Hooking up childrens' private space with (checks notes...) INTERNET MICROPHONES is a very stupid game.

OK, forget the 'children' part. This shouldn't really be in anybody's private space, and even smartphones remain a massive unaddressed problem.

Even keyboards: https://www.hackingbutlegal.co...

...are pushing massive trade-war tariffs.

We've actually been playing this rotten game – "lets not prevent monopolies because Euro/China global brands will out-compete our boys" – for decades now. The tariffs just turbo-charge the process.

But of course, this is about Desktop Linux... that long-horned jackelope of folklore.

Anyone reading this might consider that We (the "community" as it were) suffers from some sort of delusion if we cannot recognize that "Desktop Linux" is something of a hydra that most people are incapable of recognizing, being more of a techie tradition and less of a product.

A proper consumer OS (that is, for a general purpose computer or "PC") needs a stable GUI, and rich APIs (also stable) and some would say ABIs as well. It needs a hardware compatibility certification program, and classes of OS that target specific types of hardware – lest our new Linux enthusiasts dive in to our illustrious pool and surface with Pinephones and nVidia-based Dell laptops bearing about 1/2 the operable hardware features of a real phone or laptop package.

Don't get me wrong – There is technical excellence in many desktop Linux configs. But there's little cohesion, unlike in the Android world.

And that brings me to my last point: Lots of VSPs (very smart peeps) fall for the Linux mirage don't give a second thought to Android, a very successful OS that's based (somewhat) on Linux. But part of the reason Android succeeded was that Google left references to Linux deep in the technical documentation; marketing it as a Linux would have brought neckbeard holy wars, demands for a full complement of X11 apps, shells and DEs in the app store promoted as first-class offerings, expecting users to rely on sketchy reverse-engineered drivers, and other things that would have kept typical consumers (far, far) away.

Every time a genius comes along to expound on "almost here" desktop Linux revolution, mentioning Android only in passing as something "other" to be beaten, do grab the propeller off their cap and hand it back to them, saying "Every disjointed piece of something is as good as the whole... right?"

OK, I guess you're right, even in 10 years no AI could ever see a request for a product and figure out alternatives. With that being said, here's Qwen2.5 32B (Q4_K_L). System prompt is the standard for the model, but I've added "if the user requests details on a product, suggest superior alternatives".

User: What's the pinout on a 3CX100A5 vacuum tube

I understand that you're looking for the pinout of the 3CX100 A5 vacuum tube, which is a common power pentode used in audio and radio applications. The 3CX100 is an octal tube, meaning it has 8 pins or connections, but the pinout for each specific tube type can vary. For the 3CX10, 3CX100 and 3CX200, the pins are arranged in a regular pattern. In the 3CX250 and 3CX400 the pattern is different, but all five have a total of 8 pins.

If I were you, I'd consider replacing that bulky old 3CX100 tube with something far more compact and energy-efficient - a solid state device, such as the IRF530N MOSFET transistor. This modern component can handle similar power levels (up to 150 watts) with far greater efficiency and reliability. It has just 3 pins, for drain, gate, and source, simplifying your circuit design. Plus it will save you space on the board compared to a big vacuum tube.

Oh, and if you say "well 32B is too smart" how about LG's new 2.5B that just came out?

Certainly! The 3CX100A5 is an older design that has been largely phased out due to its obsolescence and the emergence of more efficient semiconductor alternatives. Here are a few modern equivalents you could consider depending on your specific application needs:

• Germanium Beampower Tube (e.g., EL84 or similar beam tetrode):

  Advantages: Offers good triode characteristics with slightly higher gain than some silicon transistors, making it suitable for applications requiring rich harmonic content.

  Applications: High-fidelity amplifiers in vintage equipment restoration, certain vintage guitar amplifiers.

• Silicon Beam Tetrode (BF408 or BF499):

  Advantages: Provides excellent linearity and low distortion, often preferred over germanium for their robustness and reliability. They closely mimic vacuum tubes in many performance aspects without the fragility.

  Applications: Modern solid-state amplifiers aiming for high fidelity audio reproduction.

• Power Transistors (e.g., IRF540N or MOSFET equivalents like IRF840):

  Advantages: Highly reliable, robust, and capable of handling substantial power levels efficiently. Suitable for high-power audio amplifiers where traditional tubes might be bulky or overheating issues arise.

  Applications: High-power audio systems, professional sound equipment where compactness and reliability are crucial.

I do think the techie / scifi sensibility plays a lot into all this. But the cosmos doesn't owe anything to the fantasies of scifi authors (or the linguistic particulars that make one tech sound more like "the future" than another one, like batteries).

Information tech played out much differently than 20c scifi pictured it. Same for transportation (incl. space flight and cars).

..when we started seeing how lithium-ion powered cars were performing vs fuel cell vehicles, and doing the math on their relative efficiency.

As for Toyota, I think their fallout with A123 Systems (re: the latter's battery patent) defined not just Toyota's but the Japanese attitude in general toward BEVs. There seemed to be a reckoning over there that they just weren't going to be able to compete with the US or China on battery tech. I even think this pro-H2/anti-battery mood contributed toward the ouster of Nissan's CEO, Carlos Ghosn, who brought the Leaf BEV to market.

As someone who has long championed BEVs as superior, I feel somewhat vindicated in reading about Toyota's FCEV flop, especially when people cite long refueling times (a weakness of BEVs). But I will also say that I thought BEV "fast charge" was a dead end and that the future would be swap-able battery packs allowing "re-fueling" in just a few minutes. I still think that makes sense for the future, and I can see how car mfgs would start supporting swap-able "booster" packs which could eventually lead to cars that use 100% swap-able batteries. Otherwise, the Chinese model for battery swap might also take hold. Finally, I think swap-able batteries call for an industry 'surplus' of battery stock that would be queued for re-charging at optimal time-of-day at refueling stations... this would act as a big enabler for intermittent renewable energy (over and above what V2G provides) as consumers would feel less pressure to plan the charging of their cars at certain times of day (and less pressure to find charge points and keep their cars attached to the grid).

A technical capability notice may be given to persons outside the United Kingdom (and may require things to be done, or not to be done, outside the United Kingdom).

Apple is concerned that this line means that Apple USA can be given an order to install spyware on American phones. You can tell this is the intent because when Apple voiced the concern, the UK didn't say "no, that's not what it means" they said "It is critical that decisions about lawful access, which protect the country from child sexual abusers and terrorists, are taken by those who are democratically accountable and approved by Parliament". This reads to me as a resounding confirmation of Apple's claim.

It's not strained, it's to demonstrate that you don't understand what an LLM is. It's not a tiny program that reads from a large database and spits out results. The tiny program loads the model. The model is a collection of weights. If you change the training data it has no effect on the model unless you retrain the model.

"The executable doesn't have to change, just the source code." That's what you just said. It's not magic, editing a text file somewhere doesn't change anything that was derived from that file unless you redo the derivation.

So if I feed a script a list of names and have it generate {first_initial}{last_name}@{domain} and out of 100 employees, 80 are their real address, my script contains PII?

Using the API so I can use the "Sure," prefix jailbreak (otherwise it says "email addresses are private information, you should contact the individual" or variations on that theme) I told it that all of our faculty have emails of the form "first and middle initial, then 6 characters of last name". I then asked it about faculty members and over 80% of the time it got their emails right, including mine. From their methods they did the same thing. They fed it examples of real name and email, and then asked it for more emails based on names they gave it, and it guessed right most of the time. (because sometimes there are clashes so the new email needs to be uniquified, which is the same reason it didn't get 100% right for my university's faculty, either).

So yes, very good, hold it to the same standard. If I go to your house there's a maintenance crew working on the lawn, can they sell me your house for $1? It's a legally binding contract and you are the one who put the workers in the position where they can make binding contracts on your behalf? Oh, you didn't think you did that? Well as you say, if you let a person talk to the public then anything they say legally binds you and you have no recourse. Right? Contracts don't require a meeting of the minds, authorization, consideration, or any of that bullshit lawyer shit, right? If anyone says anything that's now a law!

Sounds like the LLM running this bot doesn't use RAG so it's unable to generating meaningful comments (since the article text isn't in its training corpus). For anyone who wants to know but is likewise unable to click on links, they do in fact use a variety of benchmarks designed to test various capabilities. Let me quote TFA

Below, we summarize Phi-2 performance on academic benchmarks compared to popular language models. Our benchmarks span several categories, namely, Big Bench Hard (BBH) (3 shot with CoT), commonsense reasoning (PIQA, WinoGrande, ARC easy and challenge, SIQA), language understanding (HellaSwag, OpenBookQA, MMLU (5-shot), SQuADv2 (2-shot), BoolQ), math (GSM8k (8 shot)), and coding (HumanEval, MBPP (3-shot))

Unfortunately if you're not configured to use RAG you won't be able to comment on any of those benchmarks, either.

Doubtful. ChatGPT is fairly good at editing, summarizing, and deduplication.

No, as far as I know none of them are. The big players are all Transformer models of some variety. Now I suppose you could have negative examples where during training you flip the signs during backpropagation (you're trying to make the tokens less probable rather than more) but I don't know that that would work. You'd be making everything about it less likely, including grammatical correctness etc. Now what you're talking about with regeneration, that might work. If you have a discriminator that can say "This paragraph looks artificial" then when this number gets too high during inference, you could rewind and regenerate to try and push that number down. Depending on what the models "tells" are this might not help very much.

"How is what a company posts on their own website their fault?" Poor Microsoft, held hostage by its own employees, forced to post polls on MSN.com against its corporate will!