teilo

Or, just don't use JSON for config files. There are plenty of human-friendly config file options, so there is no reason to frankenstein JSON in this way.

I rebuild these things. I know the market.

Because the energy of a photon is a function of its frequency.

DX7s don't go for a high price. The market is flooded with them. If you pay more than $600, you paid too much. You can easily find one in the $400 range.

This is strangely written. NL is not a character. It is an operation. Every time he writes "NL character" he means an LF, i.e. 0x0A.

The NL operation is defined differently for different OSs. Yes, CRLF is a relic of the teletype, but in practice, it's really the Windows convention inherited from DOS. L/unix has always been LF. On old Mac operating systems it was CR alone.

So that's what I missed at the keynote.

I use Z-Wave for everything. I looked into Home Assistant, but went with a Hubitat C8 Pro hub bridged to HomeKit on my Apple TV. A Z-wave multi-relay and tilt sensor kit is under $100, and just works.

Home Assistant is more flexible, and has nicer dashboards, but it's also way more of a PITA to get everything working consistently. Hubitat, OTOH, has just worked, with very little tinkering. I had some issues early on with it running slow, but later firmwares resolved all such issues.

Why are enterprise SANs so good at dedupe, but filesystems so bad? We use HPE Nimble (yeah, they changed the name recently but I can't be bothered to remember it), and the space savings are insane for the large filesystems we work with. And there is no performance hit.

Some of this is straight up VM storage volumes for ESX virtual disks, some direct LUNs for our file servers. Our gains are upwards of 70%.

No. This is just wrong and fails to understand not just how transformers work but the conceptual mapping that results from their training.

All these arguments about whether or not LLMs think are missing the point. They do not “think” as humans think due to their intrinsically transactional nature. But calling them “fancy statistical autocomplete algorithms” is also wrong.

LLMs contain within their matrix a massively high dimensional concept map. In this coordinate space, high order vectors map the distance between abstract concepts. This is a natural result of consuming language, which by its very nature is a symbolic concept map.

The uncomfortable question becomes: Is the human brain similarly using a massively high-dimensional concept map? Can a significant part of human thought be described as a fancy autocomplete algorithm? Can a significant amount of human reasoning be mapped as a nested series of transactions?

Actual article title: Benchmarking Qualcomm's NPU on the Microsoft Surface Tablet

Because this isn't about NPUs. It's about a specific NPU, on a specific benchmark, with a specific set of libraries and frameworks. So basically, this proves nothing.

It’s an on-device RAG.

These are the same people that convinced 3rd-world mothers to feed their babies freely provided formula until their breast milk dried up, and then the formula ran out, and they could not afford more, and their babies starved to death.

That pretty much describes all the social sciences. A morass of non-reproducibility, lack of controls, cherry picked data, bias, and subjective conclusions.

Neal Stephenson predicted this in Anathem.

Sorry. That was a major brain fart. Yes. 8-bit quantization, and using 49G of RAM.