I'm old. Like, well, really REALLY old. And you would think being essentially immortal would come with perks like being able to see, but having effective ciliary muscles and them being completely dead are two different things.

I watched the preso and my first thought was "Jesus (... thank you again sir), Apple keeps making better phones for old people!" This seems like one such phone, although I don't see myself opening my wallet for one until my 15 pro max is truly utterly dead (heh), and I know for sure that the hinge / crease is going to hold up.

Just a couple of random notes about the practicality of this.

• Some new inference data centers don't have gens. The idea is that with inference if a resource is not available it can be requested from another pool (availability zone). In these cases when the power goes out, the UPSs hold the load for 4-5 minutes to give the servers time to drop and then the site goes dark.
• Imagine 32 diesel freight train engines all powering up. That's what this is like for a site-wide utility power failure for a smallish data center. Thankfully most DCs now require an enclosure for each gen which limits the noise they produce while running.
• Most providers have between 24 and 48 hours of diesel fuel on-site. This is either in large central storage tanks or in belly tanks dedicated to each gen. A provider would have typically a 4-hour SLA on fuel delivery so once the gens start running a call is made to start running fuel trucks to the DC to begin the refueling process.

It's probably a way to prevent model collapse. So they don't train LLMs on data that has been generated by LLMs.

It is more likely that Canada has 10 new reactors on-line in 14 years than Slashdot having Unicode support by then.

In the data centre business this newly-invented Megapod(tm) is called a module. My company probably has 30 or more of them, and I've seen this "data centre in a can" concept around for the past 15 years. Doing it in shipping containers was all the rage for a while.

Of course the old ones were not direct-to-chip liquid cooled, but I'm sure the providers of these modules already have that. In any case, this is in no way a new concept -- just, as you say, some shiny branding.

Self-Fulfilling Prophecy is (or at least use to be) well known in teaching circles. That is, if you call out a child for being a certain way they will often change their behaviour to make that come true, whether positive or negative. It's interesting that the same thing seems true for AI models.

Evaporative chillers are not necessary to cool a data center, certainly not in Michigan. We chill data centers in Arizona with no water usage at all. The data center designers / owners are just being cheap. Sure, if you can get the local municipality to give you water you can use that to lower your costs and increase your efficiency. But it's not necessary. All our data centers use 100% renewable power (if not available then we purchase credits), and we cool with air chillers, and despite these additional costs we're certainly not going bankrupt or being left with unsold capacity.

It's (as always) about the money. The fact that they are going ahead with the project anyway tells me that they will just switch to air chillers.

To directly respond to your comment (which is spot-on), a new facility being stood up for LANL is likely to be direct-to-chip liquid cooling. Generally we don't do full immersion because of the costs and complexity (a modern AI 52U rack is pushing 5000lbs now and fully immersing it will put additional structural strain on the slab floor), but the technology to distribute chilled water from the facility through CDUs (coolant distribution units) to manifolds in the racks and then directly to the chips needing to be cooled is finally getting mature.

About 40 years ago a friend of mine stated that this problem had been solved and that slow motion photography proved that the cat used its tail to right itself. I actually had a tailless cat nearby (lost it in an accident with a car) which I retrieved, held out in front of him about two feet off the ground, and let go. It landed on its feet.

What's curious is how long this has been "studied" and how many different theories there have been!

Trump is looking for an AI company that is headed by someone who either has far too much confidence, or far too little regard for human life.

Surprise! That ended up being Sam Altman.

I own the intellectual property for the software by the name of Amiga Flow from the '80s. So both party's argument about prior art or use can get stuffed.

You can find the source code to Amiga Flow here.

Wankers.

I can't speak to all of the areas that contribute to AI backlog (like capital allocation, systems integration, networking availability, etc). But from a data centre standpoint it is a real struggle. The general timeline to get from requirements to signature on a data centre lease is about three months if all goes well (assuming you are not self-performing). Once that is signed a DC project takes about two years from conception to RFS (ready-for-service) where you can start rolling in racks. A LOT goes into that timeframe: Ordering LLE (Long-Lead Equipment (generators have a 24+ month lead time right now)), securing permits, securing power, securing a EOR (Engineer of Record) and getting the design done, land preparation, construction crews and materials, etc.

And that is all just to get a "powered shell": what we call a building with at least two diverse sources of utility power, gens, transformers, power switch gear, static UPSs (if they are being used), chillers, pumps and piping, etc. You still can't roll in racks until you have a fit-out design (PDUs, RPPs, whips, busway, FWUs (Fan Wall Units) or CRAHs (Computer Room Air Handlers), temp and humidity sensors, etc. It takes about a YEAR from the time you get the fit-out design to the time when you can roll in racks (but it is possible to line up fit-out works with RFS if you get the design early enough).

Everybody is looking for some creative solutions to speed things up, but data centre deployments are complex (I deal with well over 1000 individual requirements per project) and there are a lot of supply chain constraints. Most of the hyperscalers are juggling 20+ simultaneous projects, it's burning people out and there is a lot of churn in the business (making it worse).

The Trump administration just pushed out the official who's unit banned Chinese vehicles.

"Jan 23 (Reuters) - The Trump administration has pushed out a Commerce Department official whose office effectively barred nearly all Chinese cars from the U.S. market for national security reasons, according to people familiar with the matter."

"The sources said that, had she not resigned, Cannon would have been reassigned, and that the new administration plans to put a political appointee in the post. Her last day is expected to be February 20, two people said."

Why this is happening is anyone's guess.

I think this may be the first time I've seen a dupe immediately following the duplicated story. Unless there is something remarkable about Trump announcing something before the company involved gets a chance to comment on it.

It's almost as if we shouldn't have included "intelligence" in the actual fucking name. But once again our language has been co-opted by marketing BS and now here we are trying to set the record straight so people aren't confused or deceived.

The most useful feature for the same people who use text messaging and emergency services when they are outside of a cell coverage area is a local weather forecast. Something that the Garmin In-Reach family of devices do, and really the only thing an avid hiker / paddler / outdoors person is missing from the Apple service that can be a matter of life or death.