Former Richard Stallman Colleague Now Argues for Open AI Models Too (fortune.com) 33
Long-time Slashdot reader theodp writes:
Recalling his initial resistance to free and open software, billionaire computer scientist David Siegel argues vigorously in FORTUNE that the stakes are too high to let AI become increasingly closed. "In the 1980s, I had the chance to spend several years arguing about free and open software, what we now call open source, with the founder of the movement, Richard Stallman. My office at the MIT AI Lab was next door to his. Stallman's position was that the source code to software should be free for everyone to use, learn from, and improve. Software encapsulates knowledge, he argued, and no one should lock something so fundamental away. To hide software inside a company was to hide knowledge itself... What I missed was that software was not just a commercial asset; it was a body of knowledge, and bodies of knowledge grow stronger when they are shared. After about two years of on-and-off debate, Stallman convinced me I was wrong."
"Now the AI fight is the same — only bigger," advises Siegel. "AI is software, and AI is increasingly closed. The frontier models — the most advanced, cutting-edge AI systems — are closed completely and the trend is accelerating. Viable open alternatives are few and far between." So, what to do...? "Yes, frontier models keep getting bigger and more expensive — that arms race may well stay with the giants. But open source AI does not have to match their scale to be useful. Much of what the world needs probably does not require the absolute frontier. And where keeping a credible open option does demand serious compute, that is precisely the kind of public good worth paying for.
"What's missing is not a path but will. The government, the private sector, and nonprofits should invest heavily in free and open source AI — the way they once invested in open software: public compute grants for open research, corporate and philanthropic support for universities and nonprofits doing the work, and a simple rule that AI built with public money is open by default.
"We have run this experiment before. We know how it turns out. Let's not unlearn it."
"Now the AI fight is the same — only bigger," advises Siegel. "AI is software, and AI is increasingly closed. The frontier models — the most advanced, cutting-edge AI systems — are closed completely and the trend is accelerating. Viable open alternatives are few and far between." So, what to do...? "Yes, frontier models keep getting bigger and more expensive — that arms race may well stay with the giants. But open source AI does not have to match their scale to be useful. Much of what the world needs probably does not require the absolute frontier. And where keeping a credible open option does demand serious compute, that is precisely the kind of public good worth paying for.
"What's missing is not a path but will. The government, the private sector, and nonprofits should invest heavily in free and open source AI — the way they once invested in open software: public compute grants for open research, corporate and philanthropic support for universities and nonprofits doing the work, and a simple rule that AI built with public money is open by default.
"We have run this experiment before. We know how it turns out. Let's not unlearn it."
Re: (Score:2)
LOL you have "use cases".
As to what I'd prefer, neither of these. I'd prefer AI developed to realize its potential rather than be corrupted and destroy civilization. This entire conversation is even relevant to that.
Re: (Score:2)
Don't forget about Meta. The first llama started got things started. But yeah, now the Chinese are the driving force behind open models.
I'd say Qwen 3.6 for coding, Gemma 4 for most other things, in particular for languages other than English and Chinese.
Public Good???? (Score:4, Interesting)
Re: (Score:2)
"public good" is a handy catchphrase useful for justifying when billionaires steal. Wasn't DOGE for the "public good"?
Re: (Score:2)
"public good" is a handy catchphrase useful for justifying when billionaires steal. Wasn't DOGE for the "public good"?
I think what we have here may be a homonym problem.
First, there's the Public Good which exists as a philosophical, political, and social ideal. That's the inclusive one we all desire and understand intuitively.
Then there's the Public Good which is a propaganda tool in the hands of patronizing corporatists. That's the one used to gaslight and hoodwink us into believing that the oligarchs have our best interests at heart and don't regard us as a mere resource base to be strip-mined and discarded like so much
Re: (Score:3)
In all seriousness, it's a question of whether you view AI as a product or as infrastructure. If you view it as a product, your goal is to make as much profit as possible from it. If you can't make a profit, it's a failure as a product. But if you view it as infrastructure, you give it away for free, or at most charge just enough to cover your expenses. You aren't trying to make money directly from the models. Your goal is to enable as many other things as possible.
For the most part, the US is treating
Re: (Score:2)
Current AI infrastructure is a scientific dead end.
Re: (Score:2)
Putting AI as infrastructure will just lead to further corporate abuses of the public's taxes.
Open training data matters too (Score:3)
Re: (Score:3, Insightful)
Yep. Open weights is closer to having precompiled binaries only (though you can do a little bit more with open weights). All training data with labeling, plus a detailed build log including any special fine-tuning logs, is needed, especially to help show how the final weights were arrived at (since there can be randomness involved). Unless someone else can both replicate your final output plus make their own changes to it, it's not fully open source.
Re: (Score:1)
The full crawling dataset is not the most interesting part, though. Who wants to download many terabytes of low quality data to retrain a model from scratch, that's already available?
The post-training data is much more interesting, because you can for example train another alignment on the base model (e.g. less "You're completely right!" attitude) and try to reduce the repetitive style of many models, or make them not to overformat everything with Markdown and emojis. Also the whole instruction following is
Problem (Score:2)
Much of an "AI model" is just the underlying statistical weights. The engine sitting atop that probability data might be either open or proprietary. But it's the weights themselves that define the generated results. And to date, the claim is that no one really understands how an LLM makes its decisions based on them. So, open/closed doesn't make that much difference.
Open source for the engine part might prevent someone from sneaking something "evil" into the process. But if the LLM was trained on faulty da
Re: Problem (Score:3)
The idea of "open weights" AI models makes about as much sense as "open binary" software.
It means you can run it locally. It doesn't mean you can modify the source and recompile and enjoy your customized version.
The following analogies aren't perfect but they are useful:
Using backpropagation to make an LLM from training data is like using a compiler to turn source code into an LLM. Using evaluations to monitor the training process is like using tests to monitor the continuous integration pipeline.
An open so
Re: (Score:3)
Not really.
To change one line in a program, you need the whole file (and possibly other files) to recompile it. To change a detail in a LLM, you need the open weights but not the training data, to finetune the LLM with that detail. A "full rebuild" would be way to costly for most people and most purposes. Do you think when OpenAI created GPT-5.4 they completely retrained with the 5.3 data plus the new one? The continued training using the weights of GPT-5.3 to create 5.4. If you'd have the 5.3 weights, you
Re: Problem (Score:2)
That's a bit like saying that compiled binaries are just as good as source code because you can modify the program's behavior by altering the machine code. Just reverse engineer it and patch the binary. Who needs source code anyway?
Also, finding the right weights to update is hard. And by hard I mean that an army of PhD students and postdocs is working on that problem.
As for continued training of an existing model... a web search for "catastrophic forgetting" will be instructive. You can mitigate that risk
Re: Problem (Score:2)
Yes and no. You're not wrong, but there's a limit to how far you can go with that approach. A web search for "catastrophic forgetting" will explain more.
It's a hard problem.
Re: (Score:2)
The "nobody understands" common wisdom never was really true (you can for example trivially visualize what attends to each other in a text), but understanding gets even better over time with recent tools like J-Lens. Look into the paper, it's really interesting how you can look into the hidden layers of the model and kinda "debug" them.
You have some hurdles (Score:2)
Re: (Score:2)
That's only true of creating the original model in the first place. Once you have a model, fine tuning it for your own purpose is much much cheaper. Few companies can afford to do the former, but lots of companies can afford to do the latter. Good quality open weights models really do democratize the whole field.
a bunch of lies and ignorance (Score:1)
"Stallman's position was that the source code to software should be free for everyone to use, learn from, and improve."
That was RMS's public-facing claim. In reality, his position was that source code should be free for HIM to use. As a student, he got upset that he couldn't have source code to a driver he wanted. Regarding software projects he was involved in, RMS loved being a dictator, do it his way or get lost.
"Software encapsulates knowledge, he argued, and no one should lock something so fundamenta
A former Richard stallman colleague? (Score:1)
We all know what AI is for (Score:3)
People keep asking, if nobody has any jobs who is going to buy their products.
Try flipping that on its head.
Imagine you're a billionaire or a trillionaire. Every single day of your life your fortune is completely dependent on these pathetic worms buying your products. If at any time they stop buying them and they stop listening to you sure you still got money but you are no longer a trillionaire. That dependency is infuriating someone is great as you.
You are being replaced by machines. Maybe they won't pull it off all at once but they are absolutely spending hundreds of billions of dollars to make it happen. And because you let them have all that money they can do it
Not really (Score:2)
rms helped popularize open source, but he didn’t begin the movement. He’s just a figurehead, and one with way too much baggage after repeated defense of pedophilia and a colleague who made repeated trips to Epstein island, even after mit told him explicitly to stop. It’s time to forget the joker ever existed.
A.I. & Open Source (Score:3)
The current situation is that the hype is off the charts. The investments are completely incomprehensible, being driven by FOMO (Fear Of Missing Out) and intense greed. Any other industry, saner minds would prevail. But for some reason Silicon Valley holds a charismatic demigod like influence over their investors. Nobody knows when the bubble is going to burst but it for sure is coming in the future. It will be a spectacular financial crash of epic proportions far worse than the US housing market snafu. Like back then, few will spot the warning signs and make a killing by shorting stocks. Nobody is paying attention to the risks currently being taken just as in 2008.
OpenAI is leveraged to the teeth by a tangled web of deals with Microsoft, Google, Amazon, etc. Scaling across many data centers with more being built. The cost of tokens is highly subsidized. Memory and storage manufacturers have abandoned retail in order to earn a cool trillion selling to these data centers. The electric grid is not capable of providing an ever increasing amount of power. There is talk of small commercial nuclear reactors but the regulatory requirements would be insane. There is concern over water and environmental pollution. The rising cost of electricity. They sell the data center to the states claiming they are creating jobs. Once the DC is built, what a couple dozen workers could manage the whole thing as most everything it's doing is remote. You just have hands and feet onsite. The jobs for construction would be temporary.
At some point the investors are going to expect an ROI and then the money is going to start drying up. With the thin margins upon which most all A.I. companies are riding things are going to implode. OpenAI may be the first and most spectacular financial failure in the history of the world.
Recently the Chinese did it a second time. First time was Deepseek. Now they have new models that again outperform US models. The Chinese models are open source. They are under severe constraints because they don't have endless top tier Nvidia GPUs so they have no choice but to increase efficiency. We landed men on the moon using extremely primitive computers. Even today's smartwatches would appear to be like a super computer in comparison. Open source would allow for far greater efficiencies. Anthropic's Claude Code CLI is using web based frontend developer React + Yoga to render the command line interface at 60 fps. It's absolutely insane and awful engineering. Their interface was flickering for a very long time. Never being fixed until they added a flag to stop the flicker. But they just added even more insane code on top of the previous badly engineered code. Then their code was leaked by their own A.I. pushing it out to GitHub wide open to everyone. Analysis of that code revealed a god awful mess. It's embarrassing really... Not only is the code from Anthropic & OpenAI closed source but they are actively forcing users to get locked into their LLMs by not allowing for API access without charging an arm and a leg to do so. i.e. charging by the token while subsidizing token usage within their own client software.
We need to stop these con artists before they destroy the worlds economies. Sam Altman may be the most dangerous man alive.The code must be freed and openly shared. You never know when some retired software engineer will make a breakthrough in code. It's happened before. Sun / Oracle had 32bit code to run Linux native binaries on Solaris and they killed it because it was too difficult to convert it to 64bit. Along came some random unicorn developer who made it work and shared his findings. In short order it was adopted by OpenSolaris. Closed source is too slow to evolve we need maximum eyeballs working on it if A.I. is truly to succeed.
A.I. is currently faking it till it makes it. There is no guarantee it will ever reach AGI or super intelligence. Even the scientists don't understand precisely how it is even doing what it is doing. But the hype machine keeps on churning. Sure, A.I. is constantly evolving and getting better and better rapidly. Will it ever live up to the hype? Nobody really knows. They just keep lighting money on fire hoping it makes it. A.I. is a significantly powerful tool but its nowhere near the hype around it.
Open models are becoming a security requirement (Score:2)
Huggingface has disclosed a security incident [huggingface.co] in which AI driven infiltrant (17k actions over a couple days) chained vulnerabilities to move through the infrastructure.
The responded with AI-driven analysis, but Anthropic and OpenAI models refused to perform that analysis because their "safety" features didn't distinguish it as a defensive measure.
So they had to fall back to an open model (GLM 5.2).
In short, attackers have unshackled AI at their disposal and if you don't have your own open models running you
the long tail (Score:2)
As the article says, "Much of what the world needs probably does not require the absolute frontier."
The frontier models are great, but I find that the ones a release or two behind are also pretty darn good. The 'open source' models do fine in many cases and are much cheaper. In a few months the second and third tier models of will have caught up to the level where the best ones are now. So I'm not very worried that there will be some kind of monopoly on AI, the competition seems healthy at this point.