zlacker

[parent] [thread] 26 comments
1. trasht+(OP)[view] [source] 2023-11-19 21:51:52
A rough guess is that Elon will pick up Ilya, with a promise to really aim for safe AGI.
replies(5): >>apppli+p >>alecco+51 >>mcmcmc+J7 >>whyeno+pe >>optima+Oo
2. apppli+p[view] [source] 2023-11-19 21:53:48
>>trasht+(OP)
If Ilya’s problem with Sam was that he was acting out on his own and deviating from founding principles, he’s not going to have a good time under Elon.
replies(3): >>CSMast+11 >>somepe+J9 >>icelan+1c
◧◩
3. CSMast+11[view] [source] [discussion] 2023-11-19 21:56:47
>>apppli+p
From everything I can piece together the underlying problem is that Sam defunded Ilya's alignment research because they only have so much compute and they needed all of it to keep up with the demand for ChatGPT and the APIs, especially after dev day.

Ilya losing access to the GPUs he needs to do his research so that the company can service a few more customers seemed like a fundamental betrayal to him and a sign that Sam was ignoring safety in order to grow marketshare.

If Elon is able to promise him the resources he needs to do his research then I think it could work out.

replies(3): >>dchich+12 >>tw04+U5 >>dwaltr+Rf
4. alecco+51[view] [source] 2023-11-19 21:57:09
>>trasht+(OP)
I'm starting to suspect this was all orchestrated by Google. Win-win. Google has the hardware, the data, and the models. Google only lacks OpenAI's secret refining sauce. Getting back Ilya would be the best outcome for them.
◧◩◪
5. dchich+12[view] [source] [discussion] 2023-11-19 22:01:18
>>CSMast+11
More likely that the WorldCoin project is doing poorly. With crypto and NFTs going down it could be a house of cards. And in a need of an urgent injection of money.
replies(1): >>spookt+F2
◧◩◪◨
6. spookt+F2[view] [source] [discussion] 2023-11-19 22:04:24
>>dchich+12
The worldcoin thing really bugs me too. Any time there is crypto, somebody is scamming somebody else. Worldcoin is no different in this regard.
replies(1): >>capabl+06
◧◩◪
7. tw04+U5[view] [source] [discussion] 2023-11-19 22:19:51
>>CSMast+11
> If Elon is able to promise him the resources he needs to do his research then I think it could work out.

Who on earth would ever trust an Elon promise at this point? The guy literally can’t open his mouth without making a promise he can’t keep.

Unless Ilya is getting something in a bulletproof contract and is willing to spend a decade fighting for it in court, he’s an idiot doing anything with Elon.

replies(3): >>static+1a >>comfys+CM >>unsupp+1V
◧◩◪◨⬒
8. capabl+06[view] [source] [discussion] 2023-11-19 22:20:18
>>spookt+F2
Did you figure out the motives behind Bitcoin too while doing your research?
replies(1): >>jazzyj+Ei
9. mcmcmc+J7[view] [source] 2023-11-19 22:29:12
>>trasht+(OP)
Ah yes, Elon who famously keeps all his promises without needing a court order.
◧◩
10. somepe+J9[view] [source] [discussion] 2023-11-19 22:38:58
>>apppli+p
Elon Musk has been very critical of OpenAI's switch from a non-profit focused on AI safety to a for-profit competing with Tesla.

It's why he fell out and left OpenAI despite investing $100 million to start it.

I'd say he's well aligned with Ilya's position. Early on I wondered if he was an instigator of the entire board coup.

replies(2): >>Rivier+8c >>Tenoke+je
◧◩◪◨
11. static+1a[view] [source] [discussion] 2023-11-19 22:40:24
>>tw04+U5
Contract wherein Musk pays all legal fees would be pretty trustworthy
replies(1): >>eganis+ne
◧◩
12. icelan+1c[view] [source] [discussion] 2023-11-19 22:50:50
>>apppli+p
Ilya has worked for Elon in the past without incident. I doubt it would be a problem.
◧◩◪
13. Rivier+8c[view] [source] [discussion] 2023-11-19 22:51:23
>>somepe+J9
He left because they didn't allow him to be in charge.
replies(1): >>somena+tV
◧◩◪
14. Tenoke+je[view] [source] [discussion] 2023-11-19 23:02:41
>>somepe+J9
The problem is Elon's approach to alignment as presented during x.ai launch is pretty different to what Ilya says and as far as I can tell pretty naive on top of that.
◧◩◪◨⬒
15. eganis+ne[view] [source] [discussion] 2023-11-19 23:03:14
>>static+1a
https://www.cnbc.com/2023/02/24/musks-twitter-has-been-sued-...

He's pretty bad at honoring contracts too

replies(1): >>static+RY
16. whyeno+pe[view] [source] 2023-11-19 23:03:33
>>trasht+(OP)
I apologize in advance for this snarky comment, but with Elon’s track record when it comes to safety at Twitter and Tesla, I would doubt his sincerity or follow-through on AI safety.
replies(1): >>trasht+zX1
◧◩◪
17. dwaltr+Rf[view] [source] [discussion] 2023-11-19 23:11:48
>>CSMast+11
Interesting, where did you hear that?
◧◩◪◨⬒⬓
18. jazzyj+Ei[view] [source] [discussion] 2023-11-19 23:28:39
>>capabl+06
despite being invented for non-scam purposes, it turns out to be a productivity multipler for scammers more than anyone else.

and also Bitcoin might be the exception that proves the rule - every other chain or token is managed by a few insiders taking get-rich-quick marks for a ride.

replies(1): >>ocdtre+4m
◧◩◪◨⬒⬓⬔
19. ocdtre+4m[view] [source] [discussion] 2023-11-19 23:49:56
>>jazzyj+Ei
I think it'd be really funny if Bitcoin was originally supposed to also be a rug pull but Satoshi died suddenly or something, and so Bitcoin was just the scam that never managed to complete itself or something.
20. optima+Oo[view] [source] 2023-11-20 00:06:21
>>trasht+(OP)
Don't do it, Ilya.
◧◩◪◨
21. comfys+CM[view] [source] [discussion] 2023-11-20 02:49:04
>>tw04+U5
Yeah, I imagine Karpathy had his fill of Musk and left. Probably with a stiff non disparagement agreement.
◧◩◪◨
22. unsupp+1V[view] [source] [discussion] 2023-11-20 04:08:31
>>tw04+U5
It would take a rocket scientist to trust Elon Musk
◧◩◪◨
23. somena+tV[view] [source] [discussion] 2023-11-20 04:13:52
>>Rivier+8c
Musk left in 2018 after disagreements over the future of the company. In 2019 OpenAI started their 'OpenAI LP' for-profit entity. It seems entirely reasonable that the profit vs mission motive that seems to be driving this issue, is also what drove Musk to leave.
replies(2): >>totall+z31 >>Rivier+Mw1
◧◩◪◨⬒⬓
24. static+RY[view] [source] [discussion] 2023-11-20 04:48:55
>>eganis+ne
That’s my point. If/when he breaches he has to pay your fees for suing him. It’s a contingency lawyer’s dream.
◧◩◪◨⬒
25. totall+z31[view] [source] [discussion] 2023-11-20 05:22:20
>>somena+tV
Come on now, does it sound like Musk to you to leave due to the prospect of profit? Surely there was some kind of power struggle there that he couldn't win, and the mission thing was a good story to tell others.
◧◩◪◨⬒
26. Rivier+Mw1[view] [source] [discussion] 2023-11-20 08:15:36
>>somena+tV
https://www.semafor.com/article/03/24/2023/the-secret-histor...

> And Musk proposed a possible solution: He would take control of OpenAI and run it himself.

◧◩
27. trasht+zX1[view] [source] [discussion] 2023-11-20 10:51:42
>>whyeno+pe
I suppose "safety" means different things to different people. Elon seems to be of the type that cares about existential risks. One reading of him, is that he sees both Tesla, Twitter and SpaceX as tools to mitigate what he sees as existential risks.

In the case of Tesla, to accelerate the development of electric cars, in the case of Twitter, to reduce the probability of civil war and in the case of SpaceX to eventually have humanity (or our descendants) spread out enough that a single catastrophic event (like a meteor, gray goo or similar) doesn't wipe us out all at once.

His detractors obviously will question both his motives and methods, but if we imagine he's acting out of good faith (whether or not he's wrong), his approach to AI fits the pattern, including his story about why he helped with the startup of OpenAI in the first place.

From someone with an ex-risk approach to AI safety, the first concern is, to quote Ilya from the recent Alignment Workshop "As a bare minimum, let's make it so that if the tech does 'bad things', it's because of its operators, rather than due to some unexpected behavior".

In other words, for someone concerned with existential risk, even intentional "bad use" such as using AI for killer robots at a large scale in war or for a dictator to use AI to suppress a population are secondary concerns.

And it appears to me that Elon and Ilya both have this outlook, while Sam may be more concerned with shorter term social impacts.

[go to top]