Timecube... Although I'm not sure if that's what you were thinking of.

The advertisers still pay, they just don't get conversions.

Having written the AI systems for Robocode bots 15 years ago, they perform at such a higher level than humans that there is no way, given all the time in the world, a human can compete with a full statistical targeting and movement system. We just don't think in that way.

Hoe much dedicated cache do these NPUs have? Because it's easy enough to saturate the memory bandwidth using the CPU for compute, never mind the GPU. Adding dark silicon for some special operations isn't going to make out memory bandwidth faster.

Does a cache help with inference workloads anyway?

I don't know much about it but my mental model is that for transformers you need random access to billions of parameters.

It's streaming access, and no not as far as I'm aware. APUs have always been hilariously bottlenecked on memory bandwidth as soon as your task actually needed to pull in data. The only exception I know of is the PS5 because it uses GDDR instead of desktop memory.

Are we going to see more memory channels for consumer desktop at some point from AMD or Intel? Apple seems to be the only one that offers it.

Am6 Socket isn't due for a couple of years. The current sTR5 used by Threadripper processors supports up to 8 channels of memory.

I'm not sure what you mean - I think the mobile 300 series can do quad channel already for its APU at least. I'd assume it can do more but do you not need more slots beyond that?

There are a single-digit number of products using the AMD Strix Halo mobile parts that have a 256-bit memory bus. All other mobile x86 processors (including AMD's mainstream Mobile silicon used for these desktop processors) have the usual 128-bit bus.

I'm talking about desktop computers, not mobile.

NPUs are more useful for prefill than decode anyway. Memory bandwidth is not the bottleneck for prefill.

When I interview people for a job I'm not looking to hire an average programmer, though.

You want to hire an above-average programmer, I assume. Do you also pay above-average salaries?

People are not uniform fungible programmers. I want someone who is a specialist and above average at the stuff I'm trying to do. If that specialization commands a higher market rate than average, then yes, that's what they get paid.

As an individual you can. Not as a corporation.

Again, I'm asking about generalized freedom of association. I get the hate for corporations, I do, but if you ban freedom of people to associate, and through that association, to pool money for a common cause: are you just straight up good with that?

The underlying law in Citizen's United didn't just target corporations: it targeted unions, it targeted non-profits. That's why I'm asking about those other cases that would be banned. I understand wanting to ban Evil Inc., but are you ready to ban World Peace Non-Profit, too? Not only that, but Citizens United itself wasn't a corporation.

If you have a defensible case, that, yes, that'd be fine, go ahead. But when only individuals can spend money on politics, are we not in the same situation, where Bezos or Musk can afford an ad, but I cannot?

Convince me that the solution is not to remove money from the process entirely.

I’m just spitballing here but why don’t we have a system where every candidate gets the same allocation of money for ads and it is outlawed to use your own fortune to bankroll your candidacy

In a world where social media companies are all run by billionaires, let them pollute the feeds with their candidate of choice. This only speedruns the complete collapse of SM in an era where we can no longer distinguish real discourse from AI

And good riddance.

Workaround: I'm rich and would just start up a newspaper that gave my candidates good coverage.

In my opinion, this isn't a easy problem if you value freedom of speech, freedom of association, and freedom of people to spend their own money.

> Workaround: I'm rich and would just start up a newspaper that gave my candidates good coverage.

Or buy Twitter.

Software that uses less RAM isn't necessarily worse, often RAM is wasted purely due to carelessness and because it didn't matter.

Your assertion that a world of lean software won't return is backwards looking; that was all driven by hardware being cheaper than developer effort.

If we now enter a world of AI-enhanced developer effort being cheaper than hardware, perhaps we can have lean efficient software again.

The fact that this is so obvious to humans is why there's no training data that LLMs can use to know the answer.

Even without AI, many (most?) orgs are held back by internal processes and politics, not development speed.