The Bradleys got eaten alive. Why do you think they're parked in the desert?
In Ukraine Bradley IFVs have been eating Russians alive. Ukrainians absolutely love them.
A goodly portion of the Abrams never even made it to the front before breaking down
The US mostly sent Abrams to get Germany to send Leopard. The things have turbine engines and opportunity costs to support them logistically are high.
No, that was propaganda, the same as the claims the Russian troops were booby trapping children's beds and the other absurd statements from Ukraine. They actually had a person in charge of writing 'atrocity propaganda', making up stories to make the Russian troops look like barbaric animals.
Earlier in the war drone and gopro videos of Russians transporting washing machines and toilets were posted to the Internet daily. There are literally hundreds if not thousands of them between telegram, twitter and reddit.
Russians don't need any help looking like barbaric animals. On a daily basis they intentionally fire dozens of long range guided weapons at civilian housing and critical infrastructure to terrorize entire populations. Official state media explicitly and repeatedly condones heinous acts of barbarism on tape for the world to see.
Drowning children, burning homes, old ladies wishing they were raped by invading Russians...
https://www.youtube.com/watch?...
Looks like cusco picked a real winner.
Hence you _cannot_ add repeaters/amplifiers in a quantum link. The original photons have to make it though all to the other side or security is lost. That should make it amply clear this is not an "Internet" technology and can at most be used for specific dedicated links.
While the underlying modalities are different the high level concept of amplifying still applies. All you really need to do is transfer quantum "information" not necessarily the original carriers of that information. Quantum teleportation does just that and has been empirically demonstrated.
The whole thing is neither "near" nor useful in any way. Ans since you need to actually encrypt conventionally after exchanging a key with quantum technology, it does not even boost security.
It is useful in some ways and does boost security. Yes the quantum channel itself still has to be authenticated classically which means you still need to guard secrets and rely on encryption algorithms. Practically you also still have to secure the classical channel by relying on classical encryption as well although I suppose you could just xor if you enjoy watching paint dry.
The advantage of the quantum scheme is stronger forward secrecy. With quantum the classical channel can be keyed in a manner that is "disconnected" from any future knowledge of secrets and access to previously recorded encrypted channels. If you consider a forward secrecy scheme like DH even though the information used to establish the channel may no longer exist it can be reconstructed in the future by brute force with a quantum computer or compromising DH itself. Once that occurs the previously recorded message can then be decrypted. With QC that same scenario isn't physically possible because there is no in-band forward secure process to attack.
The industry spent over a billion dollars lobbying over the last year and change. A 30x return on investment plus imposition of favorable regulatory regime isn't too shabby a return.
I am against both the insane spending and legislative attempts to expand copyright regimes and punish developers of tools for evil uses by others.
I believe each industry should figure out any beneficial uses of technology itself and on its own dime rather than resorting to government handouts.
AI is just a tool no different than a computer or screwdriver. AI does not have agency. Assuming technology is any more controllable or capable of being rendered safe is equal parts absurdity and insanity. It's just a bag of weights that can and will be trivially manipulated for desired effect with minimal effort.
Neither do I support any legislation that outlaws bad thing x when performed by technology y. It is both pointless and counterproductive to enumerate all instances of y. Singling out technology y for special consideration reeks of industry favoritism and corruption. If you believe bad thing should be outlawed then you should be able to justify your position without having to resort to enumerating all the ways bad thing can be accomplished.
The fact anyone sought it necessary for section 230 to exist in the first place is what I find most disturbing. These nebulous ideas of being held liable for the actions of others and the farming out of judgement and law enforcement to corporations is in my view dangerous and illogical.
Um, no.
AI presents results to queries based upon training data. There is randomness injected into various models for various reasons.
Results are not simply based on model they are based on context and that context is a product of all prior prompts and token evaluations each of which were themselves influenced by randomness. Most commonly inference time parameter temperature controls the level of randomness imposed when selecting a token.
This is what allows one to issue the exact same prompt and get seemingly infinite number of variations as a result each time the exact same prompt is issued.
The guardrails on AI apps aren't uniform in any way, model-to-model, datasets to datasets. Feedback loops exist to correct output errors.
I believe it's possible to create a basic set of guardrail standards, and make them the basis for output expectation. That's largely ignored today. While the cats are away, the mice will play. This lack of discipline is the lubricant down the slippery slope.
If you had some super human ability to judge models beyond what they themselves understand then why not just go directly to whatever technology enabled that to happen? Why fuck around with LLMs at all?
AI does what it's programmed to do
AI is trained not programmed. Nobody knows what it will do a priori. Randomness is injected for good measure just to make sure in advance outcomes are nondeterministic.
No one's built a morality guide for AI, because many want to use it to dominate, kill, and promote anti-civil results.
Could have fooled me. There are some models where I can't even ask for an opinion anymore without getting a lobotomized "As an AI.... I don't
How does one build an industry-wide moral compass into AI?
On multiple levels you can't. The technology is not smarter than the people using it making it subject to social engineering no matter what. The technology is also easy to modify by altering the model itself to impose different sets of ideology.
A common set of boundaries, the aforementioned guardrails. Treat mistakes as bugs and use feedback loops to training data to null out undesirable results. You can put a lid on AI, and we need to do this, lest the bad guys win.
Nobody can do jack diddly squat. One can spend years building some pansy ass AI that won't tell you how to swing a hammer for fear of you accidentally hitting your thumb. A few days later someone else can come along and retrain the same model to channel Goebbels while citing chapter and verse of the anarchist cookbook.
And just wait until people start closing the feedback loop allowing STM to update LTM if you thought alignment was a fools errand for static models.
This is all part of the big lie AI companies are selling. If you don't impose laws to protect our market share then Sknet. And my personal favorite
The goal is clear enough: delay AGI development for as long as possible. How exactly we should do that--given that the precursor AIs that lead to AGI are currently benefitting much of society, and AGI itself will probably also (greatly) benefit society at first--is hard to decide.
Corporations have spent a billion dollars lobbying for regulation and scaring the public with apocalyptic x-risk / nuclear weapons rhetoric. In my view this goal is misguided and counterproductive. The attempt itself is extremely likely to only serve to further aggregate power into fewer hands. The real threat is from people not things.
Never let an opportunity to pass the blame pass. This is managerial training 101 right here. "No, we didn't have a problem with our systems. It was," googles electronic interference for date, "solar flares! Yeah, man. Solar flares. Whew."
Google? GOOGLE? You don't have the BOFH excuse calendar open on your desk all set up to go? There's even several online ones if you prefer the pointy-clicky solution.
Incidentally, what was your username again?
GPS or GLONASS satellites were largely unaffected by the flare. What happened was their differential GPS system (RTK) getting screwed up
I strongly suspect that the reason has nothing to do with the solar flare and is instead caused by something like a failed software update.
Normally biggest opportunity for uncorrected time of flight error in GPS systems is accounting for ionospheric conditions directly influenced by solar storms. Errors probably got to a point where the augmentation systems got confused.
Old mail has arrived.
