It seems important to support this sort of diversity and openness in AI. Even if you don’t necessarily support the US, avoiding any single bloc dominating this technology is valuable for everyone.
Given the grind has already been specified as an espresso-level of fineness (we're talking about a cold brew recipe), most of the grinds are going to be able to get through the metal filter.
The original suggestion of plunging slowly and gently is entirely sensible, since it's the only variable available in the context.
Not the poster you're asking but they might have imagined the stirring being vigorous enough to keep the liquid moving during the entire heating period.
The “mixture of experts” concept in LLMs is a way of training a single model, it’s not based on training many different models (although that was the idea when the term was originally coined).
The main tech is a very anti-reflective coating; the best I've seen on a mainstream TV. It is backlit but in Art Mode the backlight is turned down (you can control how much and there is an ambient light sensor that fine-tunes your choice). The combination of low emission, quite low reflection and static content is pretty compelling for me in a lit room.
I have the white frame on the bezel against a light-coloured wall and when displaying art it's much less imposing on the room than an empty black screen. Obviously it uses more power than standby but it's also quite a bit less than a dynamic screensaver.
Don't know that I'd get another one; the bugs in Samsung's software are annoying (I just want it to display the art, or a picture from a single HDMI input, how hard can they possibly make it?) but perhaps that's standard for modern TVs.
Their Art Store promotion is annoying persistent but not actually required if you put images onto a USB and have it slurp them up; all you need to do is ignore their requests for subscription when setting it up. All that's required is to resize them to full screen with your own matte, or you can just crop the images to the right aspect ratio. Depending upon your desire for fidelity, you could also use ML image extension to get the right aspect ratio.
This doesn't get around all the other annoying bugs in the firmware of course.
They make it superficially closed but it's pretty simple to get around it with a USB drive. Really all you have to do is ignore their subscription system and size your images to match the screen.
If the hammer isn't mechanically coupled to the keys he could also have addressed functionality such as the width of the keys instead of just aesthetics. The standard size piano keyboard is too wide for most female hands and many men to play advanced pieces comfortably (or at all).
Hi Timo, is there a document somewhere that describes how Conduit compares feature-wise to Synapse or to the Matrix spec? It’s not clear to me if the goal to develop a fully-featured matrix server, or if there is some subset of scale or functionality that you’re aiming for; could you comment on that?
How does this relate to Synapse? Dendrite tests against https://github.com/matrix-org/sytest and has issues for “are we synapse yet” — I can’t see any discussion of this for Conduit (not even a discussion on why they don’t want to be like Synapse).