zlacker

[parent] [thread] 28 comments
1. adam_a+(OP)[view] [source] 2023-11-19 03:05:25
Pretty incredible incompetence all around if true.

From the board for not anticipating a backlash and caving immediately... from Microsoft for investing into an endeavor that is purportedly chartered as non-profit and governed by nobodies who can sink it on a whim. And having 0 hard influence on the direction despite a large ownership stake

Why bother with a non-profit model that is surreptitiously for profit? The whole structure of OpenAI is largely a facade at this point.

Just form a new for profit company and be done with it. Altman's direction for profit is fine, but shouldn't have been pursued under the loose premise of a non profit.

While OpenAI leads currently, there are so many competitors that are within striking distance without the drama. Why keep the baggage?

It's pretty clear that the best engineering will decide the winners, not the popularity of the CEO. OpenAI has first mover advantage, and perhaps better talent, but not by an order of magnitude. There is no special sauce here.

Altman may be charismatic and well connected, but the hero worship put forward on here is really sad and misplaced.

replies(4): >>x86x87+l3 >>lazyst+c4 >>rurban+ct >>jstumm+RJ
2. x86x87+l3[view] [source] 2023-11-19 03:31:15
>>adam_a+(OP)
my question is: why not both? why not pursue the profit and use that to fuel the research into AGI. seems like a best of both worlds.
replies(1): >>cthalu+h4
3. lazyst+c4[view] [source] 2023-11-19 03:38:04
>>adam_a+(OP)
look at the backgrounds of those board members... cant find any evidence that any of them have experience with corporate politics. theyre in way over their heads.
replies(2): >>robswc+b5 >>cowl+FJ
◧◩
4. cthalu+h4[view] [source] [discussion] 2023-11-19 03:38:40
>>x86x87+l3
That's the intent of the arrangement, but there's also limits - when that pursuit of profit begins to interfere with the charter of the non-profit, you end up in this situation.

https://openai.com/charter

> OpenAI’s mission is to ensure that artificial general intelligence (AGI)—by which we mean highly autonomous systems that outperform humans at most economically valuable work—benefits all of humanity.

My interpretation of events is the board believes that Altman's actions have worked against the interest of building an AGI that benefits all of humanity - concentrating access to the AI to businesses could be the issue, or the focus on commercialization of the existing LLMs and chatbot stuff causing conflict with assigning resources to AGI r&d, etc.

Of course no one knows for sure except the people directly involved here.

replies(1): >>dabock+Ze
◧◩
5. robswc+b5[view] [source] [discussion] 2023-11-19 03:43:20
>>lazyst+c4
It is also crazy that the "winning move" was to just do nothing and look like a genius and coast off that for the rest of their lives. Who in their right mind would consider them for a board position now.
replies(2): >>cthalu+U6 >>hilux+zH
◧◩◪
6. cthalu+U6[view] [source] [discussion] 2023-11-19 03:53:58
>>robswc+b5
This is assuming motivations similar to a board for a for-profit company, which the OpenAI board is not.

Insisting, no matter how painful, that the organization stays true to the charter could be considered a desirable trait for the board of a non-profit.

replies(1): >>robswc+j8
◧◩◪◨
7. robswc+j8[view] [source] [discussion] 2023-11-19 04:02:42
>>cthalu+U6
Fair. I don't know why they wouldn't just come out and say that though, if that were the case. It would be seen as admirable, instead of snake-ish.

Instead of "Sam has been lying to us" it could have been "Sam had diverged too far from the original goal, when he did X."

replies(2): >>cthalu+ec >>cowl+9K
◧◩◪◨⬒
8. cthalu+ec[view] [source] [discussion] 2023-11-19 04:35:26
>>robswc+j8
It's hard to say. Lots of things don't really make sense based on the information we have.

They could have meant that Sam had 'not been candid' about his alignment with commercial interests vs. the charter.

◧◩◪
9. dabock+Ze[view] [source] [discussion] 2023-11-19 04:57:07
>>cthalu+h4
> Of course no one knows for sure except the people directly involved here.

The IRS will know soon enough if they were indeed non-profit.

replies(1): >>cthalu+zf
◧◩◪◨
10. cthalu+zf[view] [source] [discussion] 2023-11-19 05:01:26
>>dabock+Ze
I was not implying they were not a non-profit. I am saying that we do not know the exact reason why the board fired Altman.
11. rurban+ct[view] [source] 2023-11-19 07:23:39
>>adam_a+(OP)
> It's pretty clear that the best engineering will decide the winners, not the popularity of the CEO.

This is ML, not Software engineering. Money wins, not engineering. Same as it with Google, which won because they invested massively into edge nodes, winning the ping race (fastest results), not the best results.

Ilja can follow Google's Bard by holding it back until they have countermodels trained to remove conflicts ("safety"), but this will not win them any compute contracts, nor keep them the existing GPU hours. It's only mass, not smarts. Ilja lost this one.

replies(3): >>SeanLu+BE >>rvba+FX >>foooor+dg1
◧◩
12. SeanLu+BE[view] [source] [discussion] 2023-11-19 09:17:15
>>rurban+ct
> Same as it with Google, which won because they invested massively into edge nodes, winning the ping race (fastest results), not the best results.

What in the world are you talking about? Internet search? I remember Inktomi. Basch's excuses otherwise, Google won because PageRank produced so much better results it wasn't even close.

replies(1): >>philis+t91
◧◩◪
13. hilux+zH[view] [source] [discussion] 2023-11-19 09:43:55
>>robswc+b5
A strange game. The only winning move is not to play. How about a nice game of chess?
◧◩
14. cowl+FJ[view] [source] [discussion] 2023-11-19 10:02:33
>>lazyst+c4
that's because it was never supposed to be a Corporate. It was a non-profit dedicated to AI research in the benefit of All. This is also why all this happened, they trying to stay true to the mission and not turn into a corporate.
replies(2): >>DebtDe+nT >>jkaplo+U61
15. jstumm+RJ[view] [source] 2023-11-19 10:03:42
>>adam_a+(OP)
> While OpenAI leads currently, there are so many competitors that are within striking distance without the drama.

It's hard to put into words, that do not seem contradictory: GPT-4 is barely good enough to provide tremendous value. For what I need, no other model is passing that bar, which makes them not slightly worse but entirely unusable. But again, it's not that GPT-4 is great, and I would most certainly go to whatever is better at the current price metrics in a heartbeat.

replies(1): >>skohan+ph1
◧◩◪◨⬒
16. cowl+9K[view] [source] [discussion] 2023-11-19 10:05:45
>>robswc+j8
that is what the press release says. they didn't go into specifics but it is clear that the conflict is in Comercialisation vs original purpose
replies(1): >>robswc+5w1
◧◩◪
17. DebtDe+nT[view] [source] [discussion] 2023-11-19 11:39:08
>>cowl+FJ
In which case you could say the three non-employee members of the board have no background in AI. Two of them have no real background in tech at all. One seems to have no background in anything other than being married to a famous actor.

If Sam returns, those three have to go. He should offer Ilya the same deal Ilya offered Greg - you can stay with the company but you have to step down from the board.

◧◩
18. rvba+FX[view] [source] [discussion] 2023-11-19 12:15:20
>>rurban+ct
When google came out it had the best algothitm backed by good hardware (as far as I understand often off the shelf hardware - anyway the page simply "just worked"). Difference between google and competitors was like night and day when it came out. It gained marker share very quickly because once you started using it - you didnt have any incentive to go back.

Now google search has a lot of problems, much better competition. But seriously you probably dont understand how it was years ago.

Also I thought that in ML still the best algorhitms win, since all the big companies have money. If someone came and developed a "pagerank-equivalent" for AI that is better than the current algs, customerd would switch quickly since there is no loyalty.

On a side note: Microsoft is playing the game very smart by adding AI to their products what makes you stick to them.

replies(1): >>rurban+Zj1
◧◩◪
19. jkaplo+U61[view] [source] [discussion] 2023-11-19 13:37:37
>>cowl+FJ
They don’t have experience with non-profit leadership either, do they? They have some experience leading for-profits, such as the Quora CEO, but not non-profits.
◧◩◪
20. philis+t91[view] [source] [discussion] 2023-11-19 14:00:52
>>SeanLu+BE
The faster results came after they had already won the race for best search results. Initially, Google wasn't faster than the competition in returning a full page. I vividly remeber the joy of patiently waiting 2-3 seconds for an answer, and jolting up every time Google Search came back with exactly what I wanted.
◧◩
21. foooor+dg1[view] [source] [discussion] 2023-11-19 14:47:57
>>rurban+ct
> This is ML, not Software engineering. Money wins, not engineering. Same as it with Google, which won because they invested massively into edge nodes, winning the ping race (fastest results), not the best results.

This is an absurd retcon. Google won because they had the best search. Ask Jeeves and AltaVista and Yahoo had poor search results.

Now Google produces garbage, but not in 2004.

◧◩
22. skohan+ph1[view] [source] [discussion] 2023-11-19 14:55:10
>>jstumm+RJ
What is your use-case? I have not worked with them extensively, but both PALM and LLAMA seem as good as GPT-4 for most tasks I have thrown at them
replies(1): >>barfin+tp4
◧◩◪
23. rurban+Zj1[view] [source] [discussion] 2023-11-19 15:11:23
>>rvba+FX
Oh, the pagerank myth.

Google won against initially Alta Vista, because they had so much money to buy themselves into each countries interxion to produce faster results. With servers and cheap disks.

The pagerank and more bots approach kept them in front afterwards, until a few years ago when search went downhill due to SEO hacks in this monoculture.

replies(2): >>rvba+WC1 >>int_19+I12
◧◩◪◨⬒⬓
24. robswc+5w1[view] [source] [discussion] 2023-11-19 16:22:03
>>cowl+9K
>that is what the press release says.

In the initial press release, they said Sam was a liar. Doing this without offering a hint of an example or actual specifics gave Sam the clear "win" in the court of public opinion.

IF they would have said "it is clear Sam and the board will never see eye to eye on alignment, etc. etc" they probably could have made it 50/50 or even favored.

◧◩◪◨
25. rvba+WC1[view] [source] [discussion] 2023-11-19 16:51:43
>>rurban+Zj1
This is anegdotical evidence, but I was there when Google came out and it was simply much better than the competition. I learned one day about this new websitr - and it was so much better than the other alternatives that I never went back. Same with gmail, trying to get that invite for that sweet 1GB mailbox when the ones from your country offered only 20MB and sent you 10 spammy ads per day, every day.

As an anegdote: before google I was asked to show the internet to my grandmother. So I asked her what she wants to search for. She asked me about some author, let's say William Shakespeare - guess what did the other search engine find for me and my grandma: porn...

replies(1): >>xLaszl+FN1
◧◩◪◨⬒
26. xLaszl+FN1[view] [source] [discussion] 2023-11-19 17:36:52
>>rvba+WC1
I don't remember response speed mattered until at least ten years after Google's start.

Certainly not when they won.

They were better. Basic PageRank was better than anything else. And once they figured out advertisement, they kept making it better to seal their dominance.

◧◩◪◨
27. int_19+I12[view] [source] [discussion] 2023-11-19 18:36:00
>>rurban+Zj1
Google gave better results. Few people cared about faster servers at the time, not when most of the world was still on dialup or ADSL.
◧◩◪
28. barfin+tp4[view] [source] [discussion] 2023-11-20 09:12:58
>>skohan+ph1
I’ve used all 3 a lot. Gpt 4 is definitely better. That being said if I was to rank a close second it would be Claude 2, which I think is really good
replies(1): >>skohan+Eu6
◧◩◪◨
29. skohan+Eu6[view] [source] [discussion] 2023-11-20 19:31:42
>>barfin+tp4
But would you say the others besides GPT-4 are unsuitable? That's the claim I find surprising
[go to top]