ChatGPT itself told me that it works like that when I asked it about these problems: that the LLM is used as an interface to specialized modules.

The LLMs can't even count letters in a word, because it's not a text problem, and no one created a module for this basic task. LLMs themselves are very limited in their functionality.

When I tried to get it to draw electronic circuits it always spit out total nonsense. Supply voltage always connected to ground, and all inputs and outputs of an IC were always connected to the same pin.

The AI companies have hired physics and electronics experts in the meantime. They must have created modules for electronics and physics processing.
It makes no sense to use an LLM for that, I am sure the AIs have specialized neural networks for these tasks.

The current AI tools are Large Language Models (LLMs), pretty much just advanced chatbots. These are neural networks which process and generate texts. This ability is not really helpful for an "Artificial Engineer". This has to be a different kind of neural network, one that processes formulas, design ideas and numbers. Also a simulation module certainly helps.

The current AIs can do more than just processing words, because the text generator is used as an interface to other software- and ai modules. E.g. ChatGPT seems to have a chess module since a few months. Before that it looked like what a text processor can do: copy moves from other games which look similar, meaning it creates pieces out of thin air and does illegal moves pretty often.

(I guess we use different units here)

Google uses its own Tensor Processing Units for AI. ChatGPT also already transitions from Nvidia GPUs to custom in-house designed chips. Anthropic is exploring to design its own custom chips.

These are the first google results when I search for the AI chips of these companies. Then there are also wafer-scale chips

I think Nvidia became big because Cuda is used so much. But ai companies run their own data centres, can use whatever chip design and send it themselves to TSMC. They have no need to pay Nvidia.

Scraping by the Internet Archive just happens in the same way as scraping by AI. Their automatic software probably just blocks any scraping.

Yes, that's a bit more complex and one has to be specific what they talk about.

What I mean is, the contribution of AI to the GDP lies in the construction of data centers, not in a productivity gain caused by AI. And I believe the calculations according to which the US had nearly no other economic growth. This could mean that it is just wasted money spent on servers because of a hype.

That's not about AI itself generating value. The money spent on new server farms increases the GDP.

You know Android has 72% market share and Apple only 28%?

Apple has so many phones in the top ten because there are much less choices for iOS phones..

Can you? I tried to delete the passwords which Microsoft Authenticator uploaded from my phone, because I am not using it. They don't let me.

That's a local problem. "The whole system can collapse" is nonsense because it is not one system.

Sorry, I just read that this is a persistent misconception, the mirror error did not come from reusing trhe KH-11 KENNEN equipment.

It wasn't really "defect". It was designed for earth observations, because the telescope was just the facade for the KH-11 KENNEN spy satellite program. Because of that the focal length was minimally off.