modeless

27 ms·

601.
by modeless 1y ago
The models (weights and activations and caches) can fill all the memory you have and more, and to a first (very rough) approximation every byte needs to be accessed for each token generated. You can see how that would add up. I highly recom
602.
by modeless 1y ago
Nvidia DGX Spark has 273 GB/s (2184 gigabits with your units) and people are saying it's a disappointment because that's not enough for good AI performance with large models. All the neural accelerators in the world won'
603.
by modeless 1y ago
> remarkably, nearly all the end-user consumer Internet browsing and app traffic we observed used TLS or QUIC There was a surprising amount of resistance to the push to enable TLS everywhere on the public Internet. I'm glad it was
604.
by modeless 1y ago
WDDM is one of the best and most underappreciated parts of Windows IMO. GPU drivers work better on Windows than on any other platform (including macOS with third party GPUs). An embarrassingly large percentage of my problems with Linux over
605.
by modeless 1y ago
> an inaccurate prediction market can still inform your knowledge updates I didn't say an inaccurate prediction market is completely worthless. But it is self-evident that an accurate prediction market is a lot more valuable than an
606.
by modeless 1y ago
Emperically, we have plenty of participation by the public in prediction markets without a prohibition on insider trading. We don't need to encourage any more. > Allowing insider trading seems like a nearsighted way to increase volu
607.
by modeless 1y ago
The purpose of banning insider trading in markets is not "fairness". The purpose is to encourage participation. We want to promote investment in the stock market by the general public who don't have insider information, becau
608.
by modeless 1y ago
What are the reasons why nobody uses pypy?
609.
by modeless 1y ago
Why not? It has microphones.
610.
by modeless 1y ago
We will never run out of small inconveniences. Today's world would look impossibly convenient and easy to anyone from 100 years ago, let alone 1000 or 10,000. Yet we still perceive hardships. Humanoid robot servants won't change h
611.
by modeless 1y ago
It's the marketing name for their in house AI model.
612.
by modeless 1y ago
> Charging coils in the robot’s feet allow it to simply step onto a wireless stand and charge at 2 kW. This is silly. Wireless charging is inefficient and costly compared to cables but we use it for the convenience of humans, to avoid th
613.
by modeless 1y ago
Netflix stopping DVDs by mail?
614.
by modeless 1y ago
It probably does have different modes that it selects based on the input data. I don't know that much about the implementation of image compression, but I know that PNG for example has several preprocessing modes that can be selected b
615.
by modeless 1y ago
No, not really. They are both cool but solve different problems. The problem Basis solves is that GPUs don't agree on which compressed texture formats to support in hardware. Basis is a single compressed format that can be transcoded t
616.
by modeless 1y ago
Part of curing aging would be restoring youthful brain characteristics such as openness to change. Which honestly seems like a small and easy task when compared to the whole endeavor of curing aging.
617.
by modeless 1y ago
Never thought I'd see so many people rooting for death here on HN. What a dour, pessimistic place this has become.
618.
by modeless 1y ago
Touch is a 2D field of 3D vectors. Easily stored and transmitted as images, and easily processed by neural nets. You could add temperature and pain/damage channels if you want, though they don't seem essential for most manipulatio
619.
by modeless 1y ago
Is that not authentic?
620.
by modeless 1y ago
Does Steam do something like this for game updates?
621.
by modeless 1y ago
Intel's best single core score is listed as 3240 for the Core i9-14900KS, a 250W desktop monster chip. Is this score of 4133 for an equivalent test? Is Intel that far behind?
622.
by modeless 1y ago
> or to consume interesting stories
623.
by modeless 1y ago
I can see it being interesting to create wacky fake videos of your friends for a week or two, but why would people still be using this next year? I watch videos for two reasons. To see real things, or to consume interesting stories. These v
624.
by modeless 1y ago
Shame, my impression is that ChromeOS did a lot of the foundational stuff better than Android. Would have been better to merge them the other way, but that would never fly politically.
625.
by modeless 1y ago
It's been 27 years since Google Search launched, and I still find it very useful despite some perverse incentives. A pretty good run I would say. If OpenAI declines that slowly, and is then displaced by something better, I'd say i
626.
by modeless 1y ago
Yeah, it seems obvious that this is how models will be monetized in the future. The free version of ChatGPT will stop being a loss leader for the subscription and start paying for itself with commissions. The vast majority of people will us
627.
by modeless 1y ago
OpenAI and Anthropic are both trying to automate their own AI research, which requires coding.
628.
by modeless 1y ago
It's not against YouTube's ToS to run pre-roll ads of any length, of course. But I think he was just selling get-rich-quick schemes or something. At the time YouTube ads were more rare and generally short, like 15 second TV commer
629.
by modeless 1y ago
If Google really goes through with this I might seriously consider GrapheneOS. At least Pixel hardware ought to still support unlocking the bootloader. But for how long...
630.
by modeless 1y ago
People may remember Tai Lopez from his "Here in my garage, just bought this new Lamborghini here" YouTube ad, subject of several parodies such as: https://www.youtube.com/watch?v=0GIwTG8V-Ko I think he was one of

More ›