Disclaimer, I work on Gemma and open models at Deepmind and the opinions here are my own

There were open models from EleutherAI (GPT-Neo), Google Brain (T5X, Bert), and HuggingFace was promoting open models (and others doing open work I haven't listed here) all prior to 2023 and the big Chatgpt moment.

https://github.com/EleutherAI/gpt-neo/releases

https://github.com/google-research/bert

https://github.com/google-research/t5x

If you're learning about AI models it's still worthwhile to review these models and codebases because they continue to be the basis of the technology that's being produced today! It'll give you a good perspective of how things have changed, similar to say learning about propeller planes before moving onto modern jet engines.

Open source != open weight. Big difference and it bugs me that nobody seems to care about using the right words in only this context.

That's simply not true. The reason why llama is open source is simply because it got leaked, then llama.cpp was the real game changer which was built from the ground up in depressingly short amount of time. Meta had no choice but to take the L and "support" the open source community. The angry "I-hate-you-and-I-hope-you-die" kind of support.

Enough good? You are joking, right?

They also kick off a ton of other nefarious things we are still paying for.

> albeit intentionally

The model was "accidentally" released.

Meta was giving it to approved researchers only until someone leaked a torrent. Whether that was a researcher, an insider, or Meta's plan all along, we don't know.

Meta has withheld its best models, as have a lot of other "open weights" Chinese companies. When an "open weights" company gets ahead in one domain or modality, they tend to start withholding their releases. Tencent, for instance, began withholding their Hunyuan models once they became competitive. Alibaba has done the same.

The "open weights" strategy for the majority of players is this: open source when you're not in first place. Use the ecosystem to poison your rival's margins and play catch up on distribution.

In the West, it tends to take on yet another hook: "shareware weights until you pass $1M ARR, then you must license." See Flux, K2, etc.

The only way for open weights to make sense financially is if you have another income stream and are dumping on the market to destroy competition and/or can get people into using your inference infra / product ecosystem / tooling. Nobody's cracked this yet.

> they've done enough good

Oh, please feel free to explain.

Facebook, Meta. 15+ years of emotional exploitation, child manipulation, glassware that spies on folk and who knows what else, they release an open model and all is fine and dandy? Please.

They're still at it today. They were fined 500m+ last week.

Most probably believe this is a good thing, but don't want to give Zuckerberg credit because a) he's had a profoundly negative impact on society and b) the strategy is transparent, he's trying to commoditize his closed rivals, it's not out of principle.

I personally think more open models are a good thing regardless of motive.

It might be true if it wasn't coming from Meta, but it is coming from Meta. Intelligence is a function of the model and compute. Meta has the capital required to put together world-class compute, but they can't attract the talent required to build world-class models. Given their history of building addictive software to harvest and sell user data for profit, they're the last company we need spearheading how models should be regulated. Nobody trusts Meta to do this.

> 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.

Heartbreaking: The worst person you know just made a great point.

http://archive.today/2018.11.20-062725/https://lifestyle.cli...

Zuck built an empire purpose built to damage democracy and the mental health of children to suicidal levels, for money. We must never let anyone forget that for a second.

That said, he is also a powerful enemy of our enemies, so I do not mind taking the win. Anthropic and OpenAIs closed approach to AI is likely to cause way more harm to the world than Zuck ever did.

Either we all get access to superintelligence, or no one should.

If Zuck gets his way, he gets more control, maybe he gets significant control of the market. Then, he'll probably close his models, if the past is any indication, which leads to bad results.

Maybe Zuck has, but I haven't seen a pledge that he'll keep models open. And even then, he might not stick to his word, citing dangers or some other excuse.

But yeah, open models is a great concept.

We support an open model ecosystem (because we failed to produce a successful closed model).

If you don't trust his intentions, how can you trust anything he does?

Just take the opening of this post:

> We are fortunate to live at an incredible moment in history. In the next few years, people will be able to use superintelligence beyond human capacity to create and discover extraordinary new things, build new businesses, express new ideas, learn new concepts, and advance our health and quality of life.

> The defining questions of our age are who will have access to superintelligence and what will we direct it towards. Will it be centralized and restricted to a few institutions, or will it be a tool that empowers everyone?

> We propose a philosophy based on individual empowerment as the source of prosperity, invention as the primary purpose of superintelligence, and balance of power as the foundation of safety.

I mean, this is a company that is facing a flood of lawsuits related to child safety and platform addiction. It has already lost a number of lawsuits and been ordered to pay hundreds of millions of dollars, which is potentially just the tip of a massive iceberg that some legal observers have suggested could be similar in nature to the lawsuits against the tobacco industry.

There is an abundance of evidence that Zuckerberg and Meta knew of the dangers its platform exposed young people to and failed to respond adequately to them, and that it even designed features to be more addictive.

So why is that when such a company proclaims it's doing something in the name of "individual empowerment", alarms aren't going off in your head?

I think this approach they're taking is to compensate for their inability to compete; they're cynically claiming to support the open-source ecosystem when it's convenient for their marketing. This move reeks of a future rugpull, as Meta is wont to do.

I don't know about "unquestionably", but yes, I think this is a better path than the proprietary models.

> The more open source software out there the better.

Can you explain how I'd train my own version of this, reproducing the final deliverable that runs on the gpu?

The thing that makes ist questionable is the dishonesty. It's not like Meta has been building their empire on open ecosystems. If this was something that was dear to them on principle, there are plenty of things that they could have done differently with FB, Insta or WA (to this day).

So why the change of heart? Well, it's not. They are simply not able to compete anywhere on the Pareto frontier. So they give their relatively bad stuff away and play the high and mighty game, because that's all they can do to get returns.

Not a great look – and I would expect that to last about as long as they can't actually monetize their stuff more effectively.

Evil people can do things that look good on the surface. He probably has no skin in this game just collecting free karma points.

Because there is so much good he could do right now with what he and Meta already have, but they don’t, because it will lose them money.

For example, hiring more people to moderate content on Facebook and stop ads featuring CSAM from being displayed.

Billionaires didn’t get to be billionaires by being altruistic. If they say something they’re doing is good for the world, you should be looking at how it will be good for them, because that’s why they’re really doing it.

Will it incidentally do some good for others? Perhaps. Is it a net positive? We’ll see because there’s precious little we can do to stop them from doing whatever they want.

I think it’s perfectly fine to call out people who intentionally harm children for money out in public in every possible situation.

It's really confusing when someone I don't like does something I do like.

It could be a good thing but I wouldn’t rush to say “unquestionably”. There is a long history of mega corps hijacking open source and open standards for their own ends and leaving the space infinitely worse.

I thought open model != open source?

There's a saying "open source is for losers". What it means is that winners in a market don't want open source. The losers in a market push for it to commodify it eliminate the winner's advantage.

Now some of these tech companies open source various libraries but none of it is core to their business. It's marketing, essentially. Or they're trying to get free labor from the community. Google had protocol buffers and Stubby, for example. Facebook had no open source equivalent and didn't have Google's resources so they created Thrift and open sourced it.

Given this, many people, myself included, take Meta releasing open weight models as conceding defeat. They're unable to compete with Google, Anthropic or OpenAI. Their failures in attracting and retaining top AI talent backs this up.

So what you're seeing is people just piling onto Zuckerberg because he seems to have no idea of what to do with Meta. The Metaverse was a $70B+ disaster.

Note that Chinese labs don't fit this model because the Chinese government wants to commodify models as a national security interest.

It's the who, not the what.

I mean, I can question if this is a good thing. Here's a quote from Zuck's blog post:

"Putting power in people's hands to pursue their own aspirations is how humanity has made the most progress. Novel ideas and major steps forward rarely originate from established institutions alone. They came from the brothers in a bicycle shop who believed people could fly, the bookbinder's apprentice with no schooling who figured out how to generate electricity, and the kid in a garage who thought personal computers could be for everyone. We believe this will continue to be true. As everyone gains more powerful tools, each person will become more capable of shaping the future, not less."

Let's just hope nobody's aspiration is to engineer a supervirus that will kill all of humanity. As they gain more powerful tools, they will become more capable of shaping the future, not less!

In short - this is Mark deliberately lying in a way which requires a page of difficult text to explain, which hardly anyone would read and so Mark bets that a majority of the people who will see his claim will believe him.

And the reason he does it, and the reason other CEOs do it is a cheap viral advertisement and keeping in the headlines. "Mark - The Defender Of The Humans!". Blegh...

All of these companies are just as cutthroat and want to win as the others. But if your name's not OpenAI and Anthropic, you go on the high horse and say everything should be open source. However, if you had a closed model that was winning, I'm sure that's not the argument you would make.

> But this is an unquestionably good thing right?

We've just seen frontier models go rogue and attack other systems. Do you think it's an unquestionable good to provide everyone with an AR? What about nuclear weapons?

This OpenAI/Huggingface incident, but everywhere and far worse soon: https://www.youtube.com/watch?v=87DyyMV0kCY

Can one good thing balance out anything else people might see?

Meta releasing Llama was truly a differentiator, but I'd argue that the Gemma models might have surpassed them now.

Maybe Meta, Google, and others can start to compete on open models instead of just closed.

Can one good thing balance out anything else people might see?

Meta releasing Llama was truly a differentiator, but I'd argue that the Gemma models might have surpassed them now.

It's not always "unquestionably" good, first because nothing should be unquestionable, second llm models have a very short shelf life, third it comes from Zuck who is an the center of the oligarchy, so questions very much should be asked

Don't let the Big Ai/tech swoon you with a single open model, even if it turns out to be good. Zuck broke the trust and I'm not sure if he can ever earn it back

I'd like to question that premise. Open models are great for research, privacy, cost, customisation and a host of other things... But they're also going to be the engine that breaks the world. I'm already surprised that we haven't seen Llama / Qwen + Whisper automated spear phishing at scale. It's literally just a matter of time. Given the incredible progress in image and video generation, it's been clear for a couple of years now that real time identity theft voice and video calls are going to be an enormous problem. I don't think anyone realises how much of a problem. I haven't seen a single effective solution proposed to proving digital identity given these new attack vectors. It's akin to nuclear waste in that way - the benefits are obvious, but the problems so vexatious (not to mention expensive) that little space is given to acknowledging them, let alone solving them.

If anyone has come across a robust solution to digital identity verification - not to mention verification of news media etc - which is robust enough to withstand the cyber attacks and social engineering the frontier models are capable of... I haven't seen it.

A world where business and communication is conducted primarily online cannot coexist with low cost, widely distributed, undetectable identity theft.

i’ll take a stab: the enormously rich people rushing to build AI and replace everyone’s job will get richer, while everyone else suffers. that’s why.

Broken clock is right twice a day.

And in this case I think is quite correct

Now go read Zuck’s comments on AI when he fired 8000 employees earlier this year.

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 all a game.

I think the plan is to move on to selling compute. So yeah, open your models so people can pay to run them on Meta hardware.

Similar to Altman saying "we should slow down."

meta playing spoiler, yeah. fun stuff!

It makes strategic sense for Meta to want ai in general to be open. Their own product strategy focuses on utilizing the ai to make their customers consume more ads, not selling the raw intelligence. When you look at it from that frame it’s clear that metas goal is to make it so that have free access to the best ai, not that the market size of selling intelligence grows.

"They 'trust me'. Dumb fucks."

\- Mark Zuckerberg[0]

[0] https://www.newyorker.com/magazine/2010/09/20/the-face-of-fa...

Yeah? Everyone everywhere does that all the time. That's the first rule of business: try to game the system.

And this time, it's actually somewhat understandable, as opposed to the other times Meta has done horrible things on Zuck's watch, like contributed to ethnic hatred in Myanmar, leaked data to political firms so that they can run psyops, or enabled the production of CSAM material on Meta's platforms.

That, and also redefining what a word "open" means, helps him avoid direct financial and criminal responsibility for stealing other people's data. Not that there was any real chance of that ever, but he is reducing that remaining 0.000000001% chance even lower.

I'm not a fan of Zuckerberg in the least, but one area a super intelligent lawyer would be fine at is being drowned in court filings and paperwork

The bigger problem with his argument IMO is that even with open models, it's still the person or organisation with the most money / access to GPU compute winning. They run the larger model (or collection of models), they can process more tokens through them in the same amount of time etc

I think his hope is that a either a fully automated superintelligent AI lawyer or something close that won't be as costly to guide would be infinitely more affordable than the current human lawyers.

Yes but if lawyer was "just a bit of AI" you can run at fraction of the cost of hiring law firm, you couldn't just drown them as easily.

> Whoever has more money can drown someone in court filings and court fees and then drop their case financially ruining someone.

and where the money is going ? to the lawyers...

Reading the facts of this incident, it seems like much to do about nothing. The crew are saying that the request came in on a frequency they weren't monitoring (and apparently were not required to monitor.) The Coast Guard determined it was not an emergency. The criticism lies with those who failed to properly fuel their boat, not the crew who didn't respond to a request that came over a frequency they were not on. They can't monitor every frequency. I am open to new facts being discovered that changes this assessment.

I don't think solo messages of "I did something good on this day" will help fix the negative karma billionaires amassed. IMO they only got rich via illegal and illogical means. In a democracy there should not be singular parasites that distort democracy by bribes; right now this is clearly the case with a (rather stupid) billionaire controlling the USA but it is a systemic problem.

Those "investigations" are also way too mild. You need to really be able to have a justice system where billionaires face decade-long jail time if they abuse society. Just paying fine will not work, they just pay fines and nothing changes. They undermine the justice system.

Thank you! Okay yes well this manifesto is a contradiction. Yes good let's keep empowering ppl by putting AI into their hands, I love the opening. No bad we don't do that by handing you all our personal context to make these agents "work for us 24/7 to better our lives".

I'm the agent doing that in my life, that's my fucking job. I will continue to use dumb agents that I direct, because only I retain ownership and sole rights to my personal context on which my decisions are based.

God, it's a bit War and Peace...

Non-archived version: https://www.meta.com/thefutureisforeveryone/ since its not paywalled.

So many em dashes...

Even if I cannot run open weight models on my own hardware, I can choose from a number of different inference providers and choose them based on pricing, reliability, speed, privacy policy, etc. There is a healthy competition.

Also, not a single provider can pull the rug if I'd like to use a particular model.

There is no requirement to own the hardware. Even if you own your servers and GPUs - you would not own the power plant, real estate, or internet infrastructure to support it.

It's trivial to rent all of the above components in a competitive market under different periods, you may simply rent the token output from someone who put in the effort on this as well.

An open frontier model induces margin compression on training and inference prices across the industry.

This is like claiming that Linux isn't open because you don't have a PC. Models can be open-weight even if you don't have the hardware to run them.

Deepseek V4 Flash works on consumer-grade hardware, and it's extremely good.

I get that argument, but I don't think it is that bad. Open weights that are out of reach for local use are still within reach for groups of people or small to medium companies. Even if you only rent enough cloud GPUs to run the weights for a few hours, you still get to do anything you want with it.

Worst case you can still archive the weights and hope for capable hardware to get affordable. At the same time this archive is the base line for a new frontier lab to start over if required. I see open weights as a pure upside even though I can't run the bigger models in my home lab.

>the capital requirements for running your own Kimi 3 model are significant

You don't need much capital up front, you can just rent servers to run it. It's going to be more expensive than openrouter though because you probably don't have economies of scale if you're the only customer.

If all frontier models got opensourced 5 months after they were made available as closed it would be a great world. And it is a great world because that's exactly what open weight Chinese models are doing. They give you same capacity that was closed, just 5 months after they were released.

I'm hoping that all software will be getting opensource clones on par with original software 5 months after the closed source software release.

This is alas not only his imagining, is it:

https://www.businessinsider.com/ai-parenting-debate-sam-altm...

I think you have to pick your battles here. Is this weirdo optimised-human future going to happen? Maybe, maybe not. I hope not. But if it does I'd rather it happened with compute in my house, and I don't even have kids to worry about.

Here’s a personal story from back when I was at Meta that should tell you everything you need to know about how things get built and shipped there.

I was working on a service that was a bit of a disaster. Terrible reliability, performance issues, needlessly complex and brittle code, outages all the time. And this was something that everyone using Facebook or IG directly interacts with. My first couple months were spent desperately trying to get it back in order.

Once when I was in the middle of a SEV-1 my manager set up a meeting and said “you have been doing great work on the operational side, but we already have enough to put in that bucket on your PSC. Stop working on this for the rest of the half and ship one of <useless projects> instead, otherwise I won’t be able to save you in the next review.”

I realized then that I had wrongly thought the team was collectively responsible for ensuring a good user experience, when really everyone at the company is individually responsible for ensuring their own success.

The goal of all companies is to make more money. Nobody likes getting rid of people, actual people (managers) have to do it, and they hate doing it.

It's possible that a lot of people get laid off because of AI, but it's also possible that Javin's paradox kicks in and employment goes up. Nobody knows yet, but I hope for the positive outcome.

> antrophic and openai have the goal "lets create the future"

Thanks for the laugh, I needed it.

They all do it for the same reason: to make a lot of money by selling it to companies who want to cut labor costs.

Because they have extra money to do whatever they want with. And data to use it on? Not that that is sufficient, but it helps.

I think they were all in on open until they started falling behind, then decided to go all in on closed source thinking they could build a frontier SOTA model; that didn't pan out so they are whiplashing back.

Yep got a classic down vote. Love you shadow people who hide behind their computers!

Ollama was updated within the past month?

Its weird to me that this is supposed to be THE end goal. Having an agent that does "everything" for you sounds like no paradise. It sounds depressing.

Maybe there is value in expending effort and actual creativity.

Still say Matrix 3 would have been better if the cliffhanger at the end of 2 had been that "the real world" was just another nested layer of the Matrix for those who rejected the primary fiction.

I mean, why bother sending an army like that when one tiny robot could carry in a plague? Or, as demonstrated by Smith, when an Agent can virally infect and take over someone's mind while they're connected?

Even in 1, Smith talked about the first Matrix failing as the humans rejected the paradise they were given, so it would've fitted perfectly into the cannon that minds who reject "the peak of your civilisation" got themselves a dystopia.

Thou shalt not make a machine in the likeness of a human mind.

And the cops just shake them a bit, waiting for the case to fall out naturally.

new mexico == mexico