The Powerful Chinese AI Model Experts Warned About Is Here
The “best” model is not always the “best” model for you. Why I say this, is because Z.ai's model is still outperformed in some dimensions by proprietary models from Anthropic for example. But can you afford to actually use that top end “best of the best” model?
Very usually you cannot afford them. Now consider that the new model for Z.ai is only a few percentage points (depending on what benchmarks you look at) behind Anthropic's Fable 5, but the cost is only 25% of the cost. You end up with a situation, that what you can afford through Z.ai is way better than you'll get from Anthropic or OpenAI (given like for like payment).
I have not done in-depth testing between Claude Pro and Z.ai, but I have thrown some similar challenges to both, and I got pretty good answers from them both. The difference is I'm on the free tier of Z.ai yest I have some access to their current flagship GLM-5.2 model (5.3 is only being released in a week or two), whilst on the Pro level paid Claude subscription, I'm locked out from using their flagship Fable 5 model unless I pay usage credits over and above my subscription.
It also does not seem that the proprietary US based models are fully private either. Someone was recently prosecuted after their AI model chat history was produced as evidence. We have also seen that some API keys and other information got recycled as output to others. The advice has been to not feed private notes and documents with API keys and secrets to cloud based models. All I'm saying here is that privacy is also not some feature separating proprietary AI Models from open weight models.
The very nature of how AI models are trained, means they have all “stolen” information from various sources across the Internet. None of them can claim to be fully self-taught without any hint of theft.
There is a lot depending on the future of AI, some say that can be as much as a country's entire economy being at stake. So lots of money and high stakes, as well as the weaponisation of AI, means this is becoming a lot like politics. Who can you believe is telling the truth, who is trying to sell us vapourware, who could be spying on us through AI, who could pull the rug out from under us, etc?
Either ways though, we should be thankful for two things:
1. There is right now no “Microsoft” of AI totally dominating the market, so we are seeing healthy and credible competition.
2. There is certainly the ability to self-host AI, not to the level of the big cloud providers, but certainly fit for most purposes and fully private.
But given that, it is a difficult call to make, and to decide decisively which path to take. Where do you bet your money, as a lot of customisation and work goes into adopting whichever technology you choose? Or do you take the slower more focussed middle road of trying to enable some in-house hosting and skills development. In the latter case, you build incrementally upon what you do, and there is no sudden rug pulling out from under you.
Personally, I've always been a proponent of the latter path, rather being in more control of what is built and ensuring absolute privacy. I've viewed Big Tech cloud services as land grabs and lock-ins. Yes they're convenient, they're quick, and they cost less in the short term, but that turns against you later on as your data is no longer under your control, getting it out later is very expensive and not easy, and the vendor lock-in and cost increases are very real.
Which is why my own cloud subscription is only paid month-to-month. I'm ready to switch or go fully self-hosted once I can get a more mature setup. Businesses don't always have that luxury though.
What I would bet on though, is that AI is here to stay. It is no passing fad. It is the same as moving from steam power to internal combustion engines. The world is moving on, and it is now just a question of which wave we ride, and how it helps leverage whatever we do. We're not going every going back to an age of no cellphones, no Internet, etc, much as I may sometimes wish we could.
Yet, the more digitally instant and interconnected everything becomes, I realise too how very little we own in that world, and how it can be gone in the blink of an eye. I still fear this generation may lose the most of its heritage in the future millennia. Our innovation has outstripped our abilities to preserve the volumes of data we are creating daily. In the year 3030 or 4030, if humans are even still around, they may know more about Ancient Rome than from 2026. I'm not speaking here about Wikipedia, I'm speaking about my and your hard drives at home, and in hundreds of millions of other homes, all collecting documents, photos, memories... They're all going to become lost to civilisation within just 50 to 80 years. Clouds are also not the answer, as they do shut down eventually, and memorialised accounts just get lost.
So who knows, maybe our data being sucked up (stolen) by AI, means it somehow grows and prospers on deeply embedded inside of AI, as it gets “restolen” between different AI providers
See
The Powerful Chinese AI Model Experts Warned About Is Here
Z.ai’s latest AI model release could help companies secure their systems—or find its way into the hands of hackers.
#technology #AI #archiving
