You don't seem to be thinking very logically.

What is more likely to be hit by a piece of space junk or micro meteorite, a space station sitting there for years, or a transport vessel just making a quick run up or down ?!

You also seem not to realize the kinetic energy and destructive power of these impacts - some of the most powerful weapons on earth are rail-guns using nothing other than kinetic energy to punch though multiple layers of steel, maybe pass straight through a tank.

FWIW the windows on this Chinese craft are triple layer, and the impact only cracked the outer one, so it must have been an extremely small object which is why they wouldn't have picked it up on radar and maneuvered out of the way which is normal practice.

What a weird question !

Why would it be different to any other modern spacecraft, capable of both automatic and manual control ?!

The Chinese aren't exactly a newcomer to space - they have their own space station, are currently exploring Mars with their own rover, and look likely to beat the USA to building a moon station unless we get our act together.

You see to be imagining these "taikonauts" in some primitive Chop Suey powered craft built using 1950's technology.

A TPU isn't a drop-in replacement for an Nvidia GPU or CUDA workload.

Google can build it's own Gemini models to use the TPU, and then any Gemini use will be on TPU, but if GCP is serving some other model then they also need to provide whatever that model is coded for, most likely CUDA/Nvidia.

So, the percentage of GCP AI load running on TPU is whatever precentage is using Gemini, or another Google model, as opposed to something else.

$30B is also nothing compared to what private companies are already putting into AI/ML, although there is little value in the government building yet another LLM, which is the compute-gobbler.

I wonder how this would even meant to work - if the US government made some, say, fusion or biotech breakthrough via AI/ML, spending taxpayer money, then which private companies are they going to share it with?

An untrained transformer is like a CPU. A trained transformer is like a CPU plus a program (the weights) that run on it.

You could say that the functioning of any computer program is mostly down to the CPU, but while somewhat true, that's a useless statement and fails to understand that it's the program where the logic is.

In the exact same way, saying that the functioning of a trained transformer (LLM) is down to the transformer, is somewhat true, but fails to capture what is really going on in exactly the same way as the CPU example. It's really the program running on the transformer, the weights, that is determining what it does.

Exactly .. the whole premise that you can accurately guess calories from a photo of a plate of food is bogus.

Of course in simple cases of easily recognized food, that don't vary too much, you may be able to come close, but even for something like a hotdog and fries, how big is the dog, how heavy is the portion of fries?

How is a photo going to tell you whether those carrots are just loaded with butter, or also a massive amount of sugar (an Anthony Bourdain pro-tip for delicious restaurant type carrots).

Just because you CAN ask an LLM anything, doesn't mean that you SHOULD.

He's trying, maybe badly, to get ahead of the story and control the narrative.

The fact that he's willing to admit, so soon after it's public release, that Google's Gemini 3.0 is better than their own GPT 5.1 is a bit surprising though.

Having played with "Gemini 3.0 Pro Thinkiing" at the surprisingly generous usage limits of the free tier, I have to say, even as an LLM cynic that initial impression is very good.

Yeah, this disappointing.

I was hoping it'd be delicious gooey cheese on the inside - a bit like a Brie perhaps.

Yeah - only 5% of ChatGPT users pay for it.

OpenAI's focus, and revenue, is mostly from ChatGPT, as opposed to Anthropic's being from API use (coding + business use).

API use seems more scalable than interactive ChatGPT use, which is limited by number of humans on the planet, with users willing to pay evidentially being a tiny percentage of that. OpenAI already has 800M weekly users, so you are not going to see a 10x increase than that unless they are selling to aliens.

API use has almost no limit since it's driven by computer use, not humans, and other than trial accounts, almost all API usage is paid for.

Most of the talent at OpenAI ends up leaving... typically for Anthropic. Of course if you pay enough you'll get employees, but it seems that's more despite the management rather than because of it.

AFAIK there's only 3/11 of the original founders still there (Altman, Brockman & Wojciech Zaremba).

Well, no .. that's not what he meant, even if those are some of the outcomes of the spread of AI.

Huang is clearly trying to boost Nvidia's, and his own, fortunes, by appealing to this supposed AI race with China, to get US policy changed to his own benefit.

He's obviously positioning winning of this "race" by the US as a desirable outcome, so going to be focusing on whatever aspects are seen as a positive by the US lawmakers he is trying to influence.

Actually, NVidia NOT selling to China (either because they don't want to buy - which is what they are now saying, or because of US restrictions) is going to ACCELERATE Chinese AI chip development out of necessity (even if imposed by Chinese government - maybe strategically accelerate to self-reliance).

The Chinese power advantage is real - huge amounts of hydro-electric, as well as solar. Trump is shooting US in the foot here by hampering Solar, and the AI co's themselves are talking about putting AI datacenters in space(!) which doesn't sound exactly cost competitive!

> commander Chen Dong and crewmates Chen Zhongrui and Wang Jie

Please stop making jokes - this is a serious matter.

Commander Dong and his crewmates Me Hungrui and Wang Pee just want to get home for obvious reasons.

It reminds me of the plane crew Sum Ting Wong, Wi Tu Lo, Ho Lee Fuk and Bang Ding Ow. :)

https://www.youtube.com/watch?...

Yeah - SK hynix 3D NAND is now over 300 layers

Apparently there are teeth from the Nanotyrannus embedded in the spine of the Tricerotops, and the former has a cracked skull.

https://www.theguardian.com/sc...