dheera

32 ms·

571.
by dheera 2y ago
If you want a vramfs, why would you use GPU VRAM? CPU<->GPU copy speeds are not great. I have 192GB of CPU VRAM in my desktop and that was cheap to obtain. Absolute best build decision ever.
572.
by dheera 2y ago
Sort of! If you shut your eyes off and you hear footsteps to your right you have a good idea of exactly what you're hearing -- you can probably infer if it's a child or adult, you can possibly infer if they are masculine or femini
573.
by dheera 2y ago
Idea I've had for years but never got around to testing due to lack of compute: Use a microphone array and LIDAR for ground truth, and train a diffusion model to "imagine" what the world looks like conditioned on some signal
574.
by dheera 2y ago
OK that effect is super interesting, though if you assume all the computational pathways happen in parallel on a GPU, that doesn't necessarily increase the time the model spends thinking about the question, it just conditions them to g
575.
by dheera 2y ago
> the more thinking time it gets That's not how LLMs work. These filler word tokens eat petaflops of compute and don't buy time for it to think. Unless they're doing some crazy speculative sampling pipeline where the small
576.
by dheera 2y ago
Generating good meshes sounds like a problem for a completely different machine learning algorithm to me.
577.
by dheera 2y ago
Not necessarily. If you're using it to render video you don't need to go into the mesh world.
578.
by dheera 2y ago
A lot of people say this, but you miss 100% of the shots you don't take. If you think you can't do it because there are "better people already doing it", or because academics say the markets are efficient and therefore y
579.
by dheera 2y ago
I see. I have the "Stocks Advanced" real-time data for individuals feed. I didn't realize the "professional" real time feed isn't the same feed.
580.
by dheera 2y ago
Love it. I would absolutely love if they would consider making the Core Time 2 in a slightly more timeless outer case than this "sport" look.
581.
by dheera 2y ago
I have a Polygon real-time market data subscription and everything comes real time except for darkpools. From the customer service bot: "Yes, other trades are generally reported faster than dark pool trades on our WebSocket. We stream
582.
by dheera 2y ago
I tried to train a model to predict the darkpool transactions from the rest of the transactions. It didn't work. I'm back to the drawing board trying other things now.
583.
by dheera 2y ago
I don't know, honestly. I would imagine it to be illegal to trade on that data before it's released to publicly-available trade streams. I'm of the opinion that a huge amount of manipulation happens and the SEC either turns a
584.
by dheera 2y ago
I've been wading through terabytes of financial data over the past month. (I'm an ML/AI software engineer trying to create a trading bot that makes me more consistent income during times of uncertainty that are outside my con
585.
by dheera 2y ago
I'd love to have better backup power but I'm skeptical of huge lithium batteries inside my home. Here's a tiny scooter battery going on fire spontaneously: https://www.youtube.com/watch?v=ld00r-tEEJ4 Can we u
586.
by dheera 2y ago
I just wish the stupid chatbots on customer service websites would show "Get human" as an option sooner. 99% of time there is a mistake and the bots don't have authority to correct it.
587.
by dheera 2y ago
Sure, if you want to do training get an NVIDIA card. My point is that it's not worth comparing either Mac or CPU x86 setup to anything with NVIDIA in it. For inference setups, my point is that instead of paying $10000-$15000 for this M
588.
by dheera 2y ago
Apples to oranges. NVIDIA cards have an order of magnitude more horsepower for compute than this thing. A B100 has 8 TB/s of memory bandwidth, 10 times more than this. If NVIDIA made a card with 512GB of HBM I'd expect it to cost
589.
by dheera 2y ago
It will cost 4X what it costs to get 512GB on an x86 server motherboard.
590.
by dheera 2y ago
> That was before "move fast, break things and blame the user" became the norm. When VCs only give you effectively 9 months of runway (3 months of coffees, 9 months of actual getting work done, 3 months more coffees to get the
591.
by dheera 2y ago
> We use the Musical Instrument Digital Interface (MIDI) standard to represent a musical piece as a sequence of time-stamped messages corresponding to "note-on" or "note-off" events. A message carries additional piece
592.
by dheera 2y ago
> If you're confident in your approach That's the thing. I'm not a narcissist, and my confidence in my approaches is driven by the objective statistics and uncertainties of the approach and NOT my ego. If I think there&#x2
593.
by dheera 2y ago
I'm swimming in taxes and treading water is hard enough already. Maybe if they taxed me less, I donate more to 501c's that do more per dollar than the government does.
594.
by dheera 2y ago
Oh the bullshit they brainwashed you with. What infrastructure? Non-existent healthcare? Non-existent public transportation? Non-existent fiber to the home? Non-existent policing of crimes? Schools that look like jails? Homes built of shitt
595.
by dheera 2y ago
How much do you pay in rent? The majority of that is going towards paying your landlord's taxes. Those are your second-order taxes. How much of that is left over for your landlord to spend? When they spend it, how much of that money is
596.
by dheera 2y ago
In a progressive tax system, the poor end up paying a lot of the taxes of the rich, because the rich tend to own businesses and real estate and they just pass all their income taxes onto the poor in the form of price increases. If you are a
597.
by dheera 2y ago
The related problem I have with LLMs and coding is that sometimes it does something right, I accidentally make a wrong suggestion, and it just goes with my suggestion without telling me that it thinks what I suggested is a bad idea and if I
598.
by dheera 2y ago
I really never understood why hobby servos are so shitty. Their protocol uses only a very narrow range of PWM values and not the full 0-100% scale and that alone completely wrecks their precision beyond what their motors can achieve. Also t
599.
by dheera 2y ago
Why tf isn't ollama in apt-get yet? F these curl|sh installs.
600.
by dheera 2y ago
I always thought bitter stuff would be a better repellent than spicy stuff.

More ›