If you can put off maintenance for thousands of years it's not really maintenance, it's toil.

Between 1958 and 2026 we've added 37 leap second. That's 6616 years before we'd drift a full hour at the rate we've been drifting since 1958. We're currently looking at removing one of those 37 leap seconds. So really it's even longer than 6616 years. If somehow the pace of leap seconds doubles we're still OK for well over a thousand years.

Bitcoin booms and crashes on vibes and market manipulation. https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fen.wikipedia.org%2Fwiki%2F...

The big crypto bubble will pop eventually... and if it's allowed to get be commingled into other financial institutions it's going to hurt all of us.

Sounds like your friend likes filters all iPhones have the option to shoot without post-processing; by default there is minimal processing, and people can do whatever they want and produce oversaturated AI slop. Most Android phones also have ways to disable various levels of camera post-processing but what’s available is different depending on the exact model of Android.

I like having auto-HDR, dark mode, and basic color warmth changes, but nothing else.

A transistor is still the fundamental basic unit of all logic for chips. You can put transistors together to make different things like high-speed cache memory with flip-flops as well as all types of logic gates.

DRAM memory is a bit different, only the high-speed cache(SRAM) is built from transistors. DRAM uses a capacitor to store a bit and is WAY WAY slower than SRAM which just keeps passing a bit back/forth between multiple transistors to store it.

I think your figures are way off, but the sentiment isn't wholly wrong. The US's debt is bigger than China's by most measurable dimensions. The total debts for each country are in the same ballpark.

China's debt has surged 40X in the last 20 years where China's GDP has gone up by 7X in the same time. In the last 20 years the US debt climbed 2.5X where GDP went up 2.1X.

It will be interesting to see what happens over the next 20 years... but neither the US nor China can keep just doing what they are doing and expect to not have the wheels fall off their country.

The order from the government and Anthropic's previous issues with Trump's government are quite factual. In the absence of an believable explanation for why the government has tried to place extraordinary restrictions on the Anthropic another time I think this article sums up the situation which does involve conjecture fairly well.

Microsoft tried to embrace/extend/extinguish consoles and failed . . . spectacularly.

Adyen and Stripe are both rather expensive with their normal rates... Is this just a euronationalism thing? or a way for some politician to get kickbacks? or what?

Macs... Macintosh laptops are basically the only laptops with enough RAM, memory bandwidth, and processors.

It couldn't go worse. . . right? https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.jeffgeerling.com%2Fb...

The demand for DDR5 will evaporate in 2 years whether the AI bubble pops or not. IF new factories are coming online they need to be sized for new tech, but only LPDDR6 has a ratified spec and it's not ideal for AI. So ramping up lot of extra factories for LPDDR6 makes little sense

Sandisk has been devloping an HBM-killer memory called High-Bacndwidth-Flash... but they aren't sharing much information. I'd say this announcement either means High-Bacndwidth-Flash isn't panning out or that they won't need new factories to produce it.

As far as what others are doing that makes investing in DDR5 dumb:
1. Cerebras doesn't use memory chips, just a crazy crazy amount of on-CPU cache.
2. Intel is re-entering the memory market with a sorta radical new ZAM memory that is considered an HBM-killer.
3. Turboquant is fairly useless at actually speeding things up currently, but it proves that we can shrink the need for memory if compute can handle this type of quantization in-line. In the near-term this will mean accelerators coming out in the next few years will have these schemes baked into the hardware and need far less RAM. This is roughly comparable to when we started getting hardware accelerated texture compression with 3D graphics cards. It looks like the Turboquant idea was plagiarized and overhyped, but the idea is likely amazing if applied at the CPU level.

Can't tell if this is sarcasm...

• On the one hand... "dynamic innovation" reads fairly sarcastic
  • On another hand... some people really have a raging hard on for capitalism
    • On yet another hand... the giant loopholes in this law mean people Maryland might be the first place businesses look to implement surveillance pricing.

Palantir really only has value as a middleman between various AI backends and the defense industry. Currently Palantir pretty much only uses Anthropic. Anthropic is trying to cut them out by working with the DoD directly. I see this manifesto crap mostly as a way to try to keep Palantir in the news/relevant.

It's likely Palantir is behind Anthropic getting labelled as a security threat... and if they are it is backfiring hilariously.

Yep, Trump’s judges will likely just do the same thing they did with Kalshi. They could even use the same argument for all these; “This isn’t obvious thing explicitly supposed to be governed by the state it’s market swap trading”.

Hopefully we can correct this with the midterms ...

The most likely thing AI will do is create calcification of current programming languages and methodologies; especially with the death of StackOverflow and other living tomes of public programming knowledge. Without new data to train on, things will stagnate for LLMs... which I think will either make LLMs far less relevant for programming or, more likely, entrench languages(and methodologies) that currently have good training data.

Some programming languages have been FAR more successful at getting LLMs to generate reasonable code, which mostly comes down to data that is available for training LLMs. LLMs are generally really good at generating Python because of the amount of data available to train on.