zlacker

I think the allure for Altman though would be that OpenAI already has all the pieces in place.

Going off and starting his own thing would be great, but it would be at least a year to get product out, even if he had all the same players making it. And that's just to catch up to current tech

>>Workac+(OP)
Except building something the second time around is often quicker and with the current gains of hardware capabilities and interest in the space… maybe it wouldn’t be a year behind.

>>Workac+(OP)
Further wouldn’t they not be able to create GPT-x exactly as it was even though they know it?

>>Workac+(OP)
Thats ship has sailed for him if hes not on the openAI train out of town. He'd be like a third party political candidate if he tried another run at it building his own team+product from scratch. Lots of other great things to do for sure but probably not a similar supercharged role. It just wouldn't be the same - OpenAI clearly the front runner right now

>>Workac+(OP)
If only OpenAI open-sourced its models.....

>>boring+y1
What if OAI's entire research organization follows him? Surely it's one of the best teams working today.

>>Workac+(OP)
I’m really curious about how the venture investors feel about that

>>alumin+t2
It's still tough. They won't have the data used to train the model, which is an incredibly important part. There's a lot of existing competitors in this space with headstarts. There's no guarantee that the entire research organization will follow Sam even if they leave OpenAI - they're going to have a lot of offers and opportunities at other companies that have an advantage.

It's also not clear that this is a realistic scenario - Ilya is the real deal, and there's likely plenty of people that believe in him over Altman.

Of course, the company has also expanded massively under Altman in a more commercial environment, so there are probably quite a few people that believe in Altman over him.

I doubt either side ends up with the entire research organization. I think a very real possibility is both sides end up with less than half of what OpenAI had Friday morning.

>>unytti+a2
I would be surprised but not shocked if there’s some leaks in the next few weeks.

>>Workac+(OP)
Maybe much longer. The mass of infrastructure and data housed at OpenAI will be difficult to reproduce from scratch.

Especially considering OpenAI has boosted the value of the masses of data floating around the internet. Getting access to all that juicy data is going to come at a high cost for data hungry LLM manufacturers from here on out.

>>alumin+t2
Why would the entire org follow Sam instead of Ilya?

>>cthalu+j4
Isn't also because of OpenAI scraping the internet that companies got the walls up. How else is anyone able to gathering training data these days?

>>Workac+(OP)
not only that but people greatly overestimate how hard it is to replicate the success OpenAi had. you don't just build another one.

>>sudosy+B6
Sounds like wishful thinking on the part of the authors source.

If I worked there, I would keep my job and see how things shake out. If I don’t like it, then I start looking. What I don’t do is risk my well being to take sides in a war between people way richer than me.

>>j45+i
There are also a ton of ~first mover advantages you can't benefit from, be they of untapped markets for demand or the exploitation of underpriced labor, capital, or IP. If Sam started a new company he would not get as good a deal on hardware or labor, he would get much more scrutiny on his training sets, and he would have to compete against both OpenAI and its competitors.

>>financ+N3
I'm curious about hoe the messaging and zeitgeist will evolve. Ober the past few months, the sentiment I encountered most frequently is that OpenAIs lead is unsurmountable and basically has a monopoly on genAI - or even AI in general. While I disagreed with this sentiment because there's no reason to believe LLM are the final word in AI, I think the will be many more people going back on prior messaging for partisan or astroturfing reasons and saying OpenAI is nothing special.

>>comfys+N8
That makes good sense and I think all those that are not independently wealthy already except personal friends of either Sam or high level remainers are going to do something quite similar. It's just too fluid a situation to make good decisions, especially if your livelihood is at stake, better not to make decisions that can't be easily undone.

>>jacque+Hd
Given that the total comp package is $300k base + $600k profit share, I don’t think any of their livelihoods are at stake. >>36460082

You’re probably right because people usually don’t have an appetite for risk, but OpenAI is still a startup, and one does not join a startup without an appetite for risk. At least before ChatGPT made the company famous, which was recent.

I’d follow Sam and Greg. But N=1 outsider isn’t too persuasive.

>>j45+i
I agree. Anthropic and Mistral are good examples. Both companies have key people from OpenAI and they fairly quickly developed good models, but I don’t think either are thinking too hard about real AGI, but instead are trying to create useful and ethical tools.

>>sillys+wg
> I’d follow Sam and Greg.

Once the avalanche has stopped moving that's a free decision, right now it could be costly.

>>wesapi+N6
Generally speaking for a base model this isn't nearly as important as it sounds because the specifics of the data don't matter as long as there's enough of it. You may remember this from high school as the central limit theorem.

For specific things like new words and facts this does matter, but I think they're not in real trouble as long as Wikipedia stays up.

>>cthalu+j4
thing is they can team with people that probably have that data already. Say Microsoft switches teams to a hypothetical SamCo AI most of the internet has already been indexed by bing and wants to be indexed by bing as its the number 2 search engine. that mean they either have cached or access to pretty much everything SamCo could want to feed said AI. Reddit or Twitter for example would never cut bing off as it would cut off users. Microsoft could though block openai from further access to things like github linkedin.

>>plorg+59
For sure. Getting ahead and staying ahead is one of them.

I’m just not sure it would be totally starting from scratch since there is more of a playbook and know how.

>>sillys+wg
OpenAI isn’t a normal startup. It was founded as a research focused not for profit. That 300k+ base comp isn’t what I’d consider “risky” either. Career wise it never seemed risky as some of the fields top AI researchers were there from day almost one.

>>sangno+vc
I’m wondering if the same group of investors would have willingly invested in Sam’s next company (if he doesn’t stay at Microsoft)