I think the overall trend is now moving somewhat away from having the CPU and GPU on one die. Intel's been splitting things up into several chiplets for most of their recent generations of processors, AMD's desktop processors have been putting the iGPU on a different die than the CPU cores for both of the generations that have an iGPU, their high-end mobile part does the same, even NVIDIA has done it that way.

Where we still see monolithic SoCs as a single die is mostly smaller, low-power parts used in devices that wouldn't have the power budget for a discrete GPU. But as this article shows, sometimes those mobile parts get packaged for a desktop socket to fill a hole in the product line without designing an entirely new piece of silicon.

Which was rather infamously not used in the actual research but was gatekept and used to verify the work of the domestic scientists.

1. Same input = same output. This can be called determinism, and it's technically rather trivial to achieve in the lifetime of a single model snapshot - it's just a matter of business need, because you pay extra for worse batching. It's harder if you need to extend the guarantee into the future, as you need to keep the snapshot and inference method the same. It's also a relatively niche thing, only required for build reproducibility, supply chain security, this kind of stuff.

2. Zero error rate with arbitrary inputs and outputs. This is not determinism and it's also NOT achievable in any model at all because the domain LLMs (and humans!) operate in is fundamentally ambiguous. If you want to enforce the formal rules, verify your inputs and outputs formally! Trying to solve it purely with intelligence (human or machine) is a fool's errand. You can keep the error rate low enough, but you can't guarantee the absence of errors due to the nature of intelligence.

Yeah IIRC Yudkowski famously said something about a super intelligence could derive the theory of gravity correctly by seeing only three frames of a video depicting an apple falling from a tree. This is the same Less Wrong nonsense, rejecting how vital and irreplaceable experimentation is.

There's an infinite number of explanations for the location of an object in three equally time-spaced instances. Not to mention limitations of the measuring equipment itself.

- A lot of high-quality data - Some careful design - (Not always) some external knowledge to guide the solutions

And this is using specialized NNs for physics, where you often know underlying equations. Kind of crazy that some people are so delusional about that.

Thanks for putting it the way you did. I didn't knew it was meant be that way, but it sort of confirms my suspicion that people who use the term 'rational' and 'logic' loosely often to dismiss an opposing view never really seek experimental results before having a point of view.

It all comes down to two things, player interest and the range of expertise/amount of work and coordination required for your mod to fit the base game. For example STALKER SoC has a lot more story mods than, say, Skyrim, because SoC is pretty janky: it has few animations and voiceovers are reserved to key plot moments, so it doesn't take much to match the quality.

The question is not about safety then but about "does it do what I tell it to". If the AI has the responsibility "to be safe" and to deviate from your commands according to its "judgement", if your usage of it kills someone is the AI going to be tried in court? Or you? It's you. So the AI should do what you ask it instead of assuming, lest you be tried for murder because the AI thought that was the safest thing to do. That is way more worrisome than a murderer who would already be tried anyway deciding to use AI instead of a knife to kill someone.