Again, plenty of folk, self included, have no issues with streaming from a datacenter and geforce now is quite successful.
Understand that the codecs make a huge difference. But the actual inputs are literally bytes per minute of data. MAYBE kilobytes if you are particularly good at Starcraft
I don’t know what to say. Plenty of folk, myself included, have had no issues with the majority or games. And stuff like geforce now is quite successful
And a big factor in trying to combat/delay that is to not frame it as “This doesn’t even work”. Because then it is literally one free trial away from being normalized for like 95% of the audience.
Can we not pretend the problem is solely performance based? People keep doing this with generative AI and it keeps resulting in “oh shit, ghibli AI is so awesome”.
Especially since… can you watch a twitch stream? Congrats, you can stream a desktop. Even back with Stadia it was very much viable to play games like AssCreed over streaming and have a very comparable experience to it being local. And stuff like Geforce Now actually work REALLY well.
The issue shouldn’t be “can you make this perform well enough I want to use it”. It should be about ownership and the implication for… everything if all “personal computers” exist solely in a data center and all documents exist solely in The Cloud and so forth. Preservation of anything becomes nigh impossible and you suddenly have to pay a monthly fee to ever see your kid’s pictures again.
Yes and no.
At its very core? “Agentic AI” is about the idea of having a bunch of different “agents” communicate with one another in a network with defined(-ish) communication pathways. This is an “agent network”. And if that sounds like microservices/task graphs/how every fucking app works then… you win the No-Prize!!
And, in that regard, it isn’t any difficult. This service has access to that database. It always has. Hell, this service might still have zero “AI” in it but count as an “agent” for marketing purposes. If the credentials are checked and passed in an appropriate and authorized way, it is as safe as it ever has been. Which… is a different depressing discussion.
The issue comes into play when you are looking at people rapidly rewriting existing infrastructure just to say they did. And doing so with generative AI that they fundamentally can’t vet (even if they wanted to). THAT is how you break things and THAT is how you introduce new CVEs.
The issue isn’t that you have this data stored in a SQL table that is accessed by that service which was pre-seeded with credentials in a secure way. The issue is that you have no rewritten both that service and the SQL server in a way that “optimized” things by removing that costly security check.
Yes. For true emergency/disaster relief, that is the baseline. I doubt most of the meshtastic repeaters will survive a real storm and you can bet people will be spamming/attacking longfast from the comfort of their homes a county or three over. And there is no good way to communicate proper regional channels ahead of time.
But not every internet outage is a disaster. I live in a region where it is not uncommon for construction crews to cut the fiber line and take out all traffic for the county… sometimes multiple times a month… And I can speak from experience that having a mesh network with locals is incredibly useful for “Yes, it is all of us. And Verizon/Tmobile/Spring is also out” as well as “If you go to the park on 5th and MLK you have line of sight to a working cell tower”. And even just “So… I got all of Frasier on my Plex if anyone wants to hang out for a few hours”.
If you whip out your emergency HAM radios (without a license) during that? You can bet ALL the narcs are gonna tattle on you because “you weren’t prepared”.
But even during the prelude to a disaster it can be an issue. We also have wildfires in the region and get a pretty big scare maybe once a decade. Last time we were in a state of “be ready to evacuate at a moment’s notice” for the better part of a week. And just a bit of gossip that “today is going to be the day” was enough to trigger panic and clog up cell service faster than you can say “9-11”. We even got an emergency push telling us that there were no planned evacuation orders for the day and to go about normal activities.
If you are someone frantically trying to figure out where the school took your kids? Yeah, you have an emergency. If you are someone who doesn’t have a strong support network trying to figure out what is even going on? The narcs are gonna whinge at you. But, like I said, it is very useful to coordinate your evac with that support network. You can plan ahead of time to try to all get hotels/campsites in the town a few hours North. Then you drive through the hell of the evac until you get a few cell towers away, pull over, and use an app to book a hotel/campsite. But if all the people with families have to go South to pick up their kids from the school drop off site? You can only communicate when you are all an hour or three away from town and… ain’t nobody going back through that traffic snarl.
Hopefully it ends up being a false alarm and you come back a week or two later to some smoke damage (that everyone TOTALLY fixes…) and not much else. But it’s the difference between a week or two where you are able to hang out with your friends and have some degree of normalcy versus a week or two isolated and worried that you are going to lose everything.
And that is where mesh networks thrive. I am not talking about “I have a repeater in my garden” (which I should get on…). Stuff like the t-deck is what is actually useful. Plug it in, turn it on, and the pseudo-blackberries mesh with each other well enough for coordinating because enough people in town are doing the exact same thing.
One thing that people trying to make Meshtastic/Meshpro/whatever work might want to try:
odds are that your town/community have a social media system that is generally used to discuss events and the like (probably facebook, sometimes reddit). Make a post on there basically providing a link to a “getting started” guide and the credentials/key for the local mesh.
And, most importantly: Schedule an event (maybe every other month or once a quarter) where everyone with a device should turn theirs on and either be near a window or stand outside. That is a great way to rapidly detect all the temporary nodes that might only exist during a “not emergency” and for people to debug all the messes because meshtastic is a cluster.
Got it. Nobody should consider the need for a license because, in an emergency, you don’t need one. But also get a license so that you can use it in non-emergencies otherwise you’ll get a fine.
Good talk.
If you are in a situation where you need help, the odds of someone (even the person you have been talking to for weeks on the radio) doing a day or two journey to MAYBE be able to reach you is pretty slim. And such long distance communication has other implications for bad actors.
And in the event of “rebuilding” some kind of community, you aren’t going to be using a handheld device at all. You’ll raid… I don’t even know what at this point (I miss Radio Shack) to install a radio on the tallest building you can find. Oh, a HAM Radio Nerd’s house. That’ll work.
Whereas if you are trying to communicat4e with others and signal for rescue? Whatever you can get from walking up a hill/mountain or climbing the stairs to said tall building with your handheld is probably about what you can expect.
Same with in stuff like hurricanes and the like. If you are in a region that is at all hospitable then the relief teams know to send helicopters/people to that area. And if you are in the kind of situation where even a few hours might mean the difference between life and death… odds are nobody is coming.
Maybe you’d understand more things if you continued to read after the first opportunity you see to spew whatever you want to?
But in a “the internet is out” situation? Or even a “please evacuate in a calm and orderly fashion” for a wildfire or a bad hurricane? That is where meshtastic (et al) shine and it is well worth convincing friends to pick up a t-deck or whatever. Excellent for the “is it out for everyone or just me?” checks. Also useful for letting people know which field can see a cell tower a county or two over for emergency communication or to even coordinate whether you are all gonna head North or South to hang out for (hopefully just) a few days.
I’ll also add on that it is useful to be able to practice and get familiar with a tool without risking a fine.
Yeah… if I am trying to reach people tens of miles away during The Apocalypse, I am already dead.
Anyone who is within range to be helpful (or… not) would generally be within signal range of a handheld.
In a true emergency? Yes, HAM is the way to go and I need to get around to buying one of those super sketchy Baofengs. In theory you can configure them to use without a license (which is also on the todo list) but it is super easy to tick into the licensed use. How much people will care will mostly depend on whether your local HAM folk are narcs. But, regardless, all bets are off in a true emergency and Baofengs are dirt cheap.
But in a “the internet is out” situation? Or even a “please evacuate in a calm and orderly fashion” for a wildfire or a bad hurricane? That is where meshtastic (et al) shine and it is well worth convincing friends to pick up a t-deck or whatever. Excellent for the “is it out for everyone or just me?” checks. Also useful for letting people know which field can see a cell tower a county or two over for emergency communication or to even coordinate whether you are all gonna head North or South to hang out for (hopefully just) a few days.
And anyone thinking of using any of that for stuff the government don’t want you to: You are an idiot and you need to learn about how insecure all of those are.
And, when discussing stuff like this, it is important to understand that “all VPNs” actually means corporate and public VPNs.
If you want to have an actual conversation then context matters. Rather than just fixating on nonsensical overly literal interpretations because you only want to be “technically correct” by attacking a strawman.
If your corporate VPN is routing ALL traffic then your IT department are idiots. And I am pretty sure said company would thank google for blocking youtube from their employees.
It very much is possible to ban “all” public and even commercial VPNs. VPN traffic tends to have very distinct characteristics in logs and it is not overly difficult for orgs to get the IP ranges allocated to each company.
What is not possible is banning all vpn traffic in the sense that a friend or family member sets up wireguard for you. But that is a drop in the bucket to the point of being functionally nonexistent.
The middle ground, of course, are pseudo-botnets of compromised computers. But those also tend to be a fairly small percentage (outside of DDOSing) and are likely getting blocked for other reasons.
deleted by creator
People have always defaulted to the… default.
Whether that is safari on a mac or internet explorer, and now edge, on windows. And if you are already going to deal with chromium’s bullshit… edge is perfectly fine.
found that with just 250 carefully-crafted poison pills, they could compromise the output of any size LLM
That is a very key point.
if you know what you are doing? Yes, you can destroy a model. In large part because so many people are using unlabeled training data.
As a bit of context/baby’s first model training:
Just the former is very susceptible to this kind of attack because… you are effectively labeling the training data without the trainers knowing. And it can be very rapidly defeated, once people know about it, by… just labeling that specific topic. So if your Is Hotdog? app is flagging a bunch of dicks? You can go in and flag maybe 10 dicks and 10 hot dogs and ten bratwurst and you’ll be good to go.
All of which gets back to: The “good” LLMs? Those are the ones companies are paying for to use for very specific use cases and training data is very heavily labeled as part of that.
For the cheap “build up word of mouth” LLMs? They don’t give a fuck and they are invariably going to be poisoned by misinformation. Just like humanity is. Hey, what can’t jet fuel melt again?
Western Europe had an “advantage” in that, for whatever reason, many of their factories were rebuilt in the 1950s or later. So a lot of the tooling was closer to Japan (who ALSO weirdly had to rebuild a lot of factories in the back half of the 20th…) and China (who were more or less industrializing during that period). But all the jokes about the electrical system in (Western) European cars being a mess is “truth in television” due to having a lot of the tooling but not the expertise.
Much of it goes back to the 60s-80s when Western factories were largely outdated and realizing that East Asian factories were rapidly outpacing them and able to offer better products for MUCH cheaper. Rather than acknowledge they had become complacent and didn’t want to train their worekrs they instead focused on “made in America” bullshit and insisting that that new vacuum was no longer repairable. And… mostly that boils down to the idea that if you have vacuum tube transistors you can replace them easily whereas you can’t replace a transistor on a single chip.
But, as we have learned in the intervening decades, you can… just replace the board. And many of those evil computers in cars actually drastically increased repairability/maintainability because you can actually tune many aspects with a computer and get VERY useful data out of the sensors.
Because the reality is that you can make an SOC device that is INCREDIBLY repairable by focusing on how you do chip layout and what modules can be repaired. And you can make a multi-board setup that is immensely unrepairable by locking down parts with effectively DRM. And… there are also times where it actually does make sense to lock down/register those parts just like there are times it actually does make sense to glue the fuck out of that assembly.
But that is nuance. And nuance is for women and The Gays™. So buy American and purchase a radio that you can repair until the day you die! And then buy a new radio next year.
Eh. Some stuff does make sense to centralize.
Like, the concept of a thin client (what these basically are “close enough” to) is a really good one. They drastically simplify security and costs for corporate environments. And, even in the before times, it might genuinely make sense to just pay for a month/hundred hours of GFN if you wanted to play the latest AAA game rather than upgrading your five year old computer that handles everything else you play perfectly.
The bigger issue being that it now increasingly makes sense to pay for years/thousands of hours of GFN because of how broken the everything is. And the vultures (like Amazon and nVidia) smell the decay.
And… I didn’t want to crap on the other person too much but I do think p2p is why so many people think this can’t work. There is a big difference between streaming from your computer over starbucks wifi and connecting to a major data center. And there are also arguments for power and ecological impact but that becomes a MUCH bigger mess full of bad actors and incomplete comparisons.