Post by @docyeet@mastodon.halis.io

So I have been having this idea for a while, and finally got around to making a prototype (generating with Claude, more exactly)

The principle is simple, right now authentication for Matrix bridges is rather tedious... As in, copying a request from the debug console in the browser tedious, so not very accessible to the average Mom for example (sorry IT Moms around here)

So the idea is, why don't we publish a browser extension (here in Zen Browser) to automatically detect available bridges once logged into a Matrix user account, and give the possibility to be fully walked through the login process, automagically

No finding the request manually, no copying to weird curl formatted tokens, no confused "just press F12" moments, just click the "Log in" button

What are people's thoughts ?

#homelab #matrix #synapse #element #bridges #selfhost #selfhosted #selfhosting #opensource #federated #Fediverse #elementhq #messenger #facebook

#elementhq #bridges #synapse #federated #element #matrix #messenger #homelab #selfhost #selfhosted #opensource #selfhosting #facebook #fediverse

0 Boosts

0 Likes

happyborg

@happyborg@fosstodon.org 2 hr. ago

@docyeet my thought is always, use of #GenAI is irresponsible because of the widespread serious harms it is causing.

DocYeet :verified:

@docyeet@mastodon.halis.io 2 hr. ago

@happyborg Fair, I understand
Is selfhosted LLMs falling under the same umbrella ? Taking aside the training part of the model for which I cannot disagree, it is a big source of harm

happyborg

@happyborg@fosstodon.org 1 hr. ago

@docyeet Not wishing to sound patronising, here's a chap I recommend you follow for when he dives into the reality of these models and their masters. He's picked many apart in true hacker coder style.

Here's his latest which is both jaw dropping and very amusing, in case you've not seen it. Well worth the time reading him and the replies IMO.

https://neuromatch.social/@jonny/117324790823856750

View

happyborg

@happyborg@fosstodon.org 1 hr. ago

@docyeet yes to both I'm afraid.

There are various reasons for this view. One is that any use tends to normalise and so encourage more use. Another is the 'cost' in various senses of training the model you use, local or not. And updating of course.

I don't get into debates about this because the evidence for this is out there for anyone who wants to look, but most people do not want to look and will rely on justifications around their particular use case, or method of use etc.