Digital communities (like #Fediverse instances) - where discourse about new knowledge happens - need to think seriously about their stance on conversation data being used for training large language models.

A few possibilities:

a) Ban it altogether. This will require closing down our communities and implementing proof-of-humanness, reputation scores etc.

b) Exclusively license the data to #AI companies and use that to pay for server costs.

c) Change nothing. Allow anyone to use the data.

@nik @jenniferplusplus IRI seems to be about just allowing extra characters, not content-addressing.

It would be easy enough to build content-addressability within existing URLs. For example:

`https://myblog.com/my-article-<multihash>.html`

Web browsers could have built this universally for all resources when defining the SubResource Integrity spec which is ensures that JS scripts a page downloads has not been tampered with. They chose `<script integrity="<hashalgorith>-<hashvalue>">` instead.

@nik @jenniferplusplus What's this IRI you speak of?

@jenniferplusplus I also think #ActivityPub spec seems to be designed in early 2000s where users were supposed to just trust the servers. The only way I can truly own my identity is by running my own AP server on my own domain. Content is also referred to by its URLs, not by its content-hash which makes sure that I remain dependent on other servers/intermediaries.

#Bluesky at least recognizes some of these issues.