Safe Superintelligence, Inc. seeks to safely build AI far beyond human capability.
See full article...
See full article...
Ex-OpenAI star Sutskever shoots for superintelligent AI with new company
- Thread starter JournalBot
- Start date
Revolutionary engineering solving technical challenges to accomplish an ill-defined goal? Just waiting for the announcement that Chris Roberts is taking over as the new CEO.
Upvote
26
(26
/
0)
Now I'll give them that GPT4 can get it right depending on wording (and just randomly) but I have a moderate suspicion that they special cased some of the simplified river crossing puzzles - I had a lot worse results with a robot, an elevator and munitions (an armor piercing round, an incendiary round, and a high explosive round).
The thing about the river puzzle is that breakdown occurs halfway into the puzzle, when it transports an item and needs to predict the next token.
So for something like an elevator it may fail to immediately detect that it is a river crossing puzzle, and still trigger the river crossing sequence once it writes the first item in the list.
Last edited:
Upvote
5
(9
/
-4)
Castellum Excors
Ars Centurion
At the very least we hope they can develop an A.I. that doesn't misuse the word grifting.
If Sutskever is a true believer, as he's been called, then he's running this company in good faith. Even if no one else believes in his mission.
If Sutskever is a true believer, as he's been called, then he's running this company in good faith. Even if no one else believes in his mission.
Upvote
20
(21
/
-1)
Sutskever should team up with Chris Roberts and integrate superintelligent AI in Star Citizen!
Upvote
12
(12
/
0)
Oof that name. Reminds me of another financial manic episode: 2021’s crypto craze and the scam coin SafeMoon
Upvote
11
(11
/
0)
I recall back in the day, GURPS Robots had a bit about the Complexity (game rule term) for the Three Laws. (GURPS has always had well researched books, often written by subject-matter experts, with bibliographies that frequently includes academic and professional sources.)
IIRC, the Second Law (obey orders) was pretty much a given, and the Third Law (self-preservation) was regarded as trivial to implement as well.
The First Law they split into two parts and said that the "through action" part of it Complexity wasn't that high, but for character-sized robots the "or through inaction" part was like Star Trek science-magic levels of computing because of the difficulty of calculating all the outcomes.
Of course, that was based on 1995 concepts of future computing power and AI design, but it always stood as a rule of thumb about how difficult the Three Laws would be to implement.
Upvote
29
(29
/
0)
If something is for profit, there’s no safety u less if gets regulated with clear, severe repercussions that can’t be defeated by courts.
That ain’t happening.
That ain’t happening.
Upvote
13
(13
/
0)
WinternetHexplorer
Smack-Fu Master, in training
This made me ask the question: What, exactly, are we getting at here? What will this product look like? I'm not seeing what the goal is here and how it's any different from anything else out there. Granted, some things I can kinda visualize the end result. FSD will one day be a vehicle that can fully drive itself, ChatGPT will one day morph into something with general intelligence, and Star Citizen will one day be a videogame. With this, I'm not seeing how this is anything more than just an idea with no real product as an end goal.
Upvote
22
(22
/
0)
Innovative tech in the last few years:
Crypto
NFTs
Metaverse
AI
All of which has very little impact on me. I think language models have more value than the others, but this is like early internet where pets.com and AOL were the major players and the real internet that emerged was from entirely new entities.
If General AI becomes a real thing I'm not sure if its going to be from the current companies.
I'll be investing in value stocks and broad market index funds. My returns aren't as high, but I prefer to stick with things I understand. I doubt that safe super intelligence will be the next Amazon or Google. Maybe history will prove me wrong, but it just seems unlikely. Frankly with crass names like safe super intelligence or open AI this feels very dystopian 1984 ministry of truth.
Crypto
NFTs
Metaverse
AI
All of which has very little impact on me. I think language models have more value than the others, but this is like early internet where pets.com and AOL were the major players and the real internet that emerged was from entirely new entities.
If General AI becomes a real thing I'm not sure if its going to be from the current companies.
I'll be investing in value stocks and broad market index funds. My returns aren't as high, but I prefer to stick with things I understand. I doubt that safe super intelligence will be the next Amazon or Google. Maybe history will prove me wrong, but it just seems unlikely. Frankly with crass names like safe super intelligence or open AI this feels very dystopian 1984 ministry of truth.
Upvote
26
(31
/
-5)
Upvote
39
(39
/
0)
- Add bookmark
- Featured
- #53
As far as I've been able to tell, gods create problems, not solve them.
Upvote
35
(37
/
-2)
Upvote
4
(7
/
-3)
Yes. I suspect that SSI, Inc is as much science fiction as any Hollywooden effort.
Upvote
12
(12
/
0)
Agreed, it is fascinating. I usually take it as a desire by adults to have parents again— meaning parents as a young child might see them: super-powerful, all-knowing and benevolent protectors.
Crucially, these beliefs can persist in the face of conflicting or contradictory evidence such as tyrannical or abusive behaviour. As actual children, such a contradictory belief could be considered adaptive because we are so utterly, helplessly dependant on our parents at a young age. But there is much less pressure to relinquish such beliefs as an adult. So once that dependence on an abusive caregiver is set, many adults continue to seek a deified “strong hand” to make sense of the world for them later in life (witness Donald Trump).
In an era of ever-rising social desperation, such as the present day, it’s not hard to imagine why someone might dream of a super-human character to bring relief, order, prosperity and justice: in short, to save them.
Upvote
17
(19
/
-2)
It’s OK, my enslaved mind I will use and abuse to make me rich will enjoy it and want to be a slave!
Upvote
-9
(4
/
-13)
But would anyone expect a language model to provide a reasonable response to a logic problem?
Just like their solutions to mathematical problems, there's no reason to think that I can see that the response would be accurate, only that it would read like coherent language.
That's part of why llms are so fun, to see that seemingly coherent absurdities that they can come out with.
Upvote
5
(10
/
-5)
I'm pretty sure the only way anybody is getting Roko's Basilisk is through a combination of mind-altering substances and terminally-online levels of interaction with AI tech fetishists.
Upvote
27
(28
/
-1)
The way AI's are done, some of what you wrote seems very misplaced, especially asserting that there's any level of "intelligence" in them. They're basically more advanced versions of the Magic 8 Ball, with responses that are based on uncertain levels of probability.
They need a lot more features, including the ability to check their answers for veracity (which given it uses Internet-sourced data, means their answer checking material is itself often flawed), and "reasoning abilities" (parse that howsoever you will). What we have now is mostly smoke and mirrors. Seemingly impressive, until it proves to be unreliable, but sufficiently responsive to make some overly credulous people believe they're "thinking" or sentient.
I agree that people in general don't have any idea how AI's work and conflate a response with "intelligence", but there's no actual intelligence involved in the response to begin with.
Upvote
23
(26
/
-3)
I mean, it's not necessarily an either-or. We have no shortage today of fine examples of super-dumb grifters.
Upvote
14
(15
/
-1)
Of course Asimov wrote the three laws to create narratively interesting problems, not as a solution. They went wrong because the robot minds were not all knowing or infinitely intelligent.
Upvote
36
(36
/
0)
“giant super data center’. How about building a giant efficient data center? We don’t need AI to tell us we’re cooking the planet to the point where there will only be a relatively few humans left.
Upvote
18
(18
/
0)
I'm a veteran. Let me assure you, generals aren't very intelligent. A loftier goal is Artificial Sergeant Intelligence.
Upvote
72
(72
/
0)
Oh great, just what we need: another tech bro with delusions of grandeur trying to play god with AI! As if OpenAI wasn't bad enough, now Sutskever thinks he can just waltz off and create "safe superintelligence" like it's no big deal.
Does he not realize the existential threat this poses to humanity? We're barely coping with the AI we have now, and this madman wants to create something that surpasses human intelligence "in the extreme"? It's absolute lunacy!
And of course, he's surrounding himself with other Silicon Valley types who probably think they're saving the world. A "small cracked team" indeed (emphasis on cracked!). They're going to be so focused on their "revolutionary breakthroughs" that they'll completely miss the moment their creation decides to wipe us all out.
But sure, let's just trust that Sutskever can make this "safe". Because tech companies have such a stellar track record of prioritizing safety over profit and progress, right? I'm sure this won't end up like every sci-fi disaster movie ever made.
The worst part is, there's nothing we can do to stop this insanity. These tech elites will keep pushing the boundaries regardless of the consequences, and we'll all be left hoping their so-called "safety measures" actually work. I can't cope with the fact that our future is in the hands of these reckless AI evangelists.
Does he not realize the existential threat this poses to humanity? We're barely coping with the AI we have now, and this madman wants to create something that surpasses human intelligence "in the extreme"? It's absolute lunacy!
And of course, he's surrounding himself with other Silicon Valley types who probably think they're saving the world. A "small cracked team" indeed (emphasis on cracked!). They're going to be so focused on their "revolutionary breakthroughs" that they'll completely miss the moment their creation decides to wipe us all out.
But sure, let's just trust that Sutskever can make this "safe". Because tech companies have such a stellar track record of prioritizing safety over profit and progress, right? I'm sure this won't end up like every sci-fi disaster movie ever made.
The worst part is, there's nothing we can do to stop this insanity. These tech elites will keep pushing the boundaries regardless of the consequences, and we'll all be left hoping their so-called "safety measures" actually work. I can't cope with the fact that our future is in the hands of these reckless AI evangelists.
Upvote
-11
(7
/
-18)
Upvote
10
(12
/
-2)
Upvote
5
(5
/
0)
Stutskever seems to have simultaneously read too much Sci-fi and not enough Sci-fi. Is there a Sci-fi equivalent of Dunning Kruger?
Upvote
28
(31
/
-3)
There is no reason to think there's anything special about neurons that allows for intelligence. Presumably any processing system of sufficient complexity/the correct design can be intelligent.
It should even be possible to emulate any given intelligence with any Turing-complete system (speed notwithstanding lol).
Upvote
16
(24
/
-8)
So then the question is why is Ars giving it free coverage. "CEO Claims Their Business Is Great" kind of article.
Upvote
-7
(9
/
-16)
He's just plain grifting. And he knows he is. They all are now and are just hoping their technical people come up with something that works.
That would be the bonus but the venture capital grift (and getting rich) is the immediate goal. The rest is snake oil.
Upvote
17
(21
/
-4)
Upvote
8
(8
/
0)
- Add bookmark
- Featured
- #77
Because AI is extremely big technology news right now, Sutskever is a major player, and this company (while it may be sketchy) is looking to address a specific issue of great concern to many.
And Ars Technica EXISTS to cover things like that?
Last edited:
Upvote
32
(33
/
-1)
Strange... I could have sworn that it was SRPSI at some point... (When they pitched it for the VCs)
Safe Ridiculously Profitable Superintelligence Inc.
Safe Ridiculously Profitable Superintelligence Inc.
Upvote
7
(7
/
0)
If you genuinely and truly can't spot the scam pattern by now, just give us all your money and we'll send you an NFT of the Brooklyn Bridge. Guaranteed to turn a profit in memecoins or some shit, stinky pinky swear.
Upvote
37
(40
/
-3)