Posts by CrackedNoggin 510 publicly visible posts • joined 14 Aug 2020
I'm amazed that Microsoft isn't setting up a limited liability company at arms length and hiring from there. The whole point of SMR's is that they have lower insurance requirements, and those will become even lower in 2025 when Price-Anderson comes up for renewal. But those lower insurance requirements wouldn't indemnify the parent company from damages above the insured amount.
This was wrong and displayed ignorance:
There must be some good (or maybe even bad) reasons Linus Torvalds has shut down the idea of incorporating a garbage collecting language for kernel use for non-time critical operations - e.g., sudo. The discussion seems incomplete without hearing those reasons.
Actually - the kernel manage the authentication and credentials. sudo is simply giving a command (using setuid techniques) the permissions for root (i.e. user id 0).
That was obvious if I had thought about it. The mistake doesn't affect my main point, however.
Conversely, natural fruit juice with 300 calories of fructose in a drink can cause the exact same sharp rises in blood sugar level that cola with 300 calories of sucrose does.
So if in response to a diabetes and obesity epidemic, you tax cola and tell people to drink natural fruit juice instead, you are killing them with lies.
You should instead say "eat whole fruit in moderation".
GC does *not* innately protect you against running off the end of a block you have hold of, or incorrectly sharing it with someone else.
At least in the GC languages JS and Python, memory allocation is integrated with GC, so can not run off the end of a block or share it at all. You can share files, and read from pipes, and in JS you can perform those operations, in parallel, and join asynchronously. But you cannot share variables. (Unusual extensions notwithstanding).
Stupid memory usage design that runs out of memory is not equivalent to the problem of out-of-bounds memory access.
Notably you can run out of memory while writing in RUST as well. Of course because in RUST it takes 10~100 times more programming time to set up memory usage, it's going to make you think at lot harder about how its being used, and therefore keep memory usage as parsimonious and flat as possible - and that will make dumb memory usage design less likely, at a programming time cost.
So if you can, with experience, design non-stupid memory usage in a GC language, the work will be finished sooner. (Yes - the program will be slower - that doesn't matter if it is fast enough.)
I stick by my point - for example in JS (or any other language based on V8) you cannot access out of bounds memory.
(You can have other problems like using too much memory, but you can do that in RUST too.)
RUST will let you turn off the memory checking - JS will not.
Obviously, writing in an interpretive GC language like JS is going to be slow. But for a program like sudo, which is ideally executed by a human, and processes very little data, it would be fast enough.
Note that JS can also call multiple system subprocesses and wait on them asynchronously.
Your reply is a sign you are not willing to even try to think out of the box. That's a cardinal sin. That doesn't mean your gut feeling is wrong, but if you can't explain it, it might as well be.
First a nitpick - Thomas Claburn says "Most modern programming languages, in one way or another, allow developers to write code in a way that's memory safe". That's not fair. Even writing in assembly "allowed" programmers to write perfectly safe code, and it's only because the programmers chose to die of old age or suffer a brain aneurysm or other personal memory failure before finishing the job that it didn't work out.
It ought to be - "Programming languages which use garbage collection don't allow developers to write code in a way that's memory unsafe".
Rust doesn't qualify because you can switch off the memory checking -
unsafe
{
// TODO: brain aneurysm
}
There must be some good (or maybe even bad) reasons Linus Torvalds has shut down the idea of incorporating a garbage collecting language for kernel use for non-time critical operations - e.g., sudo. The discussion seems incomplete without hearing those reasons.
Crack was a huge driver of US city violent crime and death in the 80's. Prostitutes looking like human skeletons turning a trick for the price of their next hit. The crack houses were run by crime militias armed to the teeth with high powered automatic weapons. The cops knew where they were but only had rifles, so it wouldn't have been a fair fight. Not to mention, civilian bystander casualties.
They are all linked to terror inducing crimes and that money filters up.
And if you're compiling a language, why compile javascript? It's fundamentally an untyped, dynamic language that's ill suited to compilation.
TypeScript (which is JavaScript with a layer of write/edit-time type checking - typically integrated within the IDE) is "typed", although there is plenty of room for improvement.
TypeScript (including write/edit-time type checking within an IDE) is by far a higher level language experience than writing in C++ or RUST,
not least because writing while relying on a garbage collector is easier (and typically slower at run time).
Interestingly though, TS/JS garbage collection can and does utilize runtime multithread under the hood for JS garbage collection, even as the JS user code is running single thread.
The law automatically gives federal copyright protection to works that were created but neither published nor registered before January 1, 1978. The duration of copyright in these works is generally computed the same way as for works created on or after January 1, 1978: life plus 70 years or 95 or 120 years, depending on the nature of authorship.
A quote from "Why Developers Are Flocking to LLaMA, Meta’s Open Source LLM", in "The New Stack":
LLaMA is much more adaptable for developers. This is potentially very disruptive to the current leaders in LLM, such as OpenAI and Google. Indeed, as revealed by a leaked internal Google memo this week, the big players are already concerned:
“Being able to personalize a language model in a few hours on consumer hardware is a big deal, particularly for aspirations that involve incorporating new and diverse knowledge in near real-time.”
... there are various techniques for fine-tuning LLMs, such as hard tuning, soft tuning, prefix tuning, and adapter methods. He explained that the adapter method is attractive because it allows training of the whole LLM, while keeping the rest of the transformer frozen — which results in smaller parameters and faster training time.
"... the smaller method, where I [Raschka@Lightning AI] only have these intermediate layers like LoRA, it takes only one to three hours instead of 18 hours on the same data set, basically. So it’s an advantage because you have smaller parameters."
This is a significant development and contribution, despite not being completely open.
However, it should be noted that the limitation of not being able to use Llama output as part of training is a significant obstacle to developing new learning techniques. Specifically, it disallows learning by "trial and error", which is of course how humans learn - by trying, feeling either the pain of error and/or the joy of success. We know this is vital, and we can see how stupid current LLMs are when they blithely make really dumb mistakes without "feeling" any "shame". To be clear - using Llama output is a prerequisite for any kind of "trial" activity.
That would even including prohibit teaching Llama by the "Socratic Method" - something anyone who took Philosophy 101 could understand the folly of.
Along the once tourist drag of Front Street were fire-gutted cars in a traffic jam. The cars once were filled with people trying to outrun the fast-moving flames in winds of up to 80 mph.
["Walking along Front Street provides vivid picture of panic to escape fast-moving Lahaina fire", Maui Now]
I'm in favor of it because of the long term quality and price of life benefits of living in a city where having a car is not is not a necessity. Driverless vehicles are not completely there yet, but this is an investment in terms of R&D. None of the companies doing it are making money at this point - they are all paying money.
I believe the real benefits will come with minivan or larger size driverless ridesharing vehicles with easy-to-cleans such as are found on public transport - there are per-vehicle hardware costs that make them driverless vehicles expensive. Raising the average occupancy from 1.5 to 7.5 would be an 80% reduction in effective hardware costs.
IMO it makes a lot more sense than $7500 subsidies to top %25 earners to buy an EV.
Only 73 such [GLE] events have been detected since the 1940s -- So, on average, a little under once a year. For a typical ISS stay of six months, about a 50/50 change of exposure to a GLE.
Lunar explorers could have been hit by up to 10 gray worth of radiation if they were unlucky enough to be subjected to a coronal mass ejection event in August 1972. -- If this is true, and assuming this single instance is for an 80 year period of measurement, sending humans into space is incredibly risky.
The US (applies equally to some other Western nations) has postponed two to three decades worth of inflation by outsourcing manufacturing largely to China, who can do it cheaper - and over time, better in certain key areas, as much due to atrophy of US manufacturing skills as to Chinese progress.
Winding that back would necessarily result in increasing inflation - as winding that back can only be done through increasing import taxes or subsidies, to jump start US investments at home or at least shift production to other countries.
When push comes to shove, it won't be AI chips that decide victory on the ground in Ukraine and/or Taiwan. It will be mostly dumb chips, basic materials, and a host of goods which can be manufactured, destroyed, and replenished continually with a powerful manufacturing base. LLM's will play no part at all.
One thing to consider is the AI will eventually play a large role in manufacturing, but that will only happen in countries where manufacturing takes place at large scale.
That's just "trade". Offshoring, the named phenomena, must include (1) ever increasing government deficits (2) loss of tax base (3) loss of manufacturing capacity and all the skills and OPPORTUNITIES that go with that (4) MBA next quarter culture and disconnection from business principals (5) shrinking GDP.
That happens when it seems possible to get away without reporting and the onus is pushed onto the cyber security director to take responsibility, and the cyber security director accepts it. The case of Uber and Joseph Sullivan, along with strong whistleblower laws, should fix that.
The greatest barriers lie in the cost of the equipment, and in hardware-level issues That's true just not for this heat perception and processing, but also good reliable quality cameras with quality lenses, lidar, and computer hardware. These are all per vehicle costs.
If you are running a vehicle that is usually carrying 10 to 15 people, you are getting a cost savings of 80% per head over a vehicle usually carrying 2 or 3 people. A mega-rideshare-taxi/bus service would be really convenient in residential areas - a personal car wouldn't even be an absolute requirement as it is today even in most parts of most US cities.
Because LLM will be or are already able to read a screen, guess passwords or make an account, upvote, comment, etc., all from 100% honest Chrome or Microsoft Edge browsers.
They can operate in normal human reaction time, but there could be 100's of processes each with an LLM+browser running from a single machine.
The way this is explained I hear it as being the victim process that executes "vzerouppper", similar to undoing "free" and therefore getting a register full of some else's data. But I can't reconcile this explanation because -
1. It seems to me it would be the aggressor process that would deliberately trigger this bug in a loop in order to peer at the register full of another process's data. Why is this view wrong?
2. Why doesn't this bug cause programs to crash frequently when in normal operation a register unexpectedly contains incorrect data?
What am I missing?
... high sugar, rancid fat junk food is to a healthy balanced diet.
I think most of the horrors the author was bombarded with on Twitter were specially selected by a generic social media algorithm which selects the most "viral" content. However, viral is not the same as healthy or high quality.
Because they are first to market they can do that. If/as more competition (hopefully) develops, they might cut prices. Although bundling as part of a Azure, Visual Studio, etc., means they will always have an advantage. No serious company can afford to develop without the privacy terms (no customer data used for training-data/development) that Business level offers.
I'll give you a thumbs up because, yes, being first to market should make "them" feel good. For the record, I think CoPilot is one of the more actually beneficial applications of LLMs that I've seen (even after filtering out the market hype, and remembering that CoPilot before LLM was already a thing).
It's terrible that it gives the impression of being a perfect answer. It's interesting, maybe even impressive, that LLM training on texts alone manages to get the first 3 digits and magnitude correct. For comparison, when calculators first came out there was a lot of worry that students would over trust the calculator, fat finger the input, and not notice the magnitude was completely wrong.
Yes. The Macsyma system which arose out of research on mathematical software in the AI group at MIT in the 60's was heralded as the harbinger of GAI being just around the corner, while simultaneously removing that kind of symbolic algebra from what anyone would consider AI. I think there are strong parallels to the current iteration of AI LLM's.
Entry was forced by forging email authentication tokens with an acquired Microsoft accounts (MSA) consumer signing key.
And how did an MSA consumer signing key get leaked? It doesn't help to revoke the one key if the leak isn't detected and plugged.
It's like picking up the quarter that fell out of the pocket with a hole in it, and putting it back in the same pocket.
At the moment, Perseverance is working to learn more about a large rock named "Máaz" – Navajo for "Mars." .... But learning a new language can be tricky, even for robots. Perserverence's computers don't understand accent marks, so scientists have had to adapt.
So, UTF-8 is actually harder than rocket science. Who knew? Personally, I don't believe Perseverance will be able to succeed until it learns to speak like a native. It's time to send an upgrade to handle UTF-8.
A small रिश्वत fee is required to lubricate the bureaucracy to access the incentives. Vedanta knows how to haggle that ritual because they do it everyday - Foxconn of course does not.
In the US it's called lobbying and I am worried that the same could happen with TSMC giving up while Intel knows the ropes.
I think have read on in past Reg comments somewhere a comparison between the UK/France govt sponsored development of the Concorde and the US Boeing sponsored development of the 747 - which both took place at approximately the same time - the former being a complete financial failure while the latter became an huge financial success lasting for decades. How has the US sunk this low?
NASA should simply not be in the business of developing commercial passenger aircraft. It's only being done by NASA because it has no commercial value so neither Boeing nor Airbus would ever take on that risk. NASA might as well take down this website now - climate dot nasa dot gov : "Global Climate Change, Vital Signs of the Planet, Understanding Our Planet to Benefit Humankind".
So our "EV car [and other green] plans" are to rape Myanmar for precious metals (see ‘The Sacrifice Zone': Myanmar bears cost of green energy, APNews), process them and manufacture in China using mostly (56%) coal powered energy (see China leans on coal amid energy security push, Reuters), all so that that $7,500 subsidy per vehicle will enable buying a bigger SUV, or a sporty Tesla that dose 0-60 mph (0~100 in international speed units) in 2 seconds?.
OK, a specific example, from Lawyers who cited fake cases hallucinated by ChatGPT must pay ", TheRegister.
the chatty AI model also lied when questioned about the veracity of its citation, saying the case "does indeed exist" and insisting the case can be found on Westlaw and LexisNexis
The LLM clearly passed the Turing Test. Therefore AGI?
For a free account at least, only 5 questions allowed per day.
As long as existing functionality is not destroyed, it should be a good bit of fun for the MDN developers, and they probably thrive on all the feedback, negative included.
It's not the tool, it is how you use it.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2023
