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

Mark Zuckerberg attacks 'closed' AI rivals as Meta returns to open models

ft.com

Post reply on HN
Print view View original

301–310 of 631 posts

Re: Mark Zuckerberg attacks 'closed' AI rivals as Meta returns to open models

#301

Post by JumpCrisscross »

noosphr wrote: post #288
JumpCrisscross wrote: post #276

Earlier quoted context omitted.

Yup. An extremist wing of the FOSS movement ceded the debate early on by trying to insist open source required full access to the training data. Philosophically, not wrong. But practically fucked, so the word evolved. Within tech circles, open weight != open source. Outside them, they’re synonyms.

Training data without the training regime is useless. A cake is not open source because they list the ingredients.

Open source captures a practical utility as well as a philosophy. When those two cease to converge, the practical prerogative wins.

The correct battle would have been weights + regime. But extremists insisted on data, too, which left Meta as the only other real voice arguing with anything practical. They had open weights. I think eventually open use was negotiated and that closed the case except for the folks still arguing about how to pronounce GIF.

Re: Mark Zuckerberg attacks 'closed' AI rivals as Meta returns to open models

#302

Post by sterlind »

bushido wrote: post #226

I think something that doesn't get said enough is Meta did, albeit intentionally kick off the origin of the open source race back in 2023 with the release of llama. I'm not a big fan of meta in general, but they've done enough good, and it's possible that it was intentional as well. I don't know, I wasn't in the rooms, and I think it's worth giving them some reasonable doubt. No one is purely good, and no one is pure…

llama was only three years ago? holy smokes! the industry is advancing so fast.

Re: Mark Zuckerberg attacks 'closed' AI rivals as Meta returns to open models

#303

Post by sterlind »

JumpCrisscross wrote: post #276
semiquaver wrote: post #272

Earlier quoted context omitted.

Words mean what people use them to mean. Ship has sailed whether you approve or not.

Yup. An extremist wing of the FOSS movement ceded the debate early on by trying to insist open source required full access to the training data. Philosophically, not wrong. But practically fucked, so the word evolved. Within tech circles, open weight != open source. Outside them, they’re synonyms.

also, you can do a lot to finetune or repurpose an open weight model. much more easily than you can mod closed source binaries.

Re: Mark Zuckerberg attacks 'closed' AI rivals as Meta returns to open models

#304

Post by btmiller »

bushido wrote: post #226

I think something that doesn't get said enough is Meta did, albeit intentionally kick off the origin of the open source race back in 2023 with the release of llama. I'm not a big fan of meta in general, but they've done enough good, and it's possible that it was intentional as well. I don't know, I wasn't in the rooms, and I think it's worth giving them some reasonable doubt. No one is purely good, and no one is pure…

[flagged]

Re: Mark Zuckerberg attacks 'closed' AI rivals as Meta returns to open models

#305

Post by alex1138 »

ben_w wrote: post #296
SoftTalker wrote: post #250

Earlier quoted context omitted.

Competely agree. React and whatever else they've open sourced is inconsequential compared to the harm they've caused.

I'd also argue React is on the "harm" side. :P

Check out the 'explaining react's license' thread and all the complaints over that

Re: Mark Zuckerberg attacks 'closed' AI rivals as Meta returns to open models

#306

Post by thraway3837 »

overfeed wrote: post #153
drob518 wrote: post #51

Earlier quoted context omitted.

It’s a long term pattern. When you’re winning (Anthropic), you keep the tech closed and try to monetize it as much as possible. When you’re losing (Meta), you open it up or drag it into a standards committee to either devalue it or slow the leader down while you prepare a “standard” version of it. You also highlight how altruistic and morally good you are for having done so. There is nothing new under the sun. It’s a…

> ...drag it into a standards committee to either devalue it or slow the leader down Anthropic et al recently proposed a supranational AI governing body designed to slow everyone down (euphemistically calling it "AI pacing"). Are the Pacer signatories losing?

The governing body is only the western allies. USA + AUS + EU. Or anyone else that may sign a deeply disturbing and restrictive treaty. It's to make sure that the global majority does not make a better model or have access to the same hardware. As is currently the case with embargoes on EUV machines going to China, etc.

Models are not a moat, and I think OpenAI/Anthropic know this. They know that their IPO is being weakened by the open models. Neither is hardware. We are not far from other countries catching up with silicon that matches or exceeds western capabilities, and it will be cheaper.

Re: Mark Zuckerberg attacks 'closed' AI rivals as Meta returns to open models

#307

Post by A_D_E_P_T »

applicative wrote: post #259
A_D_E_P_T wrote: post #30

Earlier quoted context omitted.

> But this is an unquestionably good thing right? Their efforts in open AI are by far the best thing Facebook/Meta have ever done. They opened the door to the Chinese, who are doing excellent work, and between them they are preventing the concentration of power and the rise of monopoly pricing. This is enough to absolve them of almost any sin. If I were a utilitarian, I'd unironically be a Zuckerberg fan right now.

The price of Kimi K3 is 'monopolistically' determined by contract with Moonshot. The weights are nominally on Hugging Face but can only be provided under contract with Moonshot, which specifies what the price can be. So it will be with the next Alibaba behemoth and, I would think, all others forever. Soon you will sing songs for the freedom Xi has given us when you read a declaration of 'open weights' ... for a model…

> The Kimi K3 'weights' are an opaque blob that can only be used by contract.

Have you looked at the contract? There are zero conditions unless you've broken $20M in revenue with their model. It doesn't even forbid distillation, lol

> Soon you will sing songs for the freedom Xi has given us when you read a declaration of 'open weights' ... for a model so huge it takes a nuclear powered data center to run and crashes the Hugging Face servers when it is uploaded.

Releasing a big open-weights model is... le bad? Am I understanding you correctly?

There are small models out there if you want them, you know.

Re: Mark Zuckerberg attacks 'closed' AI rivals as Meta returns to open models

#308

Post by s0ss »

losvedir wrote: post #289
s0ss wrote: post #270

Earlier quoted context omitted.

This is not open source; you mean open weight. These models are the antithesis of FOSS. Is it better than hosted models from other big labs? Yes, but not by much from a freedom perspective. especially considering the texts these were trained on. Grumble grumble, these details matter.

Genuine question, but why does it matter? It seems to me the vast majority of the benefit comes in the weights. Then you can self host, quantize, finetune, ablate, etc. What does having the original data get you beyond that?

open source gives you freedom to read/edit/learn from the source code. Open weights don't let you read/edit/learn from the source. open wights have more in common with traditional binary distribution of software, ala closed source software. Only in this specific context has the entire meaning of the words totally inverted.

I can run lots of binaries on my computer that don't have source available, they are not open source. These words matter.

Re: Mark Zuckerberg attacks 'closed' AI rivals as Meta returns to open models

#309

Post by aeve890 »

esafak wrote: post #188
aeve890 wrote: post #139

Earlier quoted context omitted.

>We've just seen frontier models go rogue and attack other systems. Have we though? A LLM agent doesn't have any agency at all. It can't "go rogue". To go rogue you need agency to act independently and be aware that you're breaking the rules or understand what does it mean to ignore orders. An agent it's a software that run a series of steps to reach a goal. If it have a large enough library of strategies and zero gu…

You can tell one to rewrite complex applications in a different language, solve open research problems, or develop novel viruses. This is nothing like pressing the gas pedal.

So? I'm not arguing they can't do all that. I'm arguing against the narrative that llms have agency enough to act in an adversarial way.

At best someone could argue that an agent attacks like a bacteria does, just following automated chemical and genetic programming. But you wouldn't call that an attack or attribute moral values to their actions, because they don't have moral agency. They can't "go rogue", they can't disobey.

Just like llms, their automated actions are direct product of programming. Yes they can do amazingly complex shit, exactly like a car does when you press the gas pedal.

Re: Mark Zuckerberg attacks 'closed' AI rivals as Meta returns to open models

#310

Post by ben_w »

ViktorRay wrote: post #20

Comments here are surprising to me. I get folks don’t like Zuckerberg and his company and don’t trust his intentions… I don’t either. But this is an unquestionably good thing right?. The more open source software out there the better. And the more open weights or even over source AI stuff the better too right? More competition the better generally speaking I think. Unless I’m missing something and am getting this who…

> But this is an unquestionably good thing right?. The more open source software out there the better.

I say no to both. Of the things we've made on purpose, but excluding where we were actually trying to make them opaque like e.g. cryptography, a trained artificial neural network is the most difficult thing to understand the inner workings of. This makes it a complete pain to even evaluate if one model is better or worse than another for your needs, so we have to mostly outsource these to other people's rankings and hope the score on ARC-AGI-3 or τ-bench or DeepSWE v1.1 or BioMysteryBench or whatever, actually corresponds to something we care about. Which it might do kinda but on the other hand a high score may turn out to be the curse of Goodhart.

Also, as with the Chinese models and the social media feed algorithms, the only way to tell if there's some systemic flaw in it is by analysing the aggregate outcomes. It is claimed (I can't read the laws myself*) that the Chinese government requires models to support the government's worldview about e.g. Tiananmen Square; and we have seen examples of Grok glazing Musk in amusingly stupid ways; so I fully expect something similar from Zuckerberg, e.g. requiring the model to glaze Meta products or propagandise for things Zuckerberg wants as a billionaire.

* every time I try to illustrate how mediocre Google Translate is from English to Chinese, the result is so bad that one of the replies is someone telling me the Chinese example I give is borderline gibberish.

Also, even if I could actually read it, I'm not a lawyer.

Post reply on HN