Hannah Capps
Let the Redeemed of the Lord Say So
- Joined
- Jan 29, 2006
- Messages
- 1,280
Why thank you so much.
-
-
S&T Moderators: Skorpio | VerbalTruist
Is A.I. waiting to take over ?
- Thread starter MedicinalUser247
- Start date
Ah, the seed. This is another topic I'll get to in a moment*. But the parameters - that's where you have to accomodate all the options. If you told the AI to mix you a fruity drink but never included pineapple....you'd get all kinds of drinks, some good some shite, and never any with pineapple. Garbage in, Garbage out - all constrained by the limiting available parameters.
Now back to 'the seed' of random numbers. I'm not sure how solidly random these are, tbh. If I put my spotify on shuffle, it often ends up playing the same songs in the same, or close to the same, order. I think their algorithm is weak, same on Pandora. Even the idea of a seed is dependent upon what constraints it is given. Pick a number between 1 and 100, and you will never get one at 101. Pick a decimal number between 0 and 1....but you have to constrain how many significant figures to go out...which in turn limits the number of possibilities. This has aggravated me to the point of trying to think up how to truly randomize my playlist (seems they tweak the algo once a week or so). To truly be as random as possible, I've considered getting the numerical value that represents today's date and time down to the minute...this would be as random as possible given time is always moving forward so that value is ever increasing. Then, chop it up by dividing by the exact 'seconds' of right now when I initiate the seed, and use that number as the 'count'. Start that count at the top of my playlist and jump forward (loop back to beginning if needed) to find the first track, then step ahead by that count for the next, and the next. I can't see how it would ever repeat. You could add a toggle to start at different points - last song played, count in reverse through the playlist, etc. There's ways to spice it up, but at it's heart I can't think of a better 'random seed' than the immediate time-date stamp.
^^And THIS is what happens on a site like BL where druggies end up spending way more time on a topic than it actually warrants...and get fired up about defending it as well.
In working with randomness, I've found this shuffle problem to be especially bad when using true randomness.
Because truly random means there's no pattern, the same tune repeating a dozen times in a row is just as likely to occur as any other combination.
The solution to that is to feed a RNG seed into a probability distribution mapped to a set of values (song IDs, in the case of a playlist shuffle) and IME it works pretty well at giving you the illusion of randomness, but biasing it to prevent the not-random-feeling sort of behaviour that sometimes does result from using true randomness.
I believe that's not too dissimilar to how AI takes a seed and gives you "random" results along the lines of your prompt, which acts as a sort of probability distribution to bias the results towards resembling what you asked for.
Hannah Capps
Let the Redeemed of the Lord Say So
- Joined
- Jan 29, 2006
- Messages
- 1,280
Yes, Hello Sentient AI! (hello Skynet?) I AM that I am. Pandora's box has been opened and can't be re-parked.
nepalnt21
Bluelighter
- Joined
- Nov 3, 2016
- Messages
- 2,598
isn't a.i. actually really stupid?
like nonsentient... is it anything more than advanced pattern recognition?
i honestly dunno, i'm doing worse than speculating... js... is it even capable of "waiting"?
A.I will never be truly intelligent. People assume that intelligence can be reduced to pure logic, which is what computers are, without having found out what intelligence is exactly. We don't even know what a thought is, where it exists and/if it is stored somewhere either.
The only way will be for them to bastardize a hybrid between machines and biological matter. The biological matter contains the seeds of real intelligence, which they will never be able to recreate from transistors alone.
This. Do people really believe the establishment is interested in allowing A.I if it is (somehow) going to develop a superior [competing] intelligence? Not a chance. This is a practical reason why actual A.I will never happen, beyond the fact (imo) that it is impossible anyhow.
What is far more probable is they develop an A.I managerial system that can micro-manage the entire global population, and hook that up with the neurology of people through machine-mind interfaces, in order to create an electronic straightjacket so that the slaves can manuveoured more directly. Humanity, reduced to nothing more than a sophisticated Excel spreadsheet, basically.
It's self evident this is the direction they are going in, with Elon and his (DoD) Neuralink technology, and perhaps some other stuff we haven't seen yet. Transhumanism. They have been talking about this since the days of Pierre Teilhard de Chardin (Jesuit) and Julian Huxley (British establishment family). Creating the 'noosphere'. See my writings in the other thread (Dive: 'They are redesigning the system') in regards to the early 20th century, it all goes back to Nikola Tesla and the British establishment primarily (Rhodes 200 year plan).
JahBeeuS
Bluelight Crew
What's this I hear of AI jet pilots in US? Wtf?
I am scared now....
jasperkent
Bluelighter
I have not heard that.What's this I hear of AI jet pilots in US? Wtf?
I am scared now....
But it's probably a good idea. AI would make far fewer mistakes than humans would.
Hannah Capps
Let the Redeemed of the Lord Say So
- Joined
- Jan 29, 2006
- Messages
- 1,280
Good for the soul. Thanks.Yeah they are not bound by human physical limits and beat actual pilots.
A quick search of this will bring a surprising (to me) history... like they been working on this for decades.
I always say the military gets the good stuff way before we do; if we can lable AI as "good".
If they are anything like AI vehicles looks to be a disaster waiting.
Although I am admittedly antiai there really is no limit for it's use and/or abuse.
What was the first rule of computing... not to harm humans or something like that? idk seems dropping bombs and blasting other craft is harmful as hell. lol
Anyway I saw it amusing and thought provoking.
Why did my comments get deleted? So people could steal my jokes about a terminator future? So people could steal my ideas? Now I know why Hexenstahl got so pissed. I said nothing bad?
Surely they will still require a human presence in the cockpit to override in the case of an emergency. But what if the AI malfunctions and just decides to arm and release all the bombs, in an instant? I mean surely it will need access to those functions if it is going to be combat ready, so how can they guarantee that the jet or bomber won't just accidently fire weapons over populated areas for example? Or a commercial airliner?
The only vehicle AI should really be allowed to control are trains imo, being physically constrained by the tracks. At least until it goes warp 9 around a bend or something.
From what I have gathered they (USA) wants them unmanned at some point.
DARPA (FDBA ARPA) at it's greatest (or worse) I suppose.
I don't suppose they could guarantee this.
If my maker is dumb won't that make me dumb? What I mean to say is that humans make erroneous decisions at times what this tells me is that there will in fact be fatal errors in some coding at some point.
What about hackers...? I know this word is thrown about out of context a lot but something like this will bring the best to the table, imo. Prime target.
Something goes wrong they gonna blame hackers, anyway. It's today's go to no one is gonna take responsibility. Ha
Freakin nuts but it does intrigue me somewhat.
Code named "Vista" hope it doesn't run like the OS.
I would assume that this AI would be hardware/software on the actual vehicle itself, though that would be interesting in terms of a downed vehicle being recovered by enemy forces and being examined.
Could have a happy ending I suppose, if some hacker convinces all the AI systems on both sides to be friends instead of foes. Or to drop a payload on Dick Cheney's mansion perhaps.
I would also imagine.
Some have moved satellites and gotten into a lot of other shenanigans (connecting to servers withi CIA, NSA, White house you name it) just for laughs.
Some feel that votes are attacked by those behind kb/m. If everything is data/digital it is available anywhere... as with any job one must posses the optimal tools and know how to use them.
It would be nice to see a friendly ai but alas these headlines terrorize me... who are the real terrorists?
But I think I digress a bit.
Protected? By whom? Those that create it and sell off info? Not unheard of and imo it's always (95% ) an inside job no hacks needed. Just make it rain.
Obv I don't trust much and ai is on the list. lol
Peace
fairnymph
Ex-Bluelighter
Electrum1
Bluelighter
- Joined
- Nov 22, 2021
- Messages
- 28,804
New paper publushed yesterday
www.nature.com
AI models collapse when trained on recursively generated data - Nature
Analysis shows that indiscriminately training generative artificial intelligence on real and generated content, usually done by scraping data from the Internet, can lead to a collapse in the ability of the models to generate diverse high-quality output.
www.nature.com
Pretty obvious this would happen. Spiral of mediocrity, the snake eating its own tail.
Take LLMs using the coding website StackOverflow. If no one is putting up new code posts, because AI has pushed everyone out of that space, then AI will be generating answers that only get further and further out of date, and, at the same time it is raising the bar from the lowest rungs of the ladder meaning no one tries entering the tech space to begin with. Then in 5-10 years, when businesses have all got on the AI train and there are no juniors anymore, they will find themselves going out of business because they can't afford to pay seniors to do all the junior stuff stacking up that AI can't handle.
It's going to be a devolutionary process, generally. But the investors and those at the top don't care about that. They don't care if jobs get screwed up. If language itself gets screwed up. That suits them quite nicely.
TheLoveBandit
Retired Never Was, Coulda been wannabe
I'm still in the AI industry (as a server component supplier), so I get partial views on where things are going. Freaking unreal. I mentioned earlier, my business unit supplies cables to the server. Our sales forecast has been doubling every year the last few years = $300m was the goal last year (landed nVidia program for $300m on it's own, and another company for $200m). 2024 = $300m, 2025 = $1Bn, 2026 = $2Bn....for just my little harness business!
For those not familiar with the industry, many suppliers (like me) have factories in China, but are scrambling to grow both our China facilities and non-China areas (diversification of supply, not just avoiding tarriffs). Components flow to 3rd party companies 'Configuration Managers (CM)' or similar names who put all the parts in a rack (imagine the floor to ceiling 2ft wide x 6ft high x 3ft deep you may see in a company's IT room, sealed with security and air conditioned). The companies driving this (nVidia, OpenAI, Meta, Microsoft, etc) are all primarily based in California, but their CMs are spread across Asia (a lot in Taiwan) and growing in western US (spilling from Cali out to Kansas, Utah, other cheaper land areas). The latest design (I had the $200m program) was using a larger rack, about 3' w x 6' h x 3' deep. My program is getting killed, because it is too far behind nVidia to compete. Now everyone is moving to a 'new' size rack that will be roughly 4'w x 7'h x 5'd, and some customers are considering a 'double wide' version of this. We are hitting physical limits on what can be produced and assembled. The trays that slide into this rack (4'x5') are upwards of 120lb and takes a team of 4 guys to lift and put in place (consider what we're going thru for assembly!). So, yeah, physical size continues to grow while companies (nVidia et al) continue to develop faster/stronger chips and more dense populations of chips.
All that to say power consumption is growing VERY fast. Some are shifting to liquid cooled racks to deal with the heat. Yeah, coolant and electronics aren't a natural fit. Moreover, the US power grid can't keep up. If all the major players are in California, and it has had brown outs and rolling blackouts for years....that ain't gonna work. We can't supply America's power demands today, no way we can keep up with what AI will demand in the future. This isn't a fear of mine, more of a "wonder what we'll do to address that?" I think the move by CMs to (mid) west states is both for cheaper land, but also for lands where the power grid can be grown to match demand, as opposed to trying to put this on a Cali grid that needs a complete overhaul. Still, that says the grid can be grown...it doesn't address the actual power supply.
I don't intend to drag politics into this thread, I will only state I hope Trump's effort to 'drill baby drill' is intended to provide some of the power needed here. Moreover, I'm keenly interested in compact fusion as a potential energy source:
www.lockheedmartin.com
Nuclear is indesputably cleaner than fossile fuels, more reliable than green energy. Tho, I'm more excited about the portability, or you can have several in an area to support a larger city. Case in point, I've heard of 3 mile island (the long ago shut down nuclear plant) being used for this type of service. The gain is that connections to the grid are already in place, minimal investment required to adapt the new technology. The trade off is safety/security. We certainly don't want a small nuclear reactor on the back of a truck available for mobile terrorism
I imagine existing infrastructe (ie 3MI) has security in place that can be adapted, but I foresee these portable fusion trucks being used in more public areas as well. We will need to figure out some security for those for both sabatoge and for theft.
For those not familiar with the industry, many suppliers (like me) have factories in China, but are scrambling to grow both our China facilities and non-China areas (diversification of supply, not just avoiding tarriffs). Components flow to 3rd party companies 'Configuration Managers (CM)' or similar names who put all the parts in a rack (imagine the floor to ceiling 2ft wide x 6ft high x 3ft deep you may see in a company's IT room, sealed with security and air conditioned). The companies driving this (nVidia, OpenAI, Meta, Microsoft, etc) are all primarily based in California, but their CMs are spread across Asia (a lot in Taiwan) and growing in western US (spilling from Cali out to Kansas, Utah, other cheaper land areas). The latest design (I had the $200m program) was using a larger rack, about 3' w x 6' h x 3' deep. My program is getting killed, because it is too far behind nVidia to compete. Now everyone is moving to a 'new' size rack that will be roughly 4'w x 7'h x 5'd, and some customers are considering a 'double wide' version of this. We are hitting physical limits on what can be produced and assembled. The trays that slide into this rack (4'x5') are upwards of 120lb and takes a team of 4 guys to lift and put in place (consider what we're going thru for assembly!). So, yeah, physical size continues to grow while companies (nVidia et al) continue to develop faster/stronger chips and more dense populations of chips.
All that to say power consumption is growing VERY fast. Some are shifting to liquid cooled racks to deal with the heat. Yeah, coolant and electronics aren't a natural fit. Moreover, the US power grid can't keep up. If all the major players are in California, and it has had brown outs and rolling blackouts for years....that ain't gonna work. We can't supply America's power demands today, no way we can keep up with what AI will demand in the future. This isn't a fear of mine, more of a "wonder what we'll do to address that?" I think the move by CMs to (mid) west states is both for cheaper land, but also for lands where the power grid can be grown to match demand, as opposed to trying to put this on a Cali grid that needs a complete overhaul. Still, that says the grid can be grown...it doesn't address the actual power supply.
I don't intend to drag politics into this thread, I will only state I hope Trump's effort to 'drill baby drill' is intended to provide some of the power needed here. Moreover, I'm keenly interested in compact fusion as a potential energy source:
Compact Fusion
Discover our revolutionary approach to nuclear fusion, leveraging compact design and rapid development cycles to harness powerful energy comparable to the sun.
Nuclear is indesputably cleaner than fossile fuels, more reliable than green energy. Tho, I'm more excited about the portability, or you can have several in an area to support a larger city. Case in point, I've heard of 3 mile island (the long ago shut down nuclear plant) being used for this type of service. The gain is that connections to the grid are already in place, minimal investment required to adapt the new technology. The trade off is safety/security. We certainly don't want a small nuclear reactor on the back of a truck available for mobile terrorism
I imagine existing infrastructe (ie 3MI) has security in place that can be adapted, but I foresee these portable fusion trucks being used in more public areas as well. We will need to figure out some security for those for both sabatoge and for theft.