Management has been pretty adamant that we are at the point where a developer should *never* look at source code, and work purely on prose. So some dutiful people have taken it to that extreme.

So they tell me they spend hours in planning, reviewing the resultant generated plan to give themselves confidence that the resultant code should make sense, then let it chew on making test cases, making code, running the tests, deciding whether the test cases were bad or the code was bad when failures happen, repeating until it has satisfied itself and then it lands in the builds that the still-human QA get a pass at. Then the human QA person often basically says "WTF, this is broken as hell" and the developer starts over incorporating the feedback from the tester into the flow.

Now the folks that discard that mandate and use CodeGen for code completion, more curated prompting, they get respectable speedups, but no hope for management to just toss those long term if they still need to curate the CodeGen. So management *really* wants the narative to work for the case above. They have raised the question if the real problem is the human QA, it passes all the CodeGen test cases and it's own code review, so maybe the human QA is just raising a stink to look relevant... Doesn't help that management understands neither the customer, software, or how to develop code, so they base things largely on wishful thinking on what the cheapest and most convenient answer would be that supports them getting big bonuses.

While AI might not be to blame yet for those figures, it is well within the era of TikTok where material that doesn't hit within 30 seconds, the audience is out.

Note that Gemini is responding to a user and not actually running the command.

He didn't say Gemini tried to use that to adjust brightness nor did be say he asked to code a brightness change/ automate it. So I assumed he gave the answer that Gemini would have given a random person asking how to do it.

Problem is that doesn't work either, since the asset 'value' is unknown until transacted, and also ignores the stock value entirely, which is generally not nothing.

For my net worth, you would absolutely count my stock holdings, because at that relatively mundane level, I can probably sell it at the extrapolated price. If you have even a couple million dollars of stock in a big company, you will barely move the volume of trades for the day, so the extrapolation works in isolation.

But when the amount becomes a significant portion of typical volumes, then the transaction will not pan out at 'market rates'.

So we have this awkwardness of something that is definitely worth some cash, but the extrapolated value can't predict the real value, and the other measures would fall short.

I love how it's so blatantly obvious that they are leaning into software development and the models are assuming you are going to write some C# to adjust brightness.

No idea if that's even close to right and I'm not inclined to check, but the fact the first answer is not "Well, find the brightness buttons and press them" is telling.

In a sane world, you might expect the Windows 'integrated' AI to be wired up to just... adjust the brightness..

But *fine*, you don't wire it up, you might then at least hope the AI to say "That capability is not enabled, but here is some help text telling you to do it".

But nope, "That sounds like something an ioctl would do, and I don't know the ioctl per se, but let's just submit random bullshit ioctls and *maybe* it will happen to do as user requested?"

That's the way we've been counting net worth, an extrapolation of the volume that transacted and assuming that *all* of the rest could transact at that price. Nothing particularly new.

I haven't seen really any better ideas deployed, but you are right that it's only "worth" a Trillion while most of it just sits unused. Even "unused" it bolsters the parts he is willing to leverage for loans to basically do whatever the hell he could possibly imagine.

Maybe not for *everyone*, but can be fine for folks driving on average less than 50 miles a day, which is a decent chunk of people.

That said, I think it would be a tough sell to have your car charging at a relative trickle all the time. Folks would be much more comfortable if they can plug into 240V. Which isn't a *crazy* thing to get, especially if you just get a NEMA 15-40 outlet or already have one within distance of parking.

Note I included the EVSE in my cost, and yeah, I needed a new 240V/60A circurit.

As said, I wish I had just done a 50A plug instead, the hard wired was superfluous and that outlet could have been more broadly usable. Though I could downgrade the breaker and put an outlet there easy enough if I cared that much... But just ended up spending money on an EVSE when the bundled one would have charged fast enough..

Not specific to AI, and I frankly can't speak to NHS specifically, but it sounds awfully familiar...

So many things where bureaucratic junk demands awkward forms and processes, and efforts to automate all that stuff instead of streamlining the underlying mess...

To the extent this works (and I can believe it based on other bureaucracies I've been involved with), it's because there's all sorts of dumb boilerplate crap in the process, lots of material generated that is never read, lots of fields to populate that don't matter to anyone. To the extent it ever matters that goes away as the people just stuff meaningless crap in those fields...

The human is still having to provide the crux of the important bit, but there's just so much fluff that is blatantly obvious that LLM can do whatever with that could have been omitted or dealt with better.

Look, I get it, you personally don't like EVs. There are plenty of ICE vehicles for you to choose from, you don't have to be so aggressively disdainful every time anyone mentions that their EV experience is nice.

Cost to install an EVSE for me was about $800, a *far* cry from the $7500 of the tax credit, and I find it hard to believe that any EVSE install would have gone that expensive. If anyone actually quoted that, then they are either trying to rip someone off or don't actually want to do the job and wants to scare a customer off. I could have just installed a NEMA 15-50 for less to go with the bundled mobile charger too, but I just wanted the fastest possible supported charge, which in retrospect I probably should have just done an outlet, would have been cheaper and more flexible and *plenty* fast for my needs.

A calculator can do arithmetic much faster than a human can.

A google search can find web pages way faster than any human can.

LLM is another class of traditionally human-like but not human things happening, in some contexts "good enough" or "better than a human" in others not really applicable, but it's not directly comparable to human thinking.

Well, it lets some people have long cooldowns make the rest of their ecosystem suckers.

Of course, this *also* means the people with high cooldowns get to be vulnerable to security problems longer because they will be applying cooldown to security fixes...

But yes, some sort of actual curation would be the best mitigation, particularly to allow trustworthy critical security updates through quickly instead of those getting caught in the cooldown.

Altman wants some public ownership, but not 50% which, presuming it would be a voting stake, would actually potentially matter for decision making. It's not a majority but if enough private market shareholders side with the public ownership, then it matters.

Instead, he wants enough for the public to have a stake specifically in the "approved" AI companies so that the companies are unambiguously "too big to fail". A chance to hold hostage a big enough chunk of wealth so that the government is stuck doing whatever it can to protect and ensure the selected companies, whether it be in the face of a souring market or upstart companies that didn't have the good fortune of being selected by the company. Meanwhile, the actual governance and decision making remain firmly status quo. Including decisions about how much to send back to "investors" and how much to "reinvest" (including setting their own compensation). They may even structure it so they can classify public ownership differently from private market, and reward investors in each class differently.

Just another ambition to privatize the gains and socialize the losses.

Timing *chains* are not universal, but they do help. Timing belts are quite common. Headgaskets still go and are still a massive labor cost. Within the last couple of years I've had to pay for timing belt change, a new radiator, new hoses due to leaks, a new evap canister, a new alternator. This is on a 2015. I'm told I'm "lucky" the turbo hasn't keeled over on me as this model is notorious for turbo issues. My colleague has had to pay for *two* headgaskets on a 2017 in the last few years. These aren't "solved" problems because it's pretty fundamental physics. Some things are more solved, e.g. I would have had to pay money probably for power steering problems, but the switch to electric power steering pumps greatly helps.

Because it's preventative maintenance and if you have one, you are expected to change it every 100,000 miles. Since we have interference engines, you don't want to push your luck since a failed timing belt will ruin the engine.