OpenAI's Board Approached Anthropic CEO About Top Job and Merger 30
According to The Information (paywalled), OpenAI's board of directors approached rival Anthropic's CEO about replacing Sam Altman and potentially merging the two AI startups. Anthropic CEO Dario Amodei declined on both fronts. Reuters reports: The news, reported earlier by The Information on Monday, follows various reported calls to find Altman's successor days after OpenAI's board ousted him. [...] The co-founders of Anthropic, who were also executives at OpenAI until 2020, had broken from their employer over disagreements regarding how to ensure AI's safe development and governance. Anthropic has won investments from Alphabet's Google and Amazon.com. Its Claude AI models have vied for prominence with OpenAI's GPT series.
So, basically all sellouts (Score:5, Insightful)
OpenAI sold out to Microsoft, former OpenAI dudes sold out to Google and Amazon... Great bunch.
I know you can't do AI without the kind of resources the aforementioned monopolies can line up. I know... But it really feels like the future belongs only to them and it's not a nice feeling.
Re:So, basically all sellouts (Score:5, Funny)
Right? I'm looking for the "good guys" here, defined by me as "wanting the long term best for humanity and finding ways to stifle or discourage the worst uses of AI".
And in the past 96 hours, I have no idea who to root for.
Re:So, basically all sellouts (Score:4, Funny)
https://github.com/SciSharp/LLamaSharp [github.com]
Re: (Score:2)
LLaMA's license is kind of insidious and clever. Basically:
* Hey, here's a free model!
* You can even use it for commercial use!** (*tiny* caveat in a bit)
* You can change it and redistribute it, so long as your changes are also LLaMA-licensed (hey, that's open-sourcey, right?)
That's the sort of thing that will tend to make a model get widely used if it's good, as everyone builds off each other's. But then:
* If you generate outputs from it, they can only be used to train other LLaMA models.
. .... which the
Re: (Score:2)
Are you able to use Falcon 40B with a reasonable GPU such as a 4090? Or do you need a A100 to play?
Re: (Score:2)
Yes, you need to use a quantized GGUF. In my experience, Q3_K_M will fit with a high batch size (2048) and the full number of tokens (2048). You might be able to do a Q4_K_S or possibly even more with low batch sizes and reduced token counts. I find the degradation vs. non-quantized models to be minimal. IMHO, there's no reason to NOT run models quantized at least somewhat. Quantization even makes inference faster.
(Sorry for the delay!)
Re: So, basically all sellouts (Score:2)
In case you're not sure how to read that:
Q#: how many bits are used on the majority of the weights
S, M, L: small, medium, or large: whether the other weights use 0, 1, or 2 extra bits relative to the Q number.
Re: (Score:2)
Re: (Score:3)
Seems to me like they think they have a gem and are trying to profit from it as much as possible. Maybe concentrate on the product instead of all those latest tergiversations. I mean, this must be the 100th article published on Slashdot about said tergiversations. CEO goes, CEO is coming back, new CEO, etc. etc. etc. Who cares? CEO are mostly the image the companies want you to have of them.
Re: (Score:2)
Re:So, basically all sellouts (Score:5, Insightful)
The problem now is that 500 out of 750 employees are ready to leave.
More employees have added their names to the list.
It's now up to 700 of 750.
They all have job offers from Microsoft.
Re: (Score:1)
Knowing the output from their systems, it's probably 780 of 750, and they all already work for Microsoft, and Google already.
Re: So, basically all sellouts (Score:2)
According to ChatGPT itâ(TM)s now 953 out of 750 total employees, which would make it about 28% of the total OpenAI workforce, or about 79% of the world population
Re: (Score:2)
Re: (Score:2)
They are hiring the opponents of the board, not the board itself. LOL
If the board sees 700 out of 750 employees as their opponents, then perhaps the problem is the board and not the employees.
Re: (Score:2)
> If the board sees 700 out of 750 employees as their opponents, then perhaps the problem is the board and not the employees.
But it's 500.
Twitter fired more than 2/3 of its staff successfully.
Many say it's better off because SF tech companies are full of pompous dbags.
And there are rumors of Chinese Intelligence data ties, military contracts, and other deception kept from the nonprofit's Board because the lucrative ventures were violative of the mission.
Would 2/3 of employees enrich themselves at the cos
Re: (Score:2)
Re: (Score:2)
They are our best and brightest AI experts, ready to be bought by the highest bidder.
Re: So, basically all sellouts (Score:2)
It seems to me more like the AI bubble is about to burst and the rats are leaving ship. GPT4 is just GPT3 with eye candy, GPT3 wasnâ(TM)t much better than its predecessor and the actual cost of running and training these models far exceeds what people expect vs the value they bring. GitHub CoPilot is great at building some boilerplate template but developers have seen through its limitations and at some point beyond the first day of a project I end up disabling it because the suggestions start to be co
Naive question (Score:2)
Was this all a cheap way for MS to get all of OpenAI before it got a large concrete valuation? Bit of a whispering campaign in the ears of a board member or two?
Re: (Score:2)
Cheap? No.
MS stock dropped 16% on the news of OpenAI firing Altman. It has since mostly recovered, on the rumors of Altman joining Microsoft along with most of his former employees, although it remains unstable. But that was a huge nut-punch to Microsoft stockholders.
Re: (Score:2)
Cheaper than the $86B figure that was being touted is what I was getting at. And the end result is that they've got the bulk of the personnel and none of the debts or other corporate paraphernalia. The stock drop is ephemeral as you said. Stock_holders_ wouldn't be bothered, only speculators.
But I'm just spit-balling, don't mind me.
OpenAI - who owns it? (Score:2)
The board of directors is negotiating mergers and stuff? The board is too powerful and not invested enough. OpenAI needs a goddam conductor.
The basic problem ... (Score:2)
... is the strange design that has a non-profit overseeing a for-profit. The former is trying to attenuate progress while the latter is going for the money. I think the best solution is for the profitable entity to get rid of the non-profit drag chute.
Artificial intelligence is not a religion - it's a money-maker.
Re: (Score:2)