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.

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.

What happens when Big Brother realizes that you're actually working 12 hour days, but only getting paid for 8 hours, which puts them in violation of state labor laws? I was working far more hours than I was getting paid for while I was at Meta, but that's mostly 'cause I was working 425 miles away from my home and didn't really have anything else to do.

George Orwell was only slightly off about the date. It's a trope that SciFi always gets the date wrong.

Is 30 minutes enough time to have a good wank?

If you evaporate the oceans, the water will eventually rain down somewhere.

Drones are the future of modern warfare. It doesn't seem like "neocon war propaganda" to point that out. Although I would prefer there not be any war at all, we still have madmen like Putin, Netanyahu, and Trump that go around intentionally starting wars. Pointing out that those madmen intentionally start wars isn't neocon propaganda either, it's a simple statement of fact.

Trump is on a revenge tour against everybody and anything he hallucinates has disrespected him, and apparently he hallucinates the environment has disrespected him. Honestly, that's the only logical explanation for his actions.

Wouldn't it have been more efficient to not pay a company to NOT build wind turbines on the land it had already payed to lease for wind turbines? Trump will literally waste taxpayer's money on his revenge tour against the environment. His actions are indistinguishable from the actions of a mad scientist deliberately trying to harm the planet.

Trump's "F the environment" policies are creating MORE ocean, not less ocean! Which means it will be really ironic when Mar-a-largo is finally literally underwater, won't it?

So they know damned well that a huge facet of technology is consistent, deterministic interfaces and behavior. Sure, there's a demand for AI to also provide more flexible interaction in some scenarios which is desired, but many scenarios are benefitted by the dependable interaction of apps.

So on the face of it, the concept of a platform that is a strict subset of the capabilities of platforms already out there (the "app" devices are plenty capable of running the "AI" interfaces) is dumb. This has proven out with the Humane Pin and Rabbit R1, dumb devices that did nothing better and simply signaled affinity with AI.

But Microsoft isn't quite as stupid, so what's the real point... well...

That's really the goal, to reset expectations from having capable personal devices with generally perpetual entitlement to renting every single thing you do from Microsoft. You can have your 'PC' run those deterministic applications you are used to, but it's just going to be a virtual desktop to Azure now.

You gotta train them to not use their teeth.