The improvements in AI are going to be exponential.
I am not sure. Look up how ChatGPT actually works. It just collects groups of words, from a copy of the “internet” dated 2021, and does clever pattern matching with them.
“Proper AI”, which attracted so much hope in the 1960s with the early pioneering work making good progress before hitting the rocks, has not really progressed. And I am old enough to have been there in the 1970s following that stuff.
This lack of progress may be partly because there is no need to emulate a human. Let’s be honest, to manufacture humans you just need a bike shed at the back of Lidl, after sunset, and those with taste and discretion will (according to Jeremy Clarkson who knows basically everything) will be doing it in the back of (today’s equivalent of) a Mk 1 Ford Escort. The entry level (pun fully intended) is really very low in that business, and no need for AI, or indeed any IQ above about 10 
So AI will steadily progress in specialised tasks like MRI image analysis, military drones and possibly robots, and such like.
When we turn over physical access control to systems we no longer understand, we’ll have a problem.
Anybody who does that is an idiot. Smart people don’t even connect their factory LAN to the internet.
AIs are already generating code we can’t understand.
That’s been the case since C++ arrived Or since I used to write my univ CS assignments in Fortran and then translated them line by line into Pascal (you need to be properly old to get that one)
We don’t even understand how ChatGPT works, really
Fortunately, “we” do, and it isn’t anything to do with “AI”
Fortunately, “we” do, and it isn’t anything to do with “AI”
Well, we do and we don’t. We know how the machinery works, but once you’ve fed the internet into it, it has translated that into a vast matrix with literally billions of numbers in it. And nobody has the faintest clue what that means. Not much different from the bike-shed variety – we know pretty well how neurons etc work, but we have no idea by what process you look at a picture (with an elephant in it) and say “elephant”.
I think you meant a Mk 1 Cortina (with thanks to Ian Dury and Billericay Dickie).
I had a love affair with Nina
in the back of my Cortina
A seasoned up hyena
could not have been more obscener
She took me to the cleaners
and other misdemeanours;
but I got right up between her
rum and her ribena.
Peter wrote:
Anybody who [turns over physical access control to systems we no longer understand] is an idiot.
Reminds me of the 1954 short story by Fredric Brown, about people who built a computer to answer the question about life, the universe and everything (in principle). The story ends:
He turned to face the machine. “Is there a God?”
The mighty voice answered without hesitation, without the clicking of a single relay.
“Yes, now there is a God.”
Sudden fear flashed on the face of Dwar Ev. He leaped to grab the switch.
A bolt of lightning from the cloudless sky struck him down and fused the switch shut
While I agree that ChatGPT is absolutely not intelligent in any way (and one of the biggest giveaways is its inability to just say “I don’t know” and instead spew out incorrect information, especially if you ask it anything mathematical) — its natural language processing capability is really good, much better than anything that came before (not just Eliza but all the various attempts at the Turing Prize). It doesn’t come out with the random non-sequiturs that Turing Prize attempt chatbots would come up with even just five years ago, and constantly lose context like those chatbots would.
It’s almost at the Star Trek TNG computer level where you could just speak to the computer in normal English and get sensible results. I think the ST:TNG computer might be fairly close, with the exception of the holodeck!
alioth wrote:
It’s almost at the Star Trek TNG computer level where you could just speak to the computer in normal English and get sensible results. I think the ST:TNG computer might be fairly close, with the exception of the holodeck!
[ YT URL fixed up – see here ]
alioth wrote:
It’s almost at the Star Trek TNG computer level where you could just speak to the computer in normal English and get sensible results. I think the ST:TNG computer might be fairly close, with the exception of the holodeck!
I think the one computer people like to compare AI with is HAL from 2001, a space odyssee. In Star Trek, the only “computer” which ever rejects human orders may be the android named Data in TNG, but none of the ships computers would do that. HAL however develops a intelligence of it’s own which in the end locks out the human it is “serving” by telling him “I am afraid I can’t let you do that”.
If we are talking Star Trek, I think the closest stuff where machines have taken over would be the Borg Collective.
Peter wrote:
> AIs are already generating code we can’t understand.That’s been the case since C++ arrived
I’ve long been saying that coding in C++ (or even in C) changes coders’ personality, logic and spoken language to such a degree they eventually fall out of the general society.
ChatGPT on on-wing Magnetic GoPro Mount
Last night while sitting on the couch I contemplated using a magnetic camera mount on the wing of our Bristell B23 to film the fuselage from the right side.
It wouldn’t work since the wings are made from Aluminium, which isn’t magnetic, but I was also thinking about the question, whether the mount I have — which has 6 quite strong magnets in its rubberized base — would be up to the wind force at Vne.
Not sure about the best way to calculate this I turned to ChatGPT, which once again flattened be with the quality of its responses.
Here is the raw and unedited transcript of my conversation with it:
As an IT professional myself, I find the ability of this engine to conduct real and actually helpful, technical discussions with me absolutely astonishing.
I need to actively seek out more opportunities to use its capacities, because they sure are quite impressive.
I am also an IT professional and I recently asked ChatGPT to help me with a problem. The answers were so incorrect, did not follow the conversation (context from previous messages) and never worked, that I would never (in its today’s form) trust it to calculate if I can put a camera on my airplane, which, if done wrong, could be an insurance claim, a dead person on the ground and/or me being dead with a broken airplane.
ChatGPT is a fun (BS) conversation chat bot and nothing more than that. It’s clueless about any useful stuff and provided answers and advices can be equally right or wrong.
