There are physical prototype units and they've provided as much as information (or more than) Fractal has about the FM3 at this point. That's like saying the FM3 doesn't exist.
But then, I prefer a well written manual to a video tutorial any day.
I guess I should have clarified. There may be some prototypes floating around but, unless someone actually has one to test and review, it’s probably in Fractal Forum’s best interest to monitor and limit discussions to what is actually known and can be verified independently. This isn’t a freedom of speech issue, it’s an integrity issue. Another well known forum receives significant ridicule for its constant bashing, unreliable information and inaccurate reviews.
As for why this is different, the other modelers that get discussed the most are widely in use. The Quad Cortex mostly exists in a rabbit hole on the Internet.
It has already emerged from the rabbit hole. But right now it sounds like a Behringer V-Amp reissue
The touch screen thing looks great, but who the hell wants to spend any time bending over programming stuff? desktop for deep edit / phone or handheld for quick edit? Touchscreen is so 2019
I'm a practitioner in the field. Let me 100% assure you, there is no "deep learning" in Quad Cortex.Is deep learning what makes Quad Corex high expectations
That is what I have suspected. I'm not an expert but I've been studying AI, neural nets, etc. the last few months and the claims don't hold water.I'm a practitioner in the field. Let me 100% assure you, there is no "deep learning" in Quad Cortex.
Inference is associated with deep learning and prediction. This is a neural network, it's not trying to predict something, it's processing a signal. I suppose you could say it's predicting an output but that's not really how this works.I wouldn't expect them to use a big neural network to model the entire input/output chain. It seems more likely that they approximate parts of a predefined topology using smaller local models which may very well be multilayer ("deep") NNs. Similarly for the transient response, which NDSP claims to be a "game changer". This would also allow for quicker inference (claimed 2ms latency?) and should allow them to run it on the hardware.
Exactly this. The display must be of sufficient size and the UI of sufficient simplicity for touch to be practical. I only use a touch screen for my CNC software. And even then, I only use increment/decrement buttons and direct data entry; not continuous sliders and knobs. Fine/reliable control is impossible.My main problem with touch screens is that the interface elements needs to be made bigger so a finger can touch just one thing. Too small and it’s hard to “aim” to grab things. Then each screen/menu has less objects, which means more pages, meaning things are harder to find.
To combat that, people “simplify” the gear so there are overall fewer options and things to adjust making it seem easier and touch-based. But you just artificially have less things to adjust and can’t get the sound you want.
A bit of a generalization but that’s what I see.
Inference is associated with deep learning and prediction. This is a neural network, it's not trying to predict something, it's processing a signal. I suppose you could say it's predicting an output but that's not really how this works.
Also, having smaller "local models" doesn't lower the CPU cost. Sure the NN could be simpler but that's offset by the computation of the "local model".
There's nothing inherently superior about the transient accuracy of a NN vs. other modeling techniques. In fact, giving the horrific aliasing performance I'd wager the transient accuracy is worse since transients have more high-frequency information.