mysteria

12 ms·

91.
by mysteria 2y ago
How is Tomato compared to the typical x86 choices PfSense or Opnsense?
92.
by mysteria 2y ago
For Nissan according to Mozilla [1]: "You promise to educate and inform all users and occupants of your Vehicle about the Services and System features and limitations, the terms of the Agreement, including terms concerning data collect
93.
by mysteria 2y ago
> If Ford starts snooping on our in-car conversations and injecting ads into our Spotify streams, we can just refuse to buy Fords. Well pretty much all car manufacturers today collect telemetry on driving habits, locations, and such, and
94.
by mysteria 2y ago
> I was convinced TouchID was better than FaceID but I completely changed my mind. FaceID is just more reliable and convenient, especially if you have sweaty palms Actually for me it's the opposite, and I see much more failures to u
95.
by mysteria 2y ago
The 3090s also give you access to all those CUDA inference and training frameworks used in industry. With Apple you're pretty much stuck with either MLX or llama.cpp. Honestly though I'd be curious to see a cost analysis of Apple
96.
by mysteria 2y ago
Yeah that would be awesome but sadly they'd just be confused and annoyed by it all. I mean in my writing communities I'm the only one that uses Git, Linux, and MD. Everyone else just sticks with Google Docs, Word, or some more spe
97.
by mysteria 2y ago
I write fiction using Markdown + Git and love the workflow. Now it's kind of strange treating the whole thing like a software project and having each chapter as a seperate file (along with other folders for notes, artwork, etc.) but it
98.
by mysteria 2y ago
Considering how most of the action on HN happens on the front page this site should be runnable on a little ARM SBC with good use of caching. If a site is mostly static with content in RAM a weak VM or SBC can handle a lot of traffic. I&#x2
99.
by mysteria 2y ago
Anything beyond 10 Gbps is basically the datacenter world and the switches, NICs, etc. are going to be real expensive. I wonder if their terms allow hosting on that 20Gbps plan as that's probably the only real use for it.
100.
by mysteria 2y ago
Assuming those are low latency links the line between the edge and the datacenter would be blurred. For instance it might be possible to train an LLM or have a supercomputer for weather models by combining thousands of nodes across the Inte
101.
by mysteria 2y ago
With modern hardware it's also completely viable to have a Windows VM idling in the background on a Linux desktop to run things like MS Office. Local RDP performance (with FreeRDP/etc.) is excellent and suitable for business or de
102.
by mysteria 2y ago
One cool thing about mpv is that it can transcode audio encoded in different surround sound formats (AAC, FLAC, Atmos, etc.) into regular Dolby Digital for use with old pre HDMI AVRs using SPDIF. AFAIK it's the only player that can do
103.
by mysteria 2y ago
This thing might also be usable in the datacenter as a KVM console that can sit in a rack shelf or be carried around by technicians. It's certainly more compact than the typical monitor and keyboard on a rolling cart.
104.
by mysteria 2y ago
Today a top of the line Epyc Genoa system can have a total of 256 cores and 12 TB of memory, spread out over two sockets. I wonder how that'll look 10-15 years from now.
105.
by mysteria 2y ago
Keep in mind that Gemma is a larger model but it only has 8k context. The Mistral 12B will need less VRAM to store the weights but you'll need a much larger KV cache if you intend to use the full 128k context, especially if the KV is u
106.
by mysteria 2y ago
If the game could work properly with a quantized 7B or 3B it could even be runnable directly in the user's browser with WA on CPU. I think there are a couple implementations of that already, though keep in mind that it there would be a
107.
by mysteria 2y ago
There's additional discussion on the same research in an earlier thread [1]. https://news.ycombinator.com/item?id=40787349
108.
by mysteria 2y ago
It's like any other feature in that the purchase price of the new iPhone and App Store revenue helps pay for the AI functionality. Like they hope people will want to upgrade their phones or switch to Apple for this.
109.
by mysteria 2y ago
I thought most ML engineers use their laptops as dumb terminals and just remote into a Linux GPU server.
110.
by mysteria 2y ago
> Support isn't required for linux: I think there will be people who will be keen to hack up Linux support based on the open source code available, and the same with Windows. I guess this is more of a marketing thing but I wouldn&#x
111.
by mysteria 2y ago
PBS is the Proxmox Backup Server, basically it does differential backups of your VMs, integrity checks, and allows you to back up to multiple different servers. Proxmox does have a basic backup system that can save the VM images to disk tho
112.
by mysteria 2y ago
If you're looking at HCI Proxmox is amazing for homelabbing, with a 3 node cluster or 2 nodes with a qdevice. I use PBS for automated backups and RAIDZ and while there's quite a learning curve and initial setup time actual cluster
113.
by mysteria 2y ago
Cool project, here's a few comments after a quick glance over: - There's no license on the code and design files in the Github repo, thus this currently isn't open source but rather "source-available". If you plan t
114.
by mysteria 2y ago
Since these are audio recordings an easier hardware approach would be to tap the output of the motherboard's D/A converter right before the speaker amp and capture it with an A/D. You could also tap the speaker wires which sh
115.
by mysteria 2y ago
The thing though is that it's entirely possible to have a setup with loud monitoring (e.g. performers on stage with floor wedges to hear the music and themselves) using speakers. It works well with directional dynamics and stage conden
116.
by mysteria 2y ago
If they want that they could easily have a local classifier and only upload suspected images for review. The fact that they're willing to slow down the experience for everyone implies that they're doing training or something.
117.
by mysteria 2y ago
How much do these silu and softmax improvements affect the LLM inference speed as a whole? Correct me if I'm wrong but I feel that this change will only have a small effect as the majority of the time is spent doing matrix multiplicati
118.
by mysteria 2y ago
> Instead it could even absorb every single incidental photon and particle and then actively emit one with the expected behavior as if there was no object in that region of space. E.g. absorb photon on one side, emit one photon on the ot
119.
by mysteria 2y ago
> If the tech is cheap enough, there are a lot of hypothetical upsides to constantly tracking every multi-ton machine on the road. Observe, for comparison, the FAA requirements for flight plans and running a broadcasting transponder to s
120.
by mysteria 2y ago
The bigger question is whether the fine was less than the amount they made selling the location data.

More ›