scw

As a kid, my first machine was a 486 DX2 66. My dad mandated we buy on value, and I scoured the MicroTimes looking for the best possible deal, finally settling on a machine from a small independent shop off of South Bascom Ave in San Jose. The office smelt like burnt coffee and exhaust from burning-in PCs, but the machine was awesome: a PCI video card, and as we later found out, a built-in SCSI adapter which meant we needed to throw in another $240 for an Adaptec SCSI controller when we later upgraded to Pentium.

The original post [1] now includes an update:

  UPDATE! Within a day of this blowing up on Hacker News, AMD reached back 
  out to me and said they would be looking into the matter after all.

They're unlikely to find much unless they instead try “dispositive”.

TurboQuant has a specific benefit by compressing the KV cache at a negligible cost to quality. That mainly means that the context lengths can go up in models for the same amount of memory, however the KV cache only accounts for something like 20% of the overall model size, and this will not dramatically decrease memory demands in the way that some of the more sensationalist reporting has stated.

Still active, but many fewer resources than in the past. Many backends like CUDA for Windows have been dropped and others pushed off to partners with varying levels of support. TensorFlow 2.19 is going to release soon without Python 3.13 support, it's hard not to imagine that resource constraints are at play.

The Microsoft stake finally allowed him to let loose and buy a keyboard with a working shift key.

Exciting concept! Note that the LLM corrected version does drop a full paragraph from the output at the bottom of the second page (starting with an asterisk and "My views regarding inflationary possibilities". I'm not sure if there is a simple way to mitigate this risk but would be nice to fall back on uncorrected text if the LLM can't produce valid results for some region of the document.

I recently had occasion to evaluate a database of 1200+ NVIDIA GPUs and can tell you that the only thing consistent about the model numbers is their inconsistency. For example, what is an RTX 4000? It could be the 2018 Quadro RTX 4000, the Quadro RTX 4000 Max-Q, or Quadro RTX 4000 Mobile (all Turing cards), but it could also be the RTX 4000 Mobile Ada Generation (Ada Lovelace card released 2023).

Reminds me of “boko-maru” of the made-up religion [1] in Vonnegut's Cat's Cradle.

1. https://en.m.wikipedia.org/wiki/Bokononism

The methodological approach and data sources are detailed in the associated post by JHU professor Lauren Gardner: https://systems.jhu.edu/research/public-health/ncov/

Check out: https://medium.com/microsoft-open-source-stories/how-microso...

If you're revisiting the classics I can't recommend enough Doug Metzger's Literature and History podcast [1]. It covers literature starting with Mesopostamian stories, at about the level of an undergraduate course, but is entertaining and insightful throughout. It's clearly had deep research put into every episode, but at the same time takes great effort to make the material relatable. Great stuff.

1. http://literatureandhistory.com/

Endpoints are the machines (desktops, servers, &c) in organizations which have Carbon Black installed. Their client continuously monitors for process executions, network connections, file changes, registry changes, and samples unique files on the machine, and depending on configuration, can upload these contents both within the enterprise and share them with their cloud platform. That's what is meant by "our technology uniquely collects complete, "unfiltered" endpoint data by continuously recording endpoint activity and centrally storing the collected data for advanced analytics".

This helped me: https://threadreaderapp.com/thread/976563870322999296.html

The PCEngines APU2 boards[1] are x84_64, run around 10W, and have a smaller than mini ITX footprint (6"×6").

1. http://www.pcengines.ch/apu2.htm

It's in a peer-reviewed scientific journal, written by academics in the field, not by T&F itself, nor funded by them. While there are problems with the peer reviewed system, publisher interference with research results isn't typically one of them.

Anything that depends on both boost and gdal is bound to be pain. I once spent 22 hours getting a build of GDAL on FreeBSD with the drivers I needed.

For the Vagrant VMWare support, are you using the official (paid) Vagrant provider or something else? I've been interested in hearing about experiences of using it.

That's a big ask for a site that's essentially 'morally justified' piracy.

My first summer job was doing IT for a small outfit, who worked out of an old house converted into an office. They had eight machines, all connected to a LaserJet printer via a parallel port switch. The parallel cables ran down into a crawlspace under the house, and it wasn't uncommon for one of the cables to work its way loose. No problem, tighten the parallel port connection at both ends, and you're back in business. But, being an old house, many outlets were missing ground pins. I must of electrocuted myself twenty times that summer.

Perhaps, but this patent (https://www.google.com/patents/US8407580) and a history of litigation (e.g. (http://bactra.org/reviews/wolfram/) may be enough of a barrier to prevent its adoption.

PXE booting OpenBSD is covered in the FAQ: http://www.openbsd.org/faq/faq6.html#PXE

The board has been in beta testing for a few months, folks were discussing it on the OpenBSD -tech list.

The APU is great. The APU2 was recently released: http://pcengines.ch/apu2b4.htm

Improvements over the first version: quad core, Intel Gigabit NICS, new CPU supports AES-NI and AVX (AMD-V), ECC memory, and a USB 3.0 bus. Still some issues being ironed out, but will make excellent router hardware. Note that WiFi is a relative weakness on BSD, and you'll need to carefully choose a chipset to support hostap mode. The Atheros abgn cards are generally the best.

The X220 takes 16GB, and an onboard gig-e Intel nic. It was released in 2011.

Not a direct answer, but I use the Windows port of The Silver Searcher (ag) and it works well: http://blog.kowalczyk.info/software/the-silver-searcher-for-...

It is more widespread than you might think. This article tells the story of the potential role of milk in the growth of the agrarian way of life: http://www.nature.com/news/archaeology-the-milk-revolution-1...

Most of the places selling these files don't include disclaimers about how the additional quality is only useful in studio conditions -- they use it as a differentiator in the marketplace for end listeners. The article does mention the need for higher quality in production environments:

Also, there are (and always will be) reasons to use more than 16 bits in recording and production.

None of that is relevant to playback; here 24 bit audio is as useless as 192kHz sampling. The good news is that at least 24 bit depth doesn't harm fidelity. It just doesn't help, and also wastes space.

Came to recommend the same. The great thing is the book isn't built around a 'one neat trick' approach that so much financial literature takes, but instead tries to aid in building up useful mental models -- that just happen to be applicable to financial problems.

That'd also overlap with 31C3: https://events.ccc.de/congress/2014/wiki/Main_Page