Amazon, one of the founders of Matter, is the first to add support for Matter casting, a feature of the Matter smart home standard that allows you to control TVs and streaming devices straight from a connected app. At CES this week, Amazon announced that Matter casting support is coming to its Prime Video app so that you can cast content to Amazon hardware devices.
> Is there some sense in which this isn't obvious to the point of triviality?
This is maybe a pedantic "yes", but is also extremely relevant to the outstanding performance we see in tasks like programming. The issue is primarily the size of the correct output space (that is, the output space we are trying to model) and how that relates to the number of parameters. Basically, there is a fixed upper bound on the amount of complexity that can be encoded by a given number of parameters (obvious in principle, but we're starting to get some theory about how this works). Simple systems or rather systems with simple rules may be below that upper bound, and correctness is achievable. For more complex systems (relative to parameters) it will still learn an approximation, but error is guaranteed.
I am speculating now, but I seriously suspect the size of the space of not only one or more human language but also every fact that we would want to encode into one of these models is far too big a space for correctness to ever be possible without RAG. At least without some massive pooling of compute, which long term may not be out of the question but likely never intended for individual use.
If you're interested, I highly recommend checking out some of the recent work around monosemanticity for what fleshing out the relationship between model-size and complexity looks like in the near term.
Folks should be incredibly skeptical of videos like this. The major players working to capitalize on this technology have demonstrated time and again a complete willingness to lie to us about the capabilities of their models. It's become almost a requirement to appear competitive.
Crowdsourced datasets like this and the ones produced by the OpenAssistant project could easily become the ONLY way to build foundational models if the courts decide that what OpenAI and co are doing is not Fair-Use. I don't think I would call this scenario unlikely, either.
I mean, asking them to plop their cli docker setup into a yaml file isn't exactly a difficult task. I think it's probably more correct to consider docker-compose an integral part of docker rather than a superfluous abstraction layer.
They absolutely do not. This is more of a state of the ecosystem sort of problem. I think a lot of developers were wary of getting into how some of the more esoteric AP objects should work. It's important to note too that while we have a standard in the form of AP a lot of the decisions about norms for these less common message types simply have not been made, but we are definitely moving in that direction as more services add AP integration.
I haven't seen anybody else talk about what the experience of following somebody from another service is like on the user-end.
Following pixelfed accounts from Mastodon is actually fairly straightforward. You simply see them as a tweetlike message with an image attached. One caveat there is that Mastodon does not allow for as many attached images as pixelfed, and I believe will simply drop any images after a certain point (I want to say max size is 4).
Following a peertube account or channel (you can do either) will show a post whenever new videos are uploaded. You can find the OP on mastodon at https://mastodon.social/@[email protected]. One neat bit of functionality is that if you put all of your peertube follows into a single list, you've basically got a chronological subscribed page (a la youtube). Also take note that comments can be left on either the peertube site or from within mastodon.
I think what you're describing IS activitypub. The major issue is that each implementor has a preferred message type and only basic functionality of the others. I've posted this elsewhere in the thread, but Lemmy top-level posts are implemented as Pages and communities are implemented as Groups. Currently Mastodon has implemented both but with limited functionality. There is a rework of Groups that is listed as in-progress on the roadmap (https://joinmastodon.org/roadmap).
I think the way we ended up in this situation is that there are quite a few types of messages in the AP protocol, and up until recently the up and coming ones were seeing pretty limited use. Without another service to test against, I can understand why a lot of fediverse developers have opted to kick that can down the road. That said, we are definitely in the era of determining the defacto use case for each of these message abstractions, and I suspect that will be a slow process and involve a lot of back and forth between projects.
To expand on this a little the two services make use of different well-defined objects in the AP standard. Lemmy communities are Groups in AP and top level posts are implemented as Pages (rather than Notes). As it stands Mastodon only has basic support for these two but a rework of Groups is listed on the roadmap as in-progress (https://joinmastodon.org/roadmap).
This is the way forward imo. Particularly as we've started to flesh out the relationship between model size and true context reliability. We've found that raw context-window size is not representative of what the model can actually consistently recall, but we've also found the recall is consistently reliable out to a point. I suspect more robust theoretical models around superposition will move us a long way towards understanding the limits of context reliability rather than what would currently be an experimental approach.
This also brings to mind the concept of federated search. Frequently brought up as a thought experiment, but it's a shame it's not discussed more seriously.
Very cool concept but the cost given by the AI seems to be different from what you're tool is picking up. For example, I've gotten it down to $20 on an item but the interface is showing $24.