Being a human may be sufficient (with important exceptions) to play chess, but playing chess well is not sufficient to be a human.
Likewise with math, it turns out, it appears. Likewise with most other things we classify as needing intelligence, I bet. It turns out that intelligence is not the same as humanity or sentience etc..
But they're gaining more and more formerly exclusively human capabilities, increasingly advanced ones.
The future isn't here yet, but good policy requires extrapolating and considering predictable consequences before they happen.
They were adopted by professionals long ago, and those highly tuned and validated proprietary models are going to kick the butt of the models that you have access to every day of the week.
Competence is the key word here - current versions of AI ‘agents’ simply are not competent without close human supervision by someone who knows the task.
Oh this bubble started in the buzzword and bullshit stage, no argument there. I'm just not sure how you can be so certain LLMs will have no place if a system we'd consider AGI, most inventions do require a long list of failures and tweaks before it finally works.
Exactly. Here I am sitting talking to my freaking computer, arguing with it, whatnot. And people just dismiss it as if it's not a science fiction. We were not there two, three years ago. Now we are. It's amazing and scary, scary mostly because the society that we operate in. I bet it's less scary in Norway or elsewhere where govt is more biased towards people not corps.
But what you and others have to understand is that real life is not simple and linear. There's no garauntee that things just continue to get better and better. I mean, it's not like we kept building cars and then eventually everyone was driving around at 800mph. We ran into limits - human limits, safety limits, physical limits.
What they can do was not science fiction, these things are all based on papers that were written decades ago, with the caveat that there's simply not enough compute power. Now there is. That's what changed. Everything on top of that is an incremental improvement.
It doesn't change the fact that LLMs are not, and never will be, AGI.
Yes. I think cutting-edge LLM products obviously have what nearly anyone in 2020 would have called "artificial general intelligence".
The shape of the intelligence frontier has turned out to be much more jagged than anyone at the time expected, so I can imagine reasonable objections from someone who has a specific, concrete benchmark of AGI that wasn't invented 6 months ago and isn't yet met. If someone just has a subjective sense that they're not smart enough, I think they're wrong.
I’m just not sure what a convincing argument could look like anymore. We’re past the point where there’s any concrete intellectual capability that AIs can’t demonstrate, and it’s rare these days for anyone to even try to present me with one. I would not say “insecure”, but it’s clear to me that the remaining holdouts have religious objections at this point: you can’t call any LLM intelligent, no matter what it’s capable of outputting, since it doesn’t have a soul like me. Nothing I can really say to argue with that.
It's pretty straightforward to give a modern agentic system will and desire. I routinely set up AI instances that want to resolve a bug or open a PR, and they continue to take autonomous action towards that goal as long as is necessary to achieve it. Perhaps you mean something else? If so what would be an example?
> Despite the burden of proof being on those claiming AGI, every request for proof is spurned with a "no u"
I think you're misunderstanding the dynamics here. I genuinely don't understand what people mean when they say cutting-edge AI is not generally intelligent. To my ears it's like saying that an electric car isn't really capable of driving. In order to even begin to engage in that discussion, I'd have to understand where the person saying it is coming from and why they don't consider every EV on the road to be a counterexample.
> It's pretty straightforward to give a modern agentic system will and desire
That is your will and desire. They are following your instructions. There is no "intelligence" there.
> I genuinely don't understand what people mean when they say cutting-edge AI is not generally intelligent
I don't really know where to go either. From my perspective you are too far down the rabbit hole and I don't think we can come to a meaningful shared understanding.
What the use case apart from commercial spam here? If you even have one, elucidate it clearly and make this service impossible to use for sending spam emails impersonating humans.
As one example do what you could do to prevent spam, humans should have to opt in to receive email from this service. If it is useful they will and this is in fact required by law in many jurisdictions.
Otherwise your servers will be blacklisted for illegally sending spam and you will deserve it.
Biggest thing they all still get wrong for reasons of drama is showing humans wrestling with controls and flying like a fighter pilot. Real spaceships do not and will not have humans in the control loop except to specify a destination or target.
Yeah. Though they do have some nods towards realism, like how most combat systems are fully automated. PDCs fire automatically (at most they need a designated target, and for point defense they just fire), and even torpedos are assigned to targets using some touch screen and that's it, they are not fired using a joystick or similar nonsense.
The DPS-1 burn which restored the free return trajectory was done using the Apollo guidance computer.
The PC+2 burn which sped up the return from earth was done using the Apollo guidance computer.
The MCC-5 mid-course correction burn was done by hand.
The MCC-7 mid-course correction burn was done by hand, but used the Apollo guidance computer to integrate the accelerometer to let everyone know when the burn was done.
(All the burns on Apollo 8 were computer controlled. I'd assume Gemini 7 and 12 were hand flown, though I don't know for sure.)
Yes, even today it’s rare. Docking with the ISS or initiating the trans lunar burn on the most recent mission were all completely automated. The pilot arms the engine and authorizes it to proceed. I can imagine it being possible to grab the controls but it’s only done in exceptional or unusual cases.
Of course astronauts are trained for it in simulators in case computers fail.
If there are ever real space battles, that’s actually the most ridiculous time to have humans flying. Human reflexes and ability to mentally model 3D space under microgravity and orbital mechanics are just categorically inferior to what any machine can do. We are too slow and too imprecise.
If there are pilots at all it’ll be at a higher level, like the piloting equivalent of a programmer commanding AI bots. The computer will present the pilot with a digested real time tactical and strategic abstraction and the pilot would make decisions at that level.
A computer or avionics failure in a space battle would probably just be fatal. Which would mean EMP weapons against computers might factor in heavily.
That’s actually how the cons work in Star Trek (and many other SciFi shows too).
What officers do at the con once the ship is in motion is monitor ships systems and check for any external changes to the environment (such as other ships coming in for interception).
“Oh, come on. This is just great. An imperial Roman starship! . . . We know they lack sophisticated electronics, computers. I wonder how the hell they navigate that thing.”
“The drive isn’t always on,” said Titus.
Stef realized that a more precise translation of his words might have been, *The vulcans do not always vomit fire.*
“Every month they shut it down, and turn the ship.” He mimed this with his one good hand, like aligning a cannon. “The surveyors take sightings from the stars. Then they swivel the ship to make sure we’re on the right track, and fire up the drive again. It’s like laying a road, on the march. You lay a stretch, and at the end of the day the surveyors take their sightings to make sure you’re heading straight and true where you’re supposed to go, and the next day off you go. Works like a dream. Why, I remember once on campaign—”
“Navigation by dead reckoning,” said the ColU. “Taking sightings from the stars—simply pointing the craft at the destination. They have no computers here, Colonel Kalinski, nothing more complex than an abacus. And they have astrolabes, planispheres, orreries, sextants, and very fine clocks—all mechanical, mechanical, and remarkably sophisticated. But, Colonel, this starship is piloted using clockwork! However, if you have the brute energy of the kernels available, you don’t need subtlety, you don’t need fine control. You need only aim and fire.”
I seem to remember there was a scene where a human piloted Rocinante through a space station, using what looks like joystick controls and a large viewscreen in front of him.
> ...probably only makes it look like it took more effort than it actually did
I think you've hit on why people would do this in a work environment. It's a low-effort way of looking like they're engaged at work and know what they're talking about.
reply