Chatgpt - Seriously good potential (or just some Internet fun)

I suspect Apple has an AI team too and certainly no shortage of cash. They tend to play cards close to their chest and only talk when there's a product good enough to be shown at a live event.

Last few days have been .... interesting on the corporate shenanigans route. Unless there's some real dirt (as in criminal proceedings needed) in the background we don't know about, that board needs to go. You don't fire a CEO like that. Nobody sensible will work with them in the future without some cast-iron guarantees in their contract. Be very interesting to see what deal the twitch guy is on who's come in so fast.

I suspect an exodus of talent from OpenAI to microsoft as it seems Sam Altman had a lot of personal loyalty from the staff and if the CEO can be treated like that, you've got to be at least a little concerned.
 
2 of the top 10 AI channels just commented on the situation.



I also agree on Anthropic and OpenAI merging, since Anthropic left over safety. But how can they push safety on everyone else if they lose the leading edge model?
 
Last edited:
Rockbox said:
Considering this is the 2nd major split from OpenAI after Anthropic, there are probably some governance issues. This might be their Snapchat/Vine moment. If MS has GPT at home, what do they need to give OpenAI Azure credits for? Whenever Musk takes credit for putting Ilya Sutskever in OpenAI, it almost sounds like he's his "man on the inside", acting as a spoiler. Anyway, we'll see how far they go when Google, who by most accounts has the best AI team on the planet, Meta, Amazon with Anthropic and now Microsoft are all competing for the same space with clear profit motives and boatloads of cash. Maybe a fruity company will throw them a lifeline.
Click to expand...

Well, that's part of the deal for Microsoft to get GPT4 in the first place, to provide compute to OpenAI.

The whole thing hasn't been handled too well by the board and there's still a demand by employees for Sam and Greg to be reinstated. Also, Ilya has publicly stated on twitter today that he regrets his part in it and appears to have put his name to the letter demanding they reinstate Sam and Greg.
 
This whole thing just seems wild to me. Wtf have the other 3 board members been smoking last week? Looks like MS is about to swallow their whole Company and freeze them out and put everything back in place without them. Also choosing the ex twitch CEO is hilarious. He couldn't even run that at a profit and was a complete weirdo. This is why I don't on investing in the stockmarket. I have no clue how these big decisions get made :D
 
Rockbox said:
I also agree on Anthropic and OpenAI merging, since Anthropic left over safety. But how can they push safety on everyone else if they lose the leading edge model?
Click to expand...

They still own the leading model and whatever else they were working on, if Sam and Greg and staff leave for Microsoft they don't get to take that stuff within them, they have access to what's in their heads and whatever Microsoft has the rights to in return for it's Open AI investment.

I think given the number of staff who have signed now then merging with Anthropic would make sense, if of course, something doesn't happen re: the board and Sam and Greg end up reinstated (which is still possible even now).
 
dowie said:
they don't get to take that stuff with them
Click to expand...
I meant more as in they considered themselves the arbiters of safe AI, and that won't really matter if everyone else has equal or more powerful models. They can't be the self-proclaimed world policemen of AI if they're not in the lead. They're estimated to have at most a 2 year lead on everyone, and most of their work is from open source and university research. Not to mention they don't own the compute, and if they lose so much staff that they renege on their contract with MS, they might not even get the Azure credits.

If the deceleration faction is all that's left at OpenAI, they might join the US government in order to maintain control. I already read one article calling OpenAI the most important company in existence, so we (and by that they mean the chattering classes) were entitled to see what was running inside it. But not for oil, housing or agriculture, just whatever catches the reader's attention. I expect more Connor Leahy types calling for oversight, and a no longer competitive OpenAI happy to lease themselves out as federal watchmen.
 
Rockbox said:
They're estimated to have at most a 2 year lead on everyone, and most of their work is from open source and university research
Click to expand...
OpenAI code itself presumably not open source ? doesn't OpenAI have some NDA, as many hi-tech companies apply, that they would have signed, not to reveal trade-secrets,
or at a minimum impose several years of work in another domain.
... so you can't take what is in your head directly to another company.
 
jpaul said:
doesn't OpenAI have some NDA,
Click to expand...
I was wondering this. On a severance package, sure there's probably a non-compete clause attached. There might be one in his employment contract. However, Sam's not a engineer. More a 'bringer of the right people together', since he owned y combination and is highly networked.

What I suspect is you can't copy source code, but you can tell people how your company did X. Look at Nvidia engineers joining Intel, and getting a DLSS clone out faster than AMD. Silicon Valley is so incestuous that there wouldn't be much value in hiring someone who can't use ideas learned from other companies.

Sam said most of chatgpt's value came from RLHF. He'll probably give MS advice to speedrun that part of training, and it won't be considered 'trade secrets'. He also wanted Middle Eastern funding for custom silicon. The concept of doing that isn't something you can patent.
 
Last edited:
twas a great series

Rockbox said:
Sam said most of chatgpt's value came from RLHF.
Click to expand...
interesting -so, it's reward model , so where amazon/google/apple subliminally merit their services or provide moral guidance - don't be evil - https://arize.com/blog/openai-on-rlhf/

The RTL/behavioural code used for the synthesis -> silicon would be a trade-secret , USA will soon have it's own fabs too, to keep it stateside ( Snowdon is having to live in Russia)
silicon itself can largely be obfuscated to make reverse engineering difficult, pretty sure you cant' yet just X-ray/sem it, and have to manually decap/abrade it
obfuscated like they do for secure models for hardware encryption .. although they seem to be making inroads into peoples pswd locked h/w keys with crypto currency


e: hadn't appreciated amazon use chatgpt to 'summarize' reviews - brave new world
Customers like the comfort and appearance of the pants. They mention that they are comfortable in all elements, look nice, and are a perfect fit. Customers are also happy with the weight. However, some customers are mixed on fit, quality, water resistance, and value.
 
Last edited:
Rockbox said:
I meant more as in they considered themselves the arbiters of safe AI, and that won't really matter if everyone else has equal or more powerful models. They can't be the self-proclaimed world policemen of AI if they're not in the lead. They're estimated to have at most a 2 year lead on everyone, and most of their work is from open source and university research. Not to mention they don't own the compute, and if they lose so much staff that they renege on their contract with MS, they might not even get the Azure credits.

If the deceleration faction is all that's left at OpenAI, they might join the US government in order to maintain control. I already read one article calling OpenAI the most important company in existence, so we (and by that they mean the chattering classes) were entitled to see what was running inside it. But not for oil, housing or agriculture, just whatever catches the reader's attention. I expect more Connor Leahy types calling for oversight, and a no longer competitive OpenAI happy to lease themselves out as federal watchmen.
Click to expand...

Well if the Microsoft thing goes through they'd be behind the current org by default now, by circa 9 months and the current org merging with anthropic would be in the lead (for a bit at least).

Others would have a chance to catch up too - Elon's XAI for example, google (maybe?), the other startup founded by the former deepmind guy etc..
 
You must log in or register to reply here.
Back
Top Bottom