dmw_ng

The same is true pretty much everywhere else (cryptography, sport, electrical engineering, aeronautics, ..). There is no greater teacher than a head injury or an electrical fire, the question is whether OP's fear is warranted or whether it should be encouraged. Perhaps they should not set the bit rate on a video codec until completely understanding the fundamentals of that technology too? I'd suggest like in many similar areas, gatekeeping is a not insubstantial force

Believe it or not this stuff is easy mode once you learn to ignore the terminology. It takes relatively little knowledge to know how to get data into the right format, set sensible training parameters and check results -- basically a handful of function calls. The trained models are also highly interpretable, you can literally plot them with Graphviz to understand how the model is treating values of each feature ("column")

I'm not so sure about this, their description specifically calls out one-time-only costs like domain registrations as being the only things excluded. Cautiously optimistic in addition to the recent reappearance of new GPU models being offered to existing customers, Hetzner may be prepping an hourly-billed bare metal GPU product at a great price

Hetzner bare metal already deploys in a couple of minutes when renting existing hardware, and there is plenty of precedent for hourly-billed bare metal services around. I think they're intending to move all their bare metal over to hourly billing.

I don't think it's fair to call it their free tier - it's their discretionary tier, there are numerous cases of the rug being pulled as and when it suits their business requirements to do so. Being left homeless vs. urgently coughing up is exactly the wrong problem to be dealing with mid-attack, I can't see any way to consider it free by any practical definition

Given the size of the Sindarin team (3 AFAICT), that mostly looks like a clever combination of existing tech. There are some speech APIs that offer word-by-word realtime transcription (Google has one), assuming most of the special sauce is very well thought out pipelining between speech recognition->LLM->TTS

(not to denigrate their awesome achievement, I would not be interested if I were not curious about how to reproduce their result!)

> but formed a very strong opinion about it based on basically nothing.

3.5 ms is between 5x - 53x worse than what you'd see out of gigabit ethernet, depending on the quality of the equipment, but even at the low end this is absolutely something noticeable in any application heavy on round-trips (such as those mentioned). Nothing breaks my focus more quickly than an unpredictably slow or janky UI, it is far from a made up scenario, just open an explorer window pointing at an SMB share with a few hundred JPEGs with uncached thumbnails and you've already hit it. I'd much prefer consistency of a slower network than usability that evaporates at random, else what is the point of preferring a wired network over wireless in the first place?

FWIW, listing a directory or opening a single file over SMB is at least 4 round-trips, that's 14 ms burned on lag just for a single thumbnail, in the best case, and it only multiplies from there.

AIUI the minimum latency on these is always measured in milliseconds, with 6-7ms apparently commonplace, which you're going to immediately feel using e.g. a block or network storage protocol. No references, but did look at moca for exactly the same reason a few months back. The physical size of the models I looked at also suggested you'd want to check power consumption before stuffing a home full of them - custom devices in a small market segment often drink electric.

I'd much prefer a consistent <1ms latency 10 Mbit connection than 10ish ms gigabit with 100ms spikes when next door boils their kettle, etc.

I kept the motherboard as a compromise. Serious corrosion at this point but very unlikely to ever throw it out

Wouldn't mind knowing total number of hours / audio / video codec / resolutions for that 30TB collection. Immediate thought is ridiculously overspecced codec configuration or signs of an addiction. Are they stored as lossless 10-bit or something?

I'll take 10TB perpetual for free in exchange for sorting out your web server configuration

The latency on this (or lack thereof) is the best I've seen, would love to know more about how it's achieved. I asked the bot and it claimed you're using Google's speech recognition, which I know supports streaming, but this result seems much lower lag than I remember Google's stuff being capable of

I think good in-home vision models are probably still a little bit away yet, but it seems already the case you could start to plan for their existence. It would also be possible to fine-tune a puny model to trigger a function to pass the image to a larger hosted model if explicitly requested to, there are a variety of ways things could be tiered to keep processing that can be done practically at home at home, and still make it possible to automatically (or on user's request) defer the query to a larger model operated by someone else

There must have been 100 folk with the same idea at the same time, I'm very excited for having something like this running mics in my home so long as it's running locally (and not costing $30/mo. in electricity to operate). Lots of starter projects, feels like a polished solution (e.g. easy maintainability and good home assistant integration etc) is right around the corner now

Have been tempted to try and build something out myself, there are tons of IP cameras around with 2-way audio. If the mic was reasonable enough quality, the potential for a multimodal LLM to comment contextually on the scene as well as respond through the speaker in a ceiling-mounted camera appeals to me a lot. "Computer, WTF is this old stray component I found lying under the sink?"

Ran into a variant of this when working on an IO loop, ended up with a little "Side" abstraction to wrap bidirectional/unidirectional descriptors to behave identically for the rest of the library. It's an annoying wart but pretty easy to avoid in userspace.. https://github.com/mitogen-hq/mitogen/blob/798032b9/mitogen/...

The trouble with OAuth is to get a production client ID you must pass a third party security audit.. this is in excess of $20k and AIUI must be repeated periodically. Using a developer client ID is already heavily limited, and I have no doubt now that this ladder has been pulled up, developer IDs will see further restrictions in future.

IMAP/POP passwords have long defaulted to disabled in Gmail, Gmail survived 20 years without need for these new restrictions, I can't imagine attack techniques have improved and Google's internal technical staff have regressed so substantially that they are now essential. This change seems more motivated by creating frictions for escaping the Google vortex than anything to do with security.

> putting out press releases about raids before they happen

This sounds like a Chris Morris sketch, got a link to share?

edit: I think it's referring to the Cambridge Analytica scandal

That sound you hear is the hellish screeching of a million git status monorepo users waiting for their shell prompt to reappear

You so much as blink at these cards and they mysteriously stop booting, my 4K-capable A7R IV tops out at 100 Mbit/sec write rate for video, that's almost 24 hours worth on a 1TB card.

Given the technical limitations of these devices, already storing that much footage on a card that is so slow to backup and with such limited rewrite lifetime seems insane. I love the idea of 2TB in such a tiny package, but the reality of who would actually buy these in large numbers and for what does not quite seem to line up with the tech.

It's not just cache locality, but also at least the effectiveness of cache prefetching and the ability to use wider operations.. simple array loops are something compilers can often vectorize without assistance.

Not least for the sake of power management, the central push provider needs to authenticate (and e.g. rate limit) notifications to a particular device. The identities of apps communicating with a particular device therefore seem to need to be known

This still seems like something that could be fixed with smarter design without losing functionality.. e.g. decoupling device registrations from push channels, and treating the push channel ID a particular device is using as toxic for sharing and intermingling as other kinds of personal identifiers like phone numbers, including with the OS provider itself, or creating a unique push channel ID for every app registration, etc.

Is there some place with a technical writeup of what YT are doing to frustrate circumvention so effectively?

According to stackoverflow over 70% of the industry are younger than 34, I think a few grey hairs are needed before "because security" attains its invariable meaning

Seems to run about "2x realtime" on 2015 4 core i7-6700HQ laptop, that is, 5 seconds to generate 10 seconds of output. Can imagine that being 4x or greater on a real machine

There's been instructions for manufacturing weapons useful for terrorism floating around since the BBS days, nothing new here

More interested in how the PCIe breakout board being used works.. is there some kind of bridge to a mini PCI slot?

I run two XPS 15s (2015ish 9550 workhorse / main computer, 2022ish 9250) and haven't had the slightest issue with either, although I keep the on board Nvidia disabled on both, which might explain the "reasonable power consumption" bit.

Hardware on the other hand, these things are like caring for an anorexic bubble baby with severe bone density issues. The hardware is a piece of shit, but also extremely easy to repair. Burned through at least 3 keyboards on the 2015, at least 4 pairs of speakers (new set sitting in a parcel in the room right now), 1 motherboard (mechanical stress snapped the heatsink), 2 power plugs, replacement top case because of screen latch stress snapping some tiny piece of plastic, probably more I've forgotten. Separately, the mic seems to be installed literally inside the fans, they made the screen bezel so thin on the 2022 model that they couldn't even include a web cam competitive with the 1998 Game Boy camera, it's embarrassing to videoconf from it

It's been a long day, managed to take a few seconds wondering what this interesting new "best CPUs" technical architecture was before realizing the article is SEO'd blogspam.

Maybe switch to the source article? https://www.servethehome.com/fadu-cxl-2-0-switch-and-pcie-ge...

I didn't think you meant that literally, but after paging through a few tens of google image search results for "krita samples", it seems the only thing Krita is actually used by people for is fantasy art.

Compare to "corel painter samples" where the 2nd thumbnail is a fairly impressive Audrey Hepburn rendition

Can't remember the last time I saw output of the style from the first search result being used to sell pretty much anything

I genuinely wonder how much that anime waifu logo is costing the project

Plenty of locations that used to be BT phone boxes were replaced with pillars featuring WiFi and USB charging. Is this not pretty widespread?

This device seems a bit.. daft. Renting a Santander cycle costs 1.65 GBP. Add the high energy food required to supplement your diet after a 30 minute ride and this is looking closer to 3 or 4 quid for a phone charge, and that's before considering where the homeless person is going to securely stow their belongings while cycling. Luggage storage in Victoria is 5 GBP for 3 hours..