To start, I want to try and reassure you that this blog post is as accessible as I could make it. I use the right terminology for tech people to understand, but I’ve peppered in some food analogies for non-tech people to understand. I hope that will help you understand this important topic, because I want more people to be informed about something that could severely impact your life.
I’ve written about a few different “conceptual” bubbles in AI that I see as significant: creative work, busy work, and privacy. But what about the bubble? The one that so many experts are predicting is about to pop (and have been predicting for the last 2–3 years)?
I don’t think they’re necessarily wrong, though I do wonder about the direct correlation between stock market fluctuations and normal people’s lives, which has seemingly become more disconnected in recent years. The American and EU markets are generally going up, with all-time highs reached in 2026, yet people’s ability to feed and house themselves, even in a rich country like Luxembourg, are the subject of many articles, political rallies, and even protests.
Obviously a crash is a much different beast than a fluctuation. We all remember the stories of the Great Depression. An American-triggered event, Europe’s economies were impacted severely due to their debts and loans that intertwined us with their downfall. Sounds similar to today. It’s not that globalization is bad, but being so dependent on a foreign country’s success does feel a little counterintuitive.
That’s part of the reason I’m pushing more and more of my clients and network towards Open Weight models.
Open Weight models are pretty mysterious for a lot of tech-friendly and non-tech people alike. If you’re not as AI-obsessed as I am, you might not know what these are, and if you have heard of them, your first thought when they’re mentioned might be something like “those are Chinese models” or “aren’t they open source?”
Neither of those statements are correct, but you shouldn’t be blamed for that. They’re not talked about much outside of places where AI is already deeply ingrained in ways of working and developing new technology.
So, let’s clear up some misinformation, and explain how they impact the bubble.
Where are Open Weight models from, and how are they different from Open Source?#
Open Weight models can be made anywhere, by anyone who can create their own LLM. All it means is that you let the “weights”, as in, the trained parameters, be known. This is not the original training data itself, nor does it include significant other levers used by AI labs to create a model of value.
Food analogy: It’s kind of like knowing the nutritional value of a meal, down to the number of calories, grams of protein, and sodium content, but not the recipe or the origin of the ingredients.
Open Source models, on the other hand, are few and far between, but give you a complete picture of how the model was made.
Food analogy: The entire recipe. Where the chicken was raised, and who plucked it. The name of the farmer who picked the carrots out of the ground, and his mood that day. I exaggerate, but I just want to illustrate the difference in transparency.
And both of these kinds of models, like any other kind of model, can come from anywhere. There are American, French, Chinese, and even Luxembourgish open weight models (though these ones are often based on other open weight models).
You can download and use an Open Weight or Open Source model and run it on any kind of device that has enough RAM to handle it. And there are ways to compress or “quantize” models to make them fit into smaller spaces (though you’ll lose some speed).
Food analogy: Some models are like a banquet, and some are like a meal for one. You know how much you, and your kitchen, can handle. Some people don’t mind cooking a banquet in a tiny apartment, even if it takes longer than doing it in a larger space.
How do open weight models impact the AI bubble?#
Anthropic, OpenAI, X, and even Mistral have all made a big bet: that they are going to be able to rent our intelligence back to us, using their models.
Basically, they thought they were going to own all the restaurants, and none of us would be able to cook at home or a friend’s house. Or that we might not even be all that hungry at all.
I’m not sure they bet on Open Weight models taking off the way they have. I’m certain that the prevalence of high-performance Open Weight models from government-backed labs in China is not a coincidence, either.
I am not saying there is anything purely malicious about this, but I also think the Chinese government was more cautious about their AI bubble, which popped recently, and wasn’t actually that big a deal when compared to what the American version will likely entail. China leverages Open Weights more than the West, and I believe it is to prevent their Closed Weight models from being too popular, and they have a type of AI enthusiasm that we don’t share in the West. I would hypothesize (this is purely speculation from what I hear from people who are more personally familiar with their AI ecosystem) it’s because it’s seen more like a common good than something that was created to make a small group of people rich off the labour of all of humanity.
The problem in the West is how we have focused a lot on investing in very opaque, privately-held organizations to lead the charge. They took our data and our work, and they turned it into a, frankly, really neat piece of tech. Then they said that they’d use it to replace all of us, create an underclass, and basically just make the world a miserable place to be for anyone who isn’t them. Who could have imagined that people would be eager for alternatives, or to not use AI at all?
This means that those of us who are anti-AI (definitely not me, though I have my sympathies for this group) aren’t paying them the subscription they need to justify their expenses. I think what they really wanted were a lot of “light” users who would pay €20 a month, use maybe €5, and subsidize the heavier users. Instead, they got a bunch of crazy vibe coders (and some with decades of actual engineering skills) trying to see how they could maximize a €100 plan into giving them €4,000 of value. And that’s… a lot of their non-enterprise user base.
So… it’s hard to see how they’re going to make enough money to justify the investment they’ve received thus far.
So what’s next in AI?#
Now, those same AI enthusiasts on OpenAI and other frontier labs’ services are starting to realize that the rug might get pulled soon on their cheap inference, and are looking for alternatives before it’s too late:
- Open weights
- Local LLMs
- Mixture of Experts models
- Fine tuning (heck, even abliteration)
- Loop engineering
- Custom harnesses
- The Platonic Representation Hypothesis
If you’re not in this world, some of these terms may be entirely unfamiliar. But it’s kind of like how some of the best Michelin star restaurants are hole-in-the-wall joints. Some people just really know how to cook, and a lack of wealth isn’t going to stop them from building something truly great.
None of the additional “pieces” that fit around a model would be possible to tinker with in the way we do in this world without Open Weights. So all you need to understand is what they make possible: a world without tech billionaires. Maybe a world where, if you want to do something really simple like plan a birthday party for your kid, fill in a budget spreadsheet, file your taxes, or set up a small ecommerce shop (all use cases that save you tons of time using AI), you might be able to do it via your own computer, with your own ideas.
No Sam Alt-middle-man reading your data. No company trying to push a narrative on you (like when Grok tried to deny basic reality based on Elon Musk’s feelings getting hurt). Just an easy way for you to tell your computer, in plain language, what you need it to do.
It sounds like a utopia, and it is. There was never a chance that a good outcome from AI was possible when the base level goal was to make a few rich people immensely richer. However, if the goal becomes “putting privacy back in the control of normal people and small teams, while still benefiting from computers”, then I think this AI thing might be around to stay for a while.
As long as Open Weight Models are allowed to continue to be used, of course.
Want to start using Open Weights at your organization, but don’t know where to start? Get in touch.
