This service is all sorts of weird, it feels like some sort of growth hacking thing made by someone who hasn't actually used bittorrent before, and the entirely LLM generated blog posts aren't helping. Torrents are absolutely a good way of distributing weights, but it feels like this exists entirely to suck the good will out of doing so. There doesn't need to be a weird user system, their own tracker, the goofy branding, it could have been a 1:1 tool which just made torrents of every model release and left it at that. That they distribute their own tool for downloading the torrents is especially weird, it's probably fine but also, why? I can't work out why this is using v1 torrents either, it's such a good use of v2 torrents.
> If you run an agent on that model you need those bytes, not a close copy.
A close copy will usually have close performance. Later in the post, they recognize that there are derivatives of the model with very different size and data characteristics. They posted slop anyway.
I would also fire whoever allowed "infohash" to be used in that blog post. It might have been excusable if they explained that "infohash" is Torrent jargon for the root hash of a particularly structured Meekle hash tree over the content of a single Torrent, but ... they didn't.
This service is all sorts of weird, it feels like some sort of growth hacking thing made by someone who hasn't actually used bittorrent before, and the entirely LLM generated blog posts aren't helping. Torrents are absolutely a good way of distributing weights, but it feels like this exists entirely to suck the good will out of doing so. There doesn't need to be a weird user system, their own tracker, the goofy branding, it could have been a 1:1 tool which just made torrents of every model release and left it at that. That they distribute their own tool for downloading the torrents is especially weird, it's probably fine but also, why? I can't work out why this is using v1 torrents either, it's such a good use of v2 torrents.
Cool article but i really hope we fix llm prose in some near future. How chatbots write is so grating.
The blog post is not even close to right.
> If you run an agent on that model you need those bytes, not a close copy.
A close copy will usually have close performance. Later in the post, they recognize that there are derivatives of the model with very different size and data characteristics. They posted slop anyway.
I would also fire whoever allowed "infohash" to be used in that blog post. It might have been excusable if they explained that "infohash" is Torrent jargon for the root hash of a particularly structured Meekle hash tree over the content of a single Torrent, but ... they didn't.
For fucks sake change the name of your website.
Are you trying to alienate all legitimate business?