... someone yelled "Bingo!"

Thanks for the insightful post. Yeah, if this was a practice test for our society on how to handle AGI, I agree we failed it.

As shown by the several of the AI company efforts (including OpenAI transforming into a for-profit), our current socio-economic system with its incentives to race ahead competitively regardless of the risks to society (so, privatizing gains, while socializing costs and risks) may ultimately just be incompatible with ever-more-high technology.

As Bucky Fuller wrote: "Whether it is to be Utopia or Oblivion will be a touch-and-go relay race right up to the final moment.... Humanity is in 'final exam' as to whether or not it qualifies for continuance in Universe."

I think we only have a chance of passing such a test -- whether it is about AGI, nuclear energy, nanotech, biotech, or even just plain old networked computing used by sprawling bureaucracies -- if we appreciate the humorous irony mentioned in my sig: :-)
"The biggest challenge of the 21st century is the irony of technologies of abundance in the hands of those still thinking in terms of scarcity."

Even that might not be enough -- but it is the main hope I have to offer.

I think the PP understands well enough. Why can't data centers use a closed water loop to move heat and then a water to air radiator like a car does. Sure, it would be huge. But evaporative cooling has been designed around a water supply being free or really cheap. It isn't.

So let's all piss in Google's data center intake.

It's like the tires on my truck. Not terribly worn, but well over their expiration date. Nobody will touch them, even to check air pressure, because of the liability (and federal law). One could blow out on a weak spot.

I don't even want to know what a blowout at 17,200 mph will be like.

So my house has EV and gas vehicles, and by far taking the gas vehicle to a gas station is way more annoying than plugging it in at home. Not to mention the maintenance (oil change? nope air filter change? nope brakes? just like hybrids the brakes barely get used). No random surprises like "oh great, some fluid on the ground, smoke is coming out the exhaust.

The battery is the big one, and the tendency to be heavier means faster tire wear, but the plugging in for a lot of people is a *plus* not a minus. Meanwhile the gas car has a few hundred here, a couple thousand there, and by the time I might have to replace the battery, I am pretty sure I have racked up equivalent costs over a much more annoying series of repairs and maintenance.

Tip, as soon as you say "copilot says", no one will take you seriously.

Yes, an *engine* can last a while, but that's far from the full picture.

Headgasket will likely not last that long, and while *technically* the engine "lasts" through that repair, that was an expensive repair.

Timing belt won't last that long, and that's an expensive maintenance item.

The transmissions for ICE wear more than EV, and those tend to have a relatively shorter life than cited, and that also is pretty pricey.

Further ignores things like oil changes constantly incurring expenses. Brakes on EVs and hybrids last way longer. If you can charge at home, your fuel savings are huge (public charging, maybe not so much).

Yes, when the battery goes, it will be expensive, but in that same time you probably have otherwise accumulated a comparable cumulative maintenance and repair bill for petrol-exclusive issues.

As they said, 8x code output is a flawed metric. By volume they are getting in.

One could even reasonably argue that they tend to be good at catching critical little details that are difficult for humans, like the consequences of a shallow copy buried in a sea of code creating a security disaster. A "needle in a haystack" scenario, where AI does comparatively well with the relentless attention span.

But in other ways, they are verbose messes, and will toss a whole lot of pull requests. Each of those pull requests may be a lot more volume than needed. For example, I got a CodeGen pull request for "close a gap between two elements after updating the UI framework to new version". Yes, real issue, and.. well, the referenced gap *did* close, but with a lot of dubious side effects. The issue called for a single CSS rule to be tweaked. It instead was hundreds of lines of CSS, sometimes verbatim repeated 4 times (the originator said he had to ask the prompt multiple times because it failed the first few times, it seemed Claude thought maybe adding the same rules it already added might have helped). Others might have just run the code saw the gap closed and accepted it, despite the baggage of 99% of the lines changed having no particular intent behind it.

This is in a normal where a lot of developers think even simple things need to be complex. Hello world needs to run in Azure Pipelines using Kubernetes, Helm, and Ansible spawning at least 6 microservices. They think they need all of that and yet it's unmanageably convoluted, so the AI kind of lets them have all that superfluous complexity without actually managing it.

Thanks to work, I'm using it all the time.

It is true that even opus 4.8 cannot always bring a mundane code to life credibly. It can depending on what we are taking about, but the sorts of things it fails at are well short of the complexity of what it would mean to rewrite itself in any credible way.

More importantly: Capitalism is an ECONOMY and market system. It is NOT a blueprint for a society. You can run your commerce and trade as capitalism, when you run your SOCIETY along capitalism principles you end up... essentially with the USA.

This is the part that is constantly forgotten. As a society, we have values that are not represented well within capitalism. But for some reason, we dumb shits think that we can treat everything as a market and apply capitalism to it and that will magically solve problems. But in education, just as one random example, the goal of it all is educated adults as output. It is not maximizing profit. Same for the prison system, the healthcare system and two dozen others.

It's greed, pure and simple.

Making a good product is possible. KEEPING making good products for decades is hard. Even more importantly: You will have hits and misses. Which, for a quarterly-result-bonus oriented manager is a no-no. Subscription models mean plannable revenue streams. Then all you need to do is negotiate your bonus package so that the already existing subscriptions will provide and you're home free and can already order your 2nd yacht.

Muah hahaha!

They are finding a plateau with where the LLMs can go and could use the narrative of a "pause" to explain why capabilities are going to iterate in a more 'evolutionary' way instead of the revolutionary way folks are expecting.

There isn't to my knowledge a mechanism for the models to "self-improve", whatever one may think, at least the output doesn't have access to change the model in any way. The narrative of "oops the AI started evolving itself on accident" doesn't have a way to happen.

Considering that even the vaunted Opus 4.8 can't always develop mundane traditional software beginning, it's hard to imagine it could rework the model itself even if it had such access.

That's why regulators have to be ever vigilant. Anything (within reason) that companies do which is "inward looking" is OK. The moment they address competitors in any way, that needs to be banned.

Example: A certain evil software vendor offers customers per-seat volume discounts for their product. OK. When the threshold for those discounts is coincidentally set to exactly the number of employees in a customer's facility, not OK.

I don't need self driving. The terrain here is such that autonomous systems smart enough to self stabilize would be outlandishly e pensive. Jus give me a Ferrari.