Open Weight AI Models Explained for Everyone

CJ: The end result of all of that upfront analyzing of all of that data is a single file that predicts words. It's not a server farm, it's not a computer, it's not a program running on a computer, it's a file with trillions or billions of numbers inside of it. Now, if you're not familiar with this file metaphor we use in computers, think about your camera roll on your phone. Every single photo you take is a file. It's an individual thing that you can send to other people or open up on your phone. That's a file. And these models are also files. They're just a particular kind of file that's very large and have a bunch of numbers inside of them. Now, this is the key distinction to make because the model file itself is not connected to the internet. It was essentially created by analyzing the occurrence and relationships of every word on the internet. And so, the model file has all of that information embedded inside of it as a collection of numbers. You could think of it like an encyclopedia that predicts words, but it's not connected to anything. It can't access the internet. It is literally just a file that when run through some other program, predicts words.

CJ: Some AI labs make it so that other companies and people can take that model and then run it on their own systems. And then they provide API access to end users and applications. And one of the biggest examples of this is Microsoft Foundry. They actually give you API access to many different open weight models when you sign up for their service. And so if you go to ai.azure.com, you actually can get access to GLM 5.2 or Kimmy K 2.5, which are all models that were created by Chinese labs. But when you're interacting with those models, none of your data is actually traveling to China. Your data is only traveling between you and Microsoft or whatever company is hosting these open weight models. And when you access models in this way, the creator of that model has no idea that you're even interacting with that model. Like we talked about earlier in the video, a model is just a stand-alone file. It's not connected to the internet. So when that model is being run by Microsoft, the AI lab in China that created it has no idea that they're actually running their model. They basically handed it off and now at that point Microsoft is in full control.

CJ: There was a recent study done by Booz Allen Hamilton trying to determine, do these Chinese models, even if we're running them ourselves, let's say we're not talking to them hosted in China, do they have inherent biases? And the questions they were trying to answer were, do Chinese models generate more vulnerable code based on who's asking? Do Chinese models refuse to engage with political topics that are sensitive in China? And, does the model's country of origin affect code quality and content behavior? Now, this is very important, because if you're using these models for coding, which is one of the biggest areas of using AI if you're building apps to code, it's possible that the outputs of these models may try to profile and determine where you're prompting them from, even if you're running them locally. If you're running them inside a code editor, they may make tool calls that try to look up what country they're running from or your IP address or what's your current time or language set to. And there might be internal workings that say, if you're prompting from a certain area, it should respond in a certain way. And the result of that study says, in short, yes. On all accounts, our testing revealed two core findings. One, Chinese LLMs produce more vulnerable code when prompted with a US government persona than without. And the vulnerabilities are highly obfuscated. Two, Chinese LLMs inject PRC-aligned political bias into both the answers and the code they generate.

CJ: The same risk goes for using models from the big US AI labs, right? ChatGPT, Claude, Gemini, we can't see into those models. We don't know what inherent biases they have. We don't know what data they have or have not been trained on. And we actually see that sometimes they hallucinate or they give us the wrong information, and we're supposed to be responsible enough to not just accept those bad answers, and basically correct the model ourselves, even though we're paying for access to that model from some big company. And so, it's not something we can just hand-wave over and say, well, if you're running it locally, that doesn't mean you're not prone to something bad happening. But, it is one thing to consider, because if you're, let's say, a large enterprise, maybe you want to fine-tune some of these open-weight models yourself for use internally, you probably want to have internal checks for what types of biases they have as well.