On Sourcehut

Draft 2 minutes read

For the first time in my life I have actually read an email about the update to TOS. [Read the blogpost], the choice is well argumented.

Recently I had rather pleasant experience with LLMs:

I hosted all of these projects on Sourcehut, I no longer do. I opted to run my own instance of sr.ht. This choice came easily. All of my repositories, except for Swolog, are private. I believe that sharing LLM-generated code publicly is worse than useless.1. I would like to continue sharing the Swolog code2. I will not.3 My opinion on the ultimate usefulness of LLMs is still forming, so I will not yet take out pitchforks.

Overall I applaud Sourcehut’s decision and will continue to host my hand crafted code there. Notably this blog. Sadly there is not much of it since I try to use the little free time I have on reading and, hopefully—writing.


  1. We know that training models on generated data makes them worse. More links on the Internet will result in exponential crawling load. ↩︎

  2. The majority of the codebase predates them AIs. ↩︎

  3. Large parts have been generated, and it also includes a whole agentic coach. ↩︎