The best preparation to backstop the current mass extinction is AI, it doesn't need a biosphere. We're just playing for time.

Debt to GDP, debt rarely goes down.

If you make the gate long you can always have an antenna near perpendicular with the RFID tag (antennas above or below too).

Nah, it's all the wastewater from HF steps.

All AI is just mechanical turk after all?

Supermarkets need ultracheap printable NFC tags which can be individually distinguished when you walk through a gate, then you just need to check if people aren't removing the labels or using Faraday shielded bags.

Someone has to make a logical composition and clean up the AI bullshit. There isn't enough training data in the world to fill all the "corner" cases for AI to hallucinate bullshit into. It's "just" a productivity enhancer.

Once AGI is here, we'll have different things to worry about.

They are building industry knowhow how to use something other than CUDA too. You need to get the companies to ignore the wailing of the lazy programmers.

No one else has the matching incentives and long term income to build brand in the same way in the age of ecosystem. Google being build on advertising and way too loose licensing for Android couldn't build a high end ecosystem if they tried at this point and they clearly aren't trying. Microsoft isn't even playing.

Hyping everything just because it's from Apple is a little silly, only the LLM in a Flash paper deserved it (though it's mostly a variation of the theme in "Low-Rank Approximations for Conditional Feedforward Computation in Deep Neural Networks").

Making a digital assistant work better with random app/websites is nice, but outperforming GPT-4 on something it was never designed for is hardly impressive. The whole one-shot, unlabeled data spiel from the early GPT era was just hype, multimodal doesn't make the difference ... good labeling and supervised training/finetuning makes the difference.

Were there any genetic tests done to see if he actually did anything but plain IVF? I have my doubts.

Thin film PV has a lot of room to get material consumption per m2 down. Panels which you have to pump water and gasses through though, not so much.

Just keep the synthesis centralised and the light collection distributed through PV if you want to use solar.

She's a politician.

It might be true, but what use is harping about it? Some things are better left unsaid in the public sphere, leave it to academia.

In politics I view discourse about disinformation as a run up to censorship.

On the other hand, the method can be directly used for reinforcement training and it only has to look good. You can probably even train a classifier to recognise chatbot arena queries well enough so it doesn't pollute functioning in other benchmarks.

I suspect Claude is close to an inflection point where models are reinforcement trained specifically for the public benchmarks du jour. Was GPT4 to the same level? Maybe, maybe not. Everyone in the industry is going to have to start doing it now though.

So what do you need the NPU for if you are going to use bing chat any way?