I'll add a potential differentiator...

AI will replace the software jockeys that just write code (who mostly aren't going to be that senior). It absolutely won't be going anywhere near the actual *engineers* any time soon though.

Whatever you're doing, you need to actually *run* the thing. For that, you need non-functional capabilities (scalability, manageability, observability, etc) - AI isn't getting those things right on its own, so you need engineers to direct it to do so. Those engineers may not spend much time finessing the semicolons in code, but they're still going to be "coding" in so much as the level of detail they go to will be about the same.

I'd say this is likely a bit like how no one really writes machine code any longer. Even the number of people writing C is pretty small - all replaced by higher level languages that do things well enough. So too AI.. it's replacing one with another. I'd argue there are *more* programmers because they don't have to write machine code, and so maybe AI will make more prompt-writing programmers... time will tell.

The source code for their amazing technology has been leaked:

if content.isMusic() then {
      content.containsCopyrightedMaterial = True
} else {
    raise DontCareAboutCookingRecipesException
}

I mean come on, just about everything is a derivation of something else. This just makes it like youtube moderation where "the computer said it, so it must be true".

In your normal, reasonable world it's a good idea. By the time MS have run it through their product manglers, it'll be anything but good. In short order, something as mundane as Solitaire will ask for "network access" - why? Because that's how it gets its ads and how CrapPilot integrates with it. Before you know it, just about everything you have installed has a whole raft of really broad permissions and so the whole thing becomes utterly pointless.

I'll bet my lunch that when this is implemented, the required permissions would not have prevented/mitigated (say) the Notepad markdown vulnerability just announced. Security Theatre at its finest.

I wonder if they'll also add in Fist Recognition, as I suspect their glasses may be experiencing those in the near future...?

I use an Amcrest camera on a private network to a NAS with Frigate on it. I also back up to a cloud service (B2) - which costs me about £30/month. Even with it paired down as low as I can get it for it to still be useful, it's many many times more expensive than a Ring doorbell.

Doing it yourself might well be the "right thing" to do (which is why I do it that way), but man oh man, it's not going to catch on - it's way too complicated and way too expensive.

Then again, where I live we have laws that prevent things like this Search Party feature, it also prevents Ring Nation being a thing on our TV (without each owner consenting to it). That doesn't mean Amazon aren't sharing with the cops (they absolutely are), or using the footage for their own research or whatnot, but that's most of what they can do with it.

FWIW, I think the best we can hope for is that CCTV products like this are made and sold by a responsible company. We don't want to go back to the badlands where people would use those crappy Chinese made-for-cheap cameras that all became part of botnets and illicit cam-sharing sites. That was definitely worse.

Yeah, it was pretty likely.

That said, good on Apple for delaying. If/when they release, it needs to be really good - much better than anyone elses *on the phone*. As noted elsewhere, if you ask it to alter a setting which is deep in a spaghetti of menus, then it needs to do it right first time. It needs to be at least as good as the others for the inevitable stupid questions people will ask when testing it out too, I guess.

If Apple release a half-baked product, people really lose their gussets over it. Contrast to Microsoft who most definitely released a half arsed product, and yet that's exactly what we've come to expect of them.

I agree there are lots of chancers, which by the looks of things are all financed really badly. However, SaaS isn't inherently the problem. If you had a shaky product before AI, then it's true AI makes it more shaky, but a solid product and solid financials work no matter what.

I have no idea about this company, but Doist (who run todoist.com) somehow makes money out of a to-do app (and have been at it for years and years). It's the thing just about every software tutorial tells you how to build - and yet somehow they carve out a living and there aren't dozens of super cheap vibe-coded competitors eating their lunch. It seems then that it's harder than it looks to successfully run a todo app, even though it's incredibly easy to write the core competence in software (hell, you don't even need AI to do it - just copy/paste).

To try to summarise without quoting...

Some SaaS* companies borrowed money on crazy terms, which are highly vulnerable to "blips" in the market. A downturn, perhaps like that seen in 2007, or in Covid could cause a number of those companies to fail because there's not yet enough value created (in other words: money made in a war chest) by them to withstand it.

(* I'll bet it's not just SaaS - I'm sure every man and his dog with an AI or Crypto business has been trying to do the same thing)

To simplify even further... idiots borrowed money from glorified loan sharks and have been riding something of a hype wave ever since. If that wave crashes, even a little bit, some of those debts won't be repaid and companies will fail.

Then again, any well run companies with solid financials will do just fine. As the celebrity chefs have found, just because you sell a cookbook or two doesn't mean someone else can just open up a restaurant and compete with you. AI won't immediately destroy SaaS - but it may rout out some of the crap ones with the help of a "market correction".

Here in the UK, you can lease a BYD car for something like £300/month, along with an in/out charge point attached to your house. You have to plug it in 20 times per month (so pretty much every night, I guess), but in return, you pay *zero* to drive the car. That is, all the charging and discharging of the battery will be entirely free to you (as will servicing, and I think tyres). I don't know the specifics, but there is some sort of SLA that says it won't discharge below a certain amount.

The £300/month car payment is pretty low - most new cars come in at about twice that, although you can get some BYDs pretty cheap on a Salary Sacrifice scheme (which I don't think this is). How much you save on electricity depends on your mileage, I guess. I pay £0.07/kWh on my off-peak rate, so you've got to use quite a few kWh before you're spending even £50/month. If you do a lot of miles, then this may not be for you, as you'll potentially not have a full battery before the day starts, and charging away from home isn't covered.

Either way, it's a pretty good deal for some people. The BYD doesn't meet WAF in my house though, so it'll never happen here.

Indeed, and if one person can help out (say) 5-10 cars getting to their destinations, then that's a taxi with only a tenth of a driver. They've shaved off 9/10ths, and are angling to make that 19/20ths and so on. It doesn't have to be 100% autonomous to be useful, or indeed "AI driven". In fact, if anything, it tells us something about the car manufacturers that claim to have "full self driving" coming any day now...

Way back in the day, some people found that if you played a recording of a voice agent saying "hello and welcome to Superstore, my name is Dave, how can I help you today?" rather than having the actual Dave say those words, you could actually have Dave answer $some_percentage more calls per day. Not really "AI" in the hype way it is today, but still using a computer to make a human more efficient. SSDD

> And that is why an LLM cannot and will never be able to reliably generate correct code, unless it is really simplistic code

Whilst I'm inclined to agree with you, I believe the hope is that scaling the LLM and its various input systems will push the level beyond "simplistic" up to "complex", and so "never" is unlikely.

FWIW, I think there's likely some capacity for this to be true. In my meagre experience, LLMs seem to lose the plot after a lot of context is added - that sounds like a "memory" sort of problem. In reality, it can be thought of as the prompt getting too long to be fully used in the inference process, which is a hard problem to solve, but feels like it would be solvable if you have enough compute power to do it. That said, there may be a limit even if you have endless compute, perhaps because there's an exponential requirement to keep looking back at the components of the prompt whilst also trying to perform inference (or something like that).

To break this out into an analogy, there's a natural limit to how many balls a person could juggle. They could juggle more if they could throw the balls higher, so perhaps arm strength and throwing skills have a part of play. Eventually though, physics takes over and there's no way the jugglers arms could be any stronger, or their throwing skill be any better. At that point, the only way to scale is to give the juggler additional arms - that's a hard problem to solve, but is (notionally) solvable. It's not an endless solution though, at some point all those arms get in the way of each other and there's no airspace left for balls to travel. We know where humans are on that "continuum", but much less clear where LLMs are.

I've been using Gemini alongside a Vue web app. By preference I'm a backend sort of guy, but I've quite enjoyed working with Vue. I really dislike some aspects of Javascript, HTML and CSS, but there Gemini (usually) saves me quite well. It comes up with some useful chunks of code, which I accept maybe half the time. Sometimes that saves me one line of code, sometimes it saves maybe 10.

I would stress that in my view, however the code was generated, it's still my responsibility. As such, I do spend a lot of time looking over what it (and I) have written to try to refine it or ensure it's safe or whatever else. As such, I wouldn't really describe my project as "AI generated" any more than I'd have said it was "autocomplete generated", but AI's definitely got me to my destination a little quicker than I'd have been without it.

I have of course prompted for "give me a function that..." type stuff, and it's quite good for things that are likely quite common. It does a reasonable job of things that I'd say are probably quite niche too though. Again, it all needs a bit of review and refinement, but again, it saves a lot of time. By way of a backend sort of example, I needed a script to make some API calls into a Gitlab server. Nothing difficult, just a chain of calls, each using some information given by the last. I could probably have written that from scratch with a bit of googling in about an hour. I asked Gemini for it, and I got it in moments, and had tried/tested it and used it all inside about 10 minutes. Quality isn't a high concern here, as it's not 'critical path', so actually AI was exactly the solution to it.

So in my meagre opinion, AI's quite good - it's not awesome by any means, but it has its uses, and it makes me a more efficient developer (by how much is up for debate, mind you).

If it could go to one of those "new generation" websites that says almost nothing in really big letters, and then figure out what the hell it's all about and report back to me in something approaching usable information, then MAYBE it has some use.

Otherwise, I'll do my browsing myself, thanks. I don't need a computer to take over my idle-but-trying-to-look-like-I'm-working time, thanks.

Are they still harping on about this? It's been a week already: https://tech.slashdot.org/stor...

Here's what I said last time around: https://slashdot.org/comments....

Age based limits aren't the solution to all ills - just like having an age limit for alcohol doesn't stop alcoholism. However, it gives the person a chance to grow up enough to be able to think through what they're doing - at least a little bit before taking it on.

In the case of social media, it's obviously designed to be addictive. At 13 you've got zero skills to resist that addiction. At 16 you might have a few, and at (say) 20 you'll have a few more and at 40+ you'll have lots. You might still become addicted, but at least you get a chance not to be.

Personally, I can't see why we wouldn't *all* have age limits set by government. After all, they're only set by the pushers themselves at the moment - and that seems far too slanted in their favour. Also, an age limit doesn't preclude other measures being introduced later (just as we have anti-drink-driving laws for alcohol, for example).

As for defining 'engagement' in order to ban some of it, that's going to be tricky. Is slashdot working for your engagement, for example? Irrefutably it is, but we wouldn't consider it a 'social media network' like we do facebook. Privacy violations are probably easier to determine though - and separates out the likes of slashdot from the likes of facebook quite clearly.