zlacker

[parent] [thread] 43 comments
1. mrtksn+(OP)[view] [source] 2023-11-19 22:06:42
Currently, it’s very unclear who operates under what motives. How much is it about ego? How much is it about money and how much is it due to intellectual positions? Maybe there’re are no heroes and maybe there’re no antiheroes? With the recent news about other investments and deals, the facade doesn’t seem to even resemble the OpenAI’s reality.

I can’t wait to read the autobiography of involved parties.

replies(6): >>jessen+t >>resour+G2 >>thepas+v3 >>ioulau+o4 >>zyang+t8 >>YeBanK+9m
2. jessen+t[view] [source] 2023-11-19 22:08:45
>>mrtksn+(OP)
Sam Altman by Walter Isaacson

Releasing 2037...

replies(1): >>spking+Q2
3. resour+G2[view] [source] 2023-11-19 22:18:47
>>mrtksn+(OP)
Are you sure they will be "consistently candid" in their autobiographies? :-)
◧◩
4. spking+Q2[view] [source] [discussion] 2023-11-19 22:19:41
>>jessen+t
Shouldn't it be written by ChatGPT?
replies(2): >>dboreh+N7 >>mickda+FR
5. thepas+v3[view] [source] 2023-11-19 22:23:08
>>mrtksn+(OP)
>Ego

I can absolutely empathize with Ilya here, though. As far as I know the tech making openai function is largely his life’s work. It would be extremely frustrating to have Sam be the face of it, and be given the credit for it.

Sam is clearly a very accomplished businessman and networker. Those people are super important, I wish I had a person like him on my team.

I’ve had the experience of other people tacitly taking credit for my work. Giving talks about it, receiving praise for their vision. It’s incredibly demoralizing.

I’m not necessarily saying Sam did this, since I don’t know any of these people. Just speculating on how it might feel to ge Ilya watching Sam go on a world tour meeting heads of state to talk about what is largely Ilya’s work.

replies(8): >>buggle+R3 >>kwant_+h6 >>slavet+Ga >>loboch+Ye >>danena+hf >>deevia+5p >>TheCon+Jz >>kernal+cE
◧◩
6. buggle+R3[view] [source] [discussion] 2023-11-19 22:25:03
>>thepas+v3
…and yet a very familiar story in the industry.
7. ioulau+o4[view] [source] 2023-11-19 22:27:47
>>mrtksn+(OP)
Given that nothing criminal happened, canning Sam with no chance for discussion was just overkill.

It's probably more of an intellectual / philosophical position, given that they just did not think through the real impact on the business (and thus the mission itself)

I'm inclined to assume that something stupid was done. It happens. They should resolve it, fix the rules for how the board can behave, and move on.

Despite the bungling, Ilya is probably still a good voice to have on the board. His key responsibility (super-alignment), is a key part of OpenAI's mission.

replies(3): >>cmdli+x5 >>jart+3b >>belter+wd
◧◩
8. cmdli+x5[view] [source] [discussion] 2023-11-19 22:32:52
>>ioulau+o4
While we don’t know the whole story, I don’t think Sam is innocent in this matter. It seems likely that this was a recurring disagreement, and perhaps this was simply a step too far where the board had to act. When you fire somebody, typically you don’t give them a heads up.
replies(1): >>Aeolun+ss
◧◩
9. kwant_+h6[view] [source] [discussion] 2023-11-19 22:37:16
>>thepas+v3
What makes you think it is 'his' work and not theirs? I remember when OpenAI was just a joke compared to Deepmind. The turning point (as I remember) was when they used [1] deep reinforcement learning on dota2. clearly iyla (also one of the authors) contributed, but so did many others on the team I assume?

[1]: https://arxiv.org/pdf/1912.06680.pdf

replies(2): >>Wander+ef >>mi_lk+Kf
◧◩◪
10. dboreh+N7[view] [source] [discussion] 2023-11-19 22:44:39
>>spking+Q2
ChatGPT "with Walter Isaacson".
11. zyang+t8[view] [source] 2023-11-19 22:47:54
>>mrtksn+(OP)
Musk recruited Ilya from Google to start OpenAI. Musk was ousted by Altman. Now Altman is ousted by Ilya.
◧◩
12. slavet+Ga[view] [source] [discussion] 2023-11-19 22:59:39
>>thepas+v3
I can personally understand this, and am currently struggling with it. It's hard that the world works this way.
◧◩
13. jart+3b[view] [source] [discussion] 2023-11-19 23:01:23
>>ioulau+o4
Is this the kind of alignment we can expect from superintelligence? Gossiping, surprise trials, and purges?
replies(1): >>frabcu+6d
◧◩◪
14. frabcu+6d[view] [source] [discussion] 2023-11-19 23:12:35
>>jart+3b
Worse. Which is exactly why superintelligence is scary - it'll make the humans around it go wild for power, and then it will be impossible (by definition) to predict.
replies(1): >>TeMPOr+Vf
◧◩
15. belter+wd[view] [source] [discussion] 2023-11-19 23:15:21
>>ioulau+o4
Sounds clear what he was not candid about to the Board: The promises to Microsoft...
◧◩
16. loboch+Ye[view] [source] [discussion] 2023-11-19 23:24:46
>>thepas+v3
Ideas are like children. You don't just need to give birth to them; you also need to raise them, teach them, challenge them, and show them the world.

Giving birth to an idea is a necessary condition and sets the boundaries for so much of what it can achieve. But if you're unable to raise it to become a world champion, it isn't worth anything.

I've been on the raising ideas side way more in my 20+ career in tech. I know some people became bitter and scornful of me because I pushed their ideas to become something big and received a lot of credit for that. And I try to give credit where credit is due. But often enough, when I try to share the spotlight (in front of a customer or when presenting at BoD, for example), the brilliant engineer withers under pressure or actively harms his idea by pointing out its flaws excessively. It's a delicate balance.

replies(1): >>sahila+1p
◧◩◪
17. Wander+ef[view] [source] [discussion] 2023-11-19 23:26:06
>>kwant_+h6
The robotic hand Rubiks cube manipulation using RL was no joke either!
◧◩
18. danena+hf[view] [source] [discussion] 2023-11-19 23:26:26
>>thepas+v3
I think Sam has been given credit for being a good CEO and leader, which clearly is deserved. I've never heard him take credit for technical accomplishments. Ilya has been doing plenty of talks, podcasts, etc.--if anyone's the technical face of OpenAI, it's him. There's no lack of praise or credit given to him.

"Just speculating on how it might feel to Ilya watching Sam go on a world tour meeting heads of state to talk about what is largely Ilya’s work."

The whole point of a CEO is to do this kind of stuff. If your best engineers are going on world tours, talking to politicians, and preparing for keynotes, that's a pretty terrible use of their time. Not to mention that most of them would hate doing it.

replies(1): >>jjeaff+Zj
◧◩◪
19. mi_lk+Kf[view] [source] [discussion] 2023-11-19 23:29:36
>>kwant_+h6
It for sure doesn't have sama's name on it. The point stands
◧◩◪◨
20. TeMPOr+Vf[view] [source] [discussion] 2023-11-19 23:30:48
>>frabcu+6d
Huh. I imagined many scenarios, including the more obvious and dangerous one, "AI manipulating people unaware of its existence" - but I never considered a scenario in which the AI makes its existence widely known, and perhaps presents itself as more dangerous than it is, and then it just starts slightly nudging all the people racing to take control over it.
replies(1): >>hotnfr+uG
◧◩◪
21. jjeaff+Zj[view] [source] [discussion] 2023-11-19 23:55:51
>>danena+hf
I'm a developer and have used Open AI as a beta user from before their public launch and been interested in the structure and business side of AI and had never heard of Ilya until this recent blowup. I'm just one data point, but my guess is that the vast, vast majority of the public that knows anything about AI has also never heard of Ilya.
replies(2): >>machom+iv >>jatins+OS
22. YeBanK+9m[view] [source] 2023-11-20 00:08:46
>>mrtksn+(OP)
Initially I thought it iwas about money. Now it seems to be about intellectual position: Sama wants to move fast and break things, Ilya does not. I don’t want my bank to replace customer support with LLM agent, that has access to internal APIs or LLM driving a medical decision just yet.

As an example, couple years ago Crisis Text Line decided to sell data to a for profit spin off. Their justification was that data was anonymized, which was bs for it’s unstructured text data, and that it’s not against terms of service, which users had agreed to. Mind you, these users were people in crisis maybe even on a brink of a suicide. This was highly unethical and caused a backlash. Then one of the bod members wrote a half assed “reflection” post [1]. If some core employees of CTL did a “coup” to stop this decision, because they believed it’s unethical and dangerous, wouldn’t it be justifies?

[1] http://www.zephoria.org/thoughts/archives/2022/01/31/crisis-...

replies(1): >>s3p+lV
◧◩◪
23. sahila+1p[view] [source] [discussion] 2023-11-20 00:25:08
>>loboch+Ye
This isn’t a given and not everyone’s view. Doing a thing and choosing what to do with said thing is that person’s prerogative. The specifics will matter but I don’t agree that someone else’s idea is something someone else must push and profit of if they don’t. The idea of patents also agree with this too.
replies(1): >>xwolfi+MP
◧◩
24. deevia+5p[view] [source] [discussion] 2023-11-20 00:25:46
>>thepas+v3
I was under the impression that the transformer is the tech making openai function, and that Ilya's name is not on the 2017 paper introducing the idea.
replies(1): >>numeri+mx
◧◩◪
25. Aeolun+ss[view] [source] [discussion] 2023-11-20 00:44:32
>>cmdli+x5
You don’t typically do it this quickly either.
replies(2): >>tempes+pV >>rblatz+sX
◧◩◪◨
26. machom+iv[view] [source] [discussion] 2023-11-20 01:04:33
>>jjeaff+Zj
Yes, you are only one data point. Check the views on Ilya's interviews on youtube. E.g. his interview on Lex (which he did years before Sam Altman) has 400k views, which demonstrated that he is a very well known entity in tech/AI space.
replies(1): >>thepas+vE
◧◩◪
27. numeri+mx[view] [source] [discussion] 2023-11-20 01:18:32
>>deevia+5p
I believe you're underestimating how key RLHF seems to be to getting a functioning chatbot with human-like behaviors.
replies(1): >>p1esk+FJ
◧◩
28. TheCon+Jz[view] [source] [discussion] 2023-11-20 01:35:26
>>thepas+v3
A LOT of people have put a ton of energy in to OpenAI, and a lot have put A LOT of money into it. If it was as petty as credit, then screw them all as they just don’t get it. It’s all on the shoulders of others too….
◧◩
29. kernal+cE[view] [source] [discussion] 2023-11-20 02:08:25
>>thepas+v3
Ilya doesn’t want to be known as the Steve Wozniak in this relationship while Sam is perceived as the Steve Jobs. Unless you’re technically inclined no one remembers or praises the contributions of the Woz.
replies(1): >>jacque+SS
◧◩◪◨⬒
30. thepas+vE[view] [source] [discussion] 2023-11-20 02:11:37
>>machom+iv
Lex has over 3M subs and most of his videos get way over 400k views. I think this makes the opposite point you’re trying to use it for.

Sama also went on Lex and got over 5M views. The title was: OpenAi ceo on, ChatGPT, GPT4, and the future of AI.

replies(1): >>machom+bL
◧◩◪◨⬒
31. hotnfr+uG[view] [source] [discussion] 2023-11-20 02:27:03
>>TeMPOr+Vf
Both are too complicated. All a true AI has to do to get control of everything is promise 10% annual returns and guaranteed victory in battle. Limited-time offer, sign up today.

Done.

Any actual AI takeover will be boring and largely voluntary. For certain definitions of voluntary.

replies(1): >>jacque+6T
◧◩◪◨
32. p1esk+FJ[view] [source] [discussion] 2023-11-20 02:49:24
>>numeri+mx
Ilya’s name is not on RLHF papers.
◧◩◪◨⬒⬓
33. machom+bL[view] [source] [discussion] 2023-11-20 03:00:06
>>thepas+vE
Ilya's podcast was over 3 years ago and Lex's average IT podcasts had 50-100k views. Ilya got 400k. For reference, the absolute legend Jim Keller got 600k at the same time.

So yeah, Ilya is a very known entity. No, ordinary folks don't need to know him, but if you are in IT and especially if you have anything to do with AI, then not knowing about Ilya tells more about your informational bubble than about Ilya's alleged lack of recognition.

It is akin to claiming to be into crypto on development side and not knowing the name of Vitalik Buterin.

◧◩◪◨
34. xwolfi+MP[view] [source] [discussion] 2023-11-20 03:41:23
>>sahila+1p
Patents are a compromise: you keep your prerogative, yes, but for a limited amount of time and you agree to publicly publish it so that everyone can access it. Eventually, if you do nothing with it, why would we limit humanity from benefitting from it ?

It's like imagine a guy has a nice idea to cure cancer, but plays the princess with it and refuses to industrialize it, while people are dying left and right. Surely, it becomes indefensible, and at some point, someone brave will do the right thing and implement the idea. You have a right to reap the benefit of your ideas but you have a duty not to deprive humanity of any benefit just because you thought of it first, I feel ?

◧◩◪
35. mickda+FR[view] [source] [discussion] 2023-11-20 04:04:45
>>spking+Q2
I had ChatGPT give me some proposals for screenplays.

My favorite was Rainbow MosAIc, a Rashomon style film taking place mainly from Friday to Monday. It played with all the different potential motivations and theories. It did a half decent metaphor with representing the different points of view via the different video conferencing cameras.

◧◩◪◨
36. jatins+OS[view] [source] [discussion] 2023-11-20 04:17:20
>>jjeaff+Zj
Obviously Ilya will not be as famous as Sam since Sam is doing world tours and talking to who's who of world politics. But Ilya, Karpathy, gdb all are well respected and know in dev circles.

Even the recent OpenAI profile in one of prominent publications covered Mira, Ilya and gdb in addition to Sam.

But the fundamental question is why would a researcher expect (if they do) that they will be as well known as the CEO who is the face of organisation?

◧◩◪
37. jacque+SS[view] [source] [discussion] 2023-11-20 04:17:46
>>kernal+cE
I think there is a lot to be said for being remembered as the Steve Wozniak. I'd much rather be remembered as Wozniak than as Jobs.
replies(1): >>OJFord+Hy1
◧◩◪◨⬒⬓
38. jacque+6T[view] [source] [discussion] 2023-11-20 04:20:12
>>hotnfr+uG
That's the playbook of any dictator. Hitch your horse to my wagon and we'll go places. But stray from the wagon and I'll have you shot by someone who is loyal to me. And it works. Without their henchmen little creeps wouldn't get out of the gate because they are invariably complete cowards.
◧◩
39. s3p+lV[view] [source] [discussion] 2023-11-20 04:44:30
>>YeBanK+9m
>I don’t want my bank to replace customer support with LLM agent, that has access to internal APIs

If it weren't for the mentality you are rallying against we wouldn't have ChatGPT. Google, Meta, everyone had these LLMs sitting around. OpenAI was the only company with the balls to release it to the public.

◧◩◪◨
40. tempes+pV[view] [source] [discussion] 2023-11-20 04:45:31
>>Aeolun+ss
What do you mean? Typically (always?) the firing itself is instantaneous. We have no idea how long the issues that led to the firing went on.

The communication was certainly very poor, and we don't know if the reasons were good, but I don't understand the speed complaint.

replies(1): >>Aeolun+hW1
◧◩◪◨
41. rblatz+sX[view] [source] [discussion] 2023-11-20 05:02:06
>>Aeolun+ss
If you do it properly, from the outside it seems sudden.
◧◩◪◨
42. OJFord+Hy1[view] [source] [discussion] 2023-11-20 08:37:57
>>jacque+SS
Also Wozniak is relatively known, maybe Gates/Allen works better as an analogy.
replies(1): >>jacque+9E1
◧◩◪◨⬒
43. jacque+9E1[view] [source] [discussion] 2023-11-20 09:03:54
>>OJFord+Hy1
I wouldn't want to be remembered as either Gates or Allen.

For instance:

https://www.seattletimes.com/business/paul-allen-goes-after-...

◧◩◪◨⬒
44. Aeolun+hW1[view] [source] [discussion] 2023-11-20 11:04:12
>>tempes+pV
If it’s not a sudden thing, the person being fired generally writes a ‘no bad feelings’ instead of a ‘WTF’ message.
[go to top]