zlacker

[parent] [thread] 20 comments
1. Geee+(OP)[view] [source] 2023-11-18 00:51:05
I think I see it now. Speculation following:

They achieved AGI internally, but didn't want OpenAI to have it. All the important people will move to another company, following Sam, and OpenAI is left with nothing more than a rotting GPT.

They planned all this from the start, which is why Sam didn't care about equity or long-term finances. They spent all the money in this one-shot gamble to achieve AGI, which can be reimplemented at another company. Legally it's not IP theft, because it's just code which can be memorized and rewritten.

Sam got himself fired intentionally, which gives him and his followers a plausible cover story for moving to another company and continuing the work there. I'm expecting that all researchers from OpenAI will follow Sam.

replies(6): >>woeiru+i >>26fing+q >>window+01 >>sheeps+J1 >>cthalu+id >>ilrwbw+0f
2. woeiru+i[view] [source] 2023-11-18 00:52:31
>>Geee+(OP)
This is certainly a take, but MSFT would probably sue the hell out of them if they tried to do this.
replies(1): >>xyst+31
3. 26fing+q[view] [source] 2023-11-18 00:52:58
>>Geee+(OP)
No I don’t think that’s what happened at all. Also memorizing code and rewriting it is very much IP theft.
replies(2): >>lenerd+G1 >>Geee+B3
4. window+01[view] [source] 2023-11-18 00:56:24
>>Geee+(OP)
I thought this was a joke and would end with "/s" and now I'm just left with mouth slightly agape, completely in awe.
◧◩
5. xyst+31[view] [source] [discussion] 2023-11-18 00:56:53
>>woeiru+i
US patent lawyers can’t touch China mainland.
replies(1): >>woeiru+e2
◧◩
6. lenerd+G1[view] [source] [discussion] 2023-11-18 00:59:58
>>26fing+q
Is it?

Seriously, I’m asking. Like… if you were an engineer that worked on UNIX System V at AT&T/Bell Labs and contributed code to the BSDs from memory alone, would you really be liable?

replies(3): >>sheeps+Y2 >>26fing+k3 >>astran+Z3
7. sheeps+J1[view] [source] 2023-11-18 01:00:09
>>Geee+(OP)
Interesting view - I, and many others presumably, would really like more insight into the source and nature of this speculation.

I am not dismissing the possibility, far from it. It sounds very plausible. But are there any credible reports to back it up?

replies(1): >>Geee+Z2
◧◩◪
8. woeiru+e2[view] [source] [discussion] 2023-11-18 01:03:27
>>xyst+31
Right, so he’s going to flee the US for China?!? Come on.
◧◩◪
9. sheeps+Y2[view] [source] [discussion] 2023-11-18 01:07:44
>>lenerd+G1
GPT and transformer code has been open sourced many times over by different companies. The weights and the operational model is where the IP really is. This will include the architecture for distributing training and inference at this scale. That said m, any developer and scientist worth their salt will be able to replicate it from memory - without having to copy stuff over 1:1.

So unless any of the necessary bits are patented, I highly doubt an argument against them starting a new company will hold in the courts.

Sometimes the contracts can include a cool-down period before a person can seek employment in the same industry/niche, I don’t think that will apply in Sam’s case - as he was a founder.

Also - the wanting to get himself fired intentionally argument doesn’t have any substance. What will he gain from that? If anything, him leaving on his own terms sounds like a much stronger argument. I don’t buy the getting-fired-and-having-no-choice-but-to-start-an-AGI-company argument.

An interesting twist would be if he joins Elon in his pursuit. Pure speculation, sharing it just for amusement. I don’t think they’ll ever work together. Can’t have two people calling the shots at the top. Leaves employees confused and rarely ever works. Probably not very good for their own mental health either.

replies(1): >>astran+94
◧◩
10. Geee+Z2[view] [source] [discussion] 2023-11-18 01:07:51
>>sheeps+J1
Speculation means: the forming of a theory or conjecture without firm evidence.

It's just a fun theory, which I think is plausible. It's based on my personal view of how Sam Altman operates, i.e. very smart, very calculative, makes big gambles for the "greater purpose".

replies(1): >>sheeps+l32
◧◩◪
11. 26fing+k3[view] [source] [discussion] 2023-11-18 01:09:02
>>lenerd+G1
Probably depends on what the code is and how material it is to AT&T’s business and what agreements are in place. IANAL. Youre not gonna get sued for routine stuff.
◧◩
12. Geee+B3[view] [source] [discussion] 2023-11-18 01:10:29
>>26fing+q
It's not if it's not literal. You can easily reimplement the same ML architecture which you have written before. Also, it's not really OpenAIs IP, if they kept it secret.
replies(1): >>Paul-C+uZ
◧◩◪
13. astran+Z3[view] [source] [discussion] 2023-11-18 01:12:30
>>lenerd+G1
It is not a copyright issue unless you typed out the exact code from memory. It could be a patent issue if it behaves the same way.
◧◩◪◨
14. astran+94[view] [source] [discussion] 2023-11-18 01:13:42
>>sheeps+Y2
> Sometimes the contracts can include a cool-down period before a person can seek employment in the same industry/niche, I don’t think that will apply in Sam’s case - as he was a founder.

It's very difficult to enforce anything like this in California. They can pay him to not work, but can't just require it.

replies(1): >>Paul-C+ZY
15. cthalu+id[view] [source] 2023-11-18 02:08:03
>>Geee+(OP)
> Legally it's not IP theft, because it's just code which can be memorized and rewritten.

That is not how IP law works. Even writing new code based on the IP developed at OpenAI would be IP theft.

None of this really makes sense when you consider that Ilya Sutskever, arguably the single most important person at OpenAI, appears to have been a part of removing Sam.

replies(1): >>w-ll+Vi
16. ilrwbw+0f[view] [source] 2023-11-18 02:22:08
>>Geee+(OP)
If you understand a bit of math you would know there is no AGI and there would be no AGI on the current path.
replies(1): >>Lattic+8D
◧◩
17. w-ll+Vi[view] [source] [discussion] 2023-11-18 02:57:44
>>cthalu+id
Would Ilya maybe the one to push for AGI, and Sam didn't? And the board wants the Skynet they were promissed?
◧◩
18. Lattic+8D[view] [source] [discussion] 2023-11-18 05:24:25
>>ilrwbw+0f
What "bit of math" are you referring to? Similarly, would you have said the same things one year ago about the capabilities that ChatGPT currently possesses?
◧◩◪◨⬒
19. Paul-C+ZY[view] [source] [discussion] 2023-11-18 08:44:44
>>astran+94
It's actually easier to enforce a noncompete in California on a founder or principal of a firm than it is on an employee. I don't recall the exact, legal specifics, but it has something to do with the fact that those people are in some way attached to the "goodwill" of the original business, which is something of value that the company can protect.

Someone else can probably say it better than I can, but that's how I understand it at this moment.

◧◩◪
20. Paul-C+uZ[view] [source] [discussion] 2023-11-18 08:49:56
>>Geee+B3
Even if it is literal, it may not be infringement. See "rangeCheck" in the Oracle v. Google case.

https://www.theverge.com/2017/10/19/16503076/oracle-vs-googl...

◧◩◪
21. sheeps+l32[view] [source] [discussion] 2023-11-18 16:20:13
>>Geee+Z2
I believe the comment was edited. The original comment made a mention of “seeing a lot of speculation” (paraphrasing) that piqued my curiosity.

The source of the speculation could further enhance or remove the probability of this being true. For instance, a journalist who covers OpenAI vs. a random tweeter (now X’er?) with no direct connection. It’s a loose application of Bayesian reasoning - where knowing the likelihood of one event (occupation of speculator and their connection to AI) can significantly increase the probability of the other event (the speculation).

[go to top]