Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Greg Brockman quits OpenAI

twitter.com

Post reply on HN
Print view View original

591–600 of 701 posts

Re: Greg Brockman quits OpenAI

#591

Post by satvikpendem »

solardev wrote: post #382
015a wrote: post #214

Earlier quoted context omitted.

Sure; but its still weird that Microsoft agreed to the deal with the board in the state that it was; not just no board seat, but three absolute outsiders, two of them extremely unqualified. We may look back on their decision to buy 49% of OpenAI as a big misstep.

Wasn't it a Hail Mary for Microsoft? They're not doing anything else particularly earth shattering, and if this came out without them, they'd be even less relevant. If OpenAI fought this and won without them, Microsoft would have nothing to compete against Google and everyone else with. Did Microsoft have any other route to AI relevance?

Sure, they could simply copy the GPT papers (as they are entirely public) and implement them inside their own products, as they are doing already with GitHub and Office. There is really no need to hang onto OpenAI's word.

Re: Greg Brockman quits OpenAI

#592

Post by satvikpendem »

chairhairair wrote: post #414
krapp wrote: post #180

Earlier quoted context omitted.

Governments will have their own black-budget private LLM networks, they don't need OpenAI. The NSA probably has a whole cluster of them in its data center in Utah, trained on every public and private communication they've slurped up over the years, likely a generation or two ahead of what's available to the public.

This is immensely dumb. What secret cabal of researchers would they be hiring that would be capable of being ahead of Deepmind/OpenAI? Where exactly would they find these people? Shadow MIT? CalTech2?

I can tell you right now that the government agencies are ahead of for-profit ones. Whether you choose to believe it is up to you.

Re: Greg Brockman quits OpenAI

#593

Post by vikramkr »

Sebb767 wrote: post #333
symlinkk wrote: post #278

Earlier quoted context omitted.

it’s a causal writing style that has become popular online in the last decade or so. don’t overthink it

It's usual in chats, not for formal writing like officially announcing your resignation. At least that's my impression.

This is the sector of the economy that ipos in cargo shorts I don't think formal writing would be expected or appreciated

Re: Greg Brockman quits OpenAI

#594

Post by PeterStuer »

johnwheeler wrote: post #84

Edit: I called it https://twitter.com/karaswisher/status/1725682088639119857 nothing to do with dishonesty. That’s just the official reason. ———- I haven’t heard anyone commenting about this, but the two main figures here-consider: This MUST come down to a disagreement between Altman and Sutskever. Also interesting that Sutskever tweeted a month and a half ago https://twitter.com/ilyasut/status/1707752576077176907 Th…

Feels like Griffindor beheaded Slytherin right before Voldemort could make them his own. Hogwarts will be in turmoil but that price was unavoidable given the existential threath?

Re: Greg Brockman quits OpenAI

#595

Post by visarga »

MediumD wrote: post #564
cedws wrote: post #517

Earlier quoted context omitted.

>Ilya claims the transformer architecture, with some modification for efficiency, is actually sufficient for AGI. I thought this guy was supposed to know what he's talking about? There was a paper that shows LLMs cannot generalise[0]. Anybody who's used ChatGPT can see there's imperfections. [0] https://arxiv.org/abs/2309.12288

> We provide evidence for the Reversal Curse by finetuning GPT-3 and Llama-1 on fictitious statements such as "Uriah Hawthorne is the composer of 'Abyssal Melodies'" and showing that they fail to correctly answer "Who composed 'Abyssal Melodies?'". The Reversal Curse is robust across model sizes and model families and is not alleviated by data augmentation. This just proves that the LLMs available to them, with the t…

No, if you read this article it shows there were some issues with the way they tested.

> The claim that GPT-4 can’t make B to A generalizations is false. And not what the authors were claiming. They were talking about these kinds of generalizations from pre and post training.

> When you divide data into prompt and completion pairs and the completions never reference the prompts or even hint at it, you’ve successfully trained a prompt completion A is B model but not one that will readily go from B is A. LLMs trained on “A is B” fail to learn “B is A” when the training date is split into prompt and completion pairs

Simple fix - put prompt and completion together, don't do gradients just for the completion, but also for the prompt. Or just make sure the model trains on data going in both directions by augmenting it pre-training.

https://andrewmayne.com/2023/11/14/is-the-reversal-curse-rea...

Re: Greg Brockman quits OpenAI

#596

Post by Paul-Craft »

vikramkr wrote: post #590
Paul-Craft wrote: post #585

Earlier quoted context omitted.

I find it hard to believe that the board of OpenAI isn't smart, savvy and self-interested enough to know that not delaying the announcement by an hour or so is the wrong move. That leads me to believe that yes, this was something big and worthy enough of being announced with that timing, and that it was probably not a mistake.

They also said Greg was going to stay at the company and then he immediately quit. I find it very hard to believe that smart, savvy, and self interested are adjectives that apply to a board who doesn't know what their own chairman thinks.

Even smart, savvy, and self interested people can't always predict what individual humans are going to do. It's certainly an interesting wrinkle, but I don't think it's relevant to the limited scope of the analysis I've presented here.

Re: Greg Brockman quits OpenAI

#597

Post by satvikpendem »

lenerdenator wrote: post #289
johnwheeler wrote: post #226

Earlier quoted context omitted.

I’m not saying there is lack of truth. I’m saying that’s not the real reason. It could be there’s a scandal to be found, but my guess is the hostility from OpenAI is just preemptive. There’s really no nice way to tell someone to fuck off from the biggest thing. Ever.

I mean I'm not a lawyer (of the big city or simple country varieties, or any other variety) but if you talk to most HR people they'll tell you that if they ever get a phone call from a prospective employer to confirm details about someone having worked there previously, the three things they'll typically say are: 1) a confirmation of the dates of employment 2) a confirmation of the role/title during employment 3) whe…

There is no legal justification for ever saying those dates, much less their department and role. I have never heard of any HR department saying anything of the sort, even if this is an oft-quoted meme of HR. I suspect you have actually never worked in HR to provide such statements, you are merely speculating.

Re: Greg Brockman quits OpenAI

#598

Post by TeMPOraL »

lijok wrote: post #512
Jerrrry wrote: post #468

Earlier quoted context omitted.

>John >I'm not patronizing you (A)ssuming (G)ood (F)aith, referring to someone online by their name, even in an edge case where their username is their name, is considered patronizing as it is difficult to convey a tone via text medium that isn't perceived as a mockery/veiled threat. This may be a US-internet thing; analogous to getting within striking distance with a raised voice can be a capital offense in the US,…

> referring to someone online by their name is considered patronizing This has to be a joke, right?

No. More than that, comes off as patronizing to start a comment with the other person's first name when speaking, off-line, face-to-face, unless you're their spouse, parent, or in some other close relationship.

Re: Greg Brockman quits OpenAI

#599

Post by TeMPOraL »

Filligree wrote: post #588
jrockway wrote: post #519

Earlier quoted context omitted.

I don't think it's a joke. I would find it patronizing unless I'm already on a first name basis with the commenter through some prior relationship.

Really? Referring to someone by first name is perfectly ordinary where I’m from, regardless of relationship. If someone doesn’t want me to do that, I’d expect them to introduce themselves as “Mr. so-and-so”, instead.

It's not the first name alone, it's also the sentence structure. "Hey John, did you hear about..." sounds perfectly normal even when talking on-line to strangers. "John, you misunderstand..." is appropriate if you're their parent or spouse or otherwise in some kind of close relationship.

Re: Greg Brockman quits OpenAI

#600

Post by oooyay »

lijok wrote: post #512
Jerrrry wrote: post #468

Earlier quoted context omitted.

>John >I'm not patronizing you (A)ssuming (G)ood (F)aith, referring to someone online by their name, even in an edge case where their username is their name, is considered patronizing as it is difficult to convey a tone via text medium that isn't perceived as a mockery/veiled threat. This may be a US-internet thing; analogous to getting within striking distance with a raised voice can be a capital offense in the US,…

> referring to someone online by their name is considered patronizing This has to be a joke, right?

No. Look at examples where people hurl veiled threats at dang. They almost always use his real first name. It's a form of subtle intimidation. That kind of intimidation, whether the users real name is incorporated into their username in some way or they're using other open source intel goes back to the early days of the internet.
Post reply on HN