destruction that is the AI bubble.

In summary: Perhaps, if the hoster wants it for streamlining their dev process. Otherwise it can be replaced with a far more efficient standard algorithmic program, which is what we had before.

Criticism on LLMs has two angles:

• Data usage and energy cost while running it - this you would have under control when selfhosting
• Training of the LLM: You have no control over how the LLM was trained - not which data sources were used, the forkforce that cleaned up the data or the amount and type of energy used to do so

Maybe, maybe not.

The question I have is: How is this the best way for a non-profit to shape and steward the open web?

no

Mozilla is burning donation money on this entry into an already full market, it hooks into cloud models, and at best (if you download one instead) you'll be left with a wrapper around a closed-source binary made by a Big Tech corporation.

Don't donate to Mozilla, because this is where Mozilla donations go. Google funds Firefox, donations (to MZLA) fund Thunderbird. Mozilla Foundation funds get sunk into this.