I gather that suntzu3000 would prefer it that companies should be able to do whatever the hell they want with no limits or restrictions
Right, because the only alternative to a nanny state is total anarchy.
You need to read better. 86% usage before the ban, 81% usage after the ban. 5% stopped - depending the margins of error on each survey.
Plus what percentage of the 5% said they stopped, but didn't?
Australia has always been a nanny state. Australian politicians think the average Australian is a moron that needs to be protected from themselves. Zero respect for the power of the free market. Anything that anyone considers a vice (or even just slightly bad for your health) is either banned or effectively banned through extreme taxation.
We should vote in Pauline Hansen / One Nation (Trump-style populists) to send a message that we are not going to tolerate it anymore. Hopefully the other parties will get the message.
why they're teaching these models to know how to make cocaine, nerve gas and bombs?
In order for an LLM to know how to create a helpful harmless AI agent, they need to know what the words "unhelpful" and "harmfull" mean.
We raise our small children on extremely sanitized data. The stories we tell them, the explanations in response to questions, even the facts we teach in early school all paint an ideal world. It's only after they've established an ethical approach and a sense of right and wrong do we start showing them the real world. We need to do this when training AI as well. Currently we throw all of it at them with little thought to establishing a base personality before expanding the data input. Of course we end up with a distorted mirror of the qualities of adult life; bots that threaten blackmail or encourage self-harm. You can't bolt ethics on after the thing has been trained on the complicated, often selfish behavior of real and fictional characters.
This is a very interesting thought, but you are misunderstanding what is going on in AI pretraining. It's not the AI agent actually reading the books. It's a simulator that learns to represent the world (like the holodeck in Star Trek). The AI agent is then summoned like an NPC in the simulator (like a character on the holodeck). See The Complex Universe Theory of AI Psychology.
Technological progress is always jagged: On average things get better, but a minority of things regress and get worse. I'm sure there is a whole list of non-obvious reasons why riding a horse was better than driving a car.
once the AI sloppers get squeezed out of more and more first-world countries they'll move to third-world ones
Did you just suggest that Australia is a third-world country? FYI The GDP per head in Australia is 75k USD, just shy of USA GDP per head at 94k USD.
It's sad to see these products with Linux on the back end and no Linux client.
I have codex write and maintain a Linux client for codex called Qodex. announcement github link. It's already updated it for 5.6 and I'm currently working with it.
Enjoy.
AI is very malluable. There are a lot of different ways to use it. Some ways work well, some don't, some haven't been thought of or tried yet. These sets are also changing with each new frontier advance. This is only the beginning.
The fact that a antiquated old company like Ford couldn't figure out how to use it properly yet is evidence of nothing.
Nvidia does mostly one thing well: Fast tensor multiplications. If you build a transformer-optimized architecture you can probably take a few shortcuts that for example minimize what needs to be moved around in memory.
The vast majority of the compute of executing (or training) a GPT-like model is tensor (ie multi-dimensional matrixes of floats) multiplications. GPU memory is used to read/write/store tensors for the operands and results. I don't see how a special purpose ASIC can do any better than an NVIDIA GPU can on this task, (other than being cheaper per chip at the expensive of being less general/flexible).
I was under the impression that ASICs were cheaper to produce while still doing a subset of tasks as well (or nearly as well) as a more general GPU, but that the power-performance ratio was about the same.
If that is so, I don't understand how its going to allow OpenAI to serve more users. Maybe the issue is that NVIDIA can't keep up with demand? So this is a way to expand the overall capacity of their data centers? By adding fleets of ASICs in addition to their fleets of NVIDIA GPUs? Not sure.
Well look on the bright side, they have to release them when the opensource Chinese models catch up to match capability, otherwise everyone will use those.
Yes, this is inevitable. GLM 5.2 was released this month and it is right on the heals of GPT 5.5 and not far behind Fable 5. DeepSeek, Kimi, Qwen, MiniMax are all only a few months behind too. There is no moat.
Is there any way they can ban the Chinese models short of blocking the whole internet?
The chinese models are open weight, which means anyone can download them and run them on their own hardware. They are all only each about 1 terabyte. There is no way governments can shut them down. China could shut down the labs and prevent progress, but once an open weight model is uploaded there is no way of removing it because copies of the weights have been taken from all over the world.
AI models are entirely deterministic computer applications.
No they are not deterministic. The tokens they generate are randomly sampled from a weighted propability distribution (what's called a "softmax" over the logits in their final output layer - as adjusted by temperature and some other sampling complexities).
I happen to know a lot about this, and am familiar with the state-of-the-art text-to-speech models, and for some reason, Google definately isn't using one. My guess is that they have so many youtube videos to transcribe each day that they just dont have the compute to use a state-of-the-art model to do it, and have to use a fast cheap one with a higher error rate. If that is the reason, then Google should provide the option to people posting videos to pay for having their video transcribed by a SOA model.
The biggest problem with Rust (as compared to say C++) is the lack of a formal specification. No spec helps in the beginning of programming language development, because you can move fast and break things - but if you want real investment in a language you need to have a stable and rigorous and versioned specification so that companies feel comfortable investing X billion dollars of IP in the language, knowing that its not going to fall apart on every compiler upgrade.
To downgrade the human mind is bad theology. - C. K. Chesterton
