I'm certainly not sold on the idea that "government controls AI technology" is better than "bunch of silicon valley bros control AI technology". I don't trust either of them. I would absolutely rather have private development with more strict regulatory oversight, which at least keeps things relatively public and gives voters at least some say. The more groups involved in overseeing what goes on, the better. I get that red tape is out of vogue recently, but I suspect that's only a temporary thing.

I'm still impressed that the Teams web client can't edit a message with an emoji in it without messing up the keyboard controls and often corrupting the message and losing data. It's had this bug for years. It's literally a messaging app where you can't reliably send messages.

I've only used Claude code for a matter of weeks at work. While it's supposed to stay contained in the folder you point it at, I've noticed that it has no problem *looking* outside that folder, and even acting outside that folder if you gave it implied permission. In one case it used the connection string from the application source code it was working on, and the fact that my windows account had read-only access to some tables in the production database, to connect to the database and query tables in order to answer a question it had about the potential ramifications of a code change. The problem is that while we're OK with this particular source code leaving the premises, we're not ok with data from a production database going offsite. In this case it wasn't sensitive data, but after that incident I only run Claude Code from a VM under a user with minimal permissions. Clearly it would have been prudent to do that from the start, but let this be a warning to anyone else who just downloaded Claude Code, turned off the "use my data to train future models" and gave it a go... it will *not* keep itself inside the folder you give it. Put it in a sandbox. I heard a similar story this week from an IT friend who said devs at his company were using Cursor.ai and it was doing scans of the network drives, so they did the same thing and moved it all to VMs.

The simplest explanation is usually the correct one, and in this case, given what I've seen people do with AI, and given my own experience with Claude Code (Opus and Fable models), the normal human inclination to trust it's probably right and click OK is by far the most likely explanation. And remember that frontier models have grown increasingly better over the last year, and one of the newest improvements is that the reasoning models use two models in an adversarial mode where the second model checks and challenges the output of the first. The timeline of the events in Iran probably indicate that this was an older AI model without those newer advances. I can totally believe this story. When consumers were expected to keep their hands on the wheel of self-driving cars, I absolutely knew they wouldn't, and when humans are told to check the conclusions of AI, they mostly just click OK. I do actually look at the code that Claude generates before it goes into our codebase, and I can tell you, the desire to just skim read is real.

As they've been telling us software developers who are being asked to adapt to probably the most rapid transformation of any career that's ever been seen... "it's ok to take a moment to grieve." I was spending most of my time writing code up until about the middle of August, and then I realized that Fable was finally good enough to work on the rather large code base that I work on, and now I write less than 5% of the new code, and I'm pushing out new features at probably 3 times as fast as I was before, while now spending almost all my time creating and reviewing specifications, reading code, updating documentation, reviewing software architecture, and making bigger plans. It's jarring, exhilarating, and humbling at the same time. It's gonna take a minute for us to adjust. And yes, things are still changing fast. Software development has embraced this. It's coming for 3D CAD software and mechanical engineers next. There's been pushback within the more creative arts, like graphic design, writing, and making videos, and I totally understand that. Those industries will grapple with it in different ways than we have in the engineering space. And we won't lose all the creative stuff. As someone recently said to me, "we still ride horses, but we don't ride them to work."

... did Slashdot become a god damned commie circle jerk?

We had a 2004(?) nissan x-trail for a while (in Canada), and the traditional analog instrument cluster was in the middle of the dash on that one. It was pretty easy to get used to.

We're well past the point where we all have a calculator app in our pocket, but if you've ever felt the frustration of having a discussion with someone who can't do simple arithmetic because they never had to do flash cards in school, you probably understand why students should learn the basics before moving on to more power tools. There are just times you need to pickup a screwdriver, and times you reach for the impact driver.

Renaming a lake and bullying Canada are distractions from the real issues he cares about: the war with Iran, and the Epstein files. Looks like it's working.

These are all the reasons those of us who live in suburbs give when we're asked why we don't live in the downtown core of cities.

The problem with this argument, from the shop's point of view, is that the UK is in the middle of a shoplifting epidemic. The policing resources simply aren't there to deal with small scale (or even flagrant and much larger scale) shoplifting. It now seems to be rare for police officers to attend reports of shoplifting at all. And so effectively shoplifting becomes legalised, which is obviously hurting the big chain stores financially and potentially devastating for smaller shops.

Shops are trying to find some way to respond to this to protect themselves. They typically tell staff, even security staff, not to get physical with shoplifters as they walk out, not least because the risk of something worse happening and the business becoming liable for it is significant. But then the shoplifters know this and some of them will just walk in and openly empty a shelf into a bag and walk out again without paying because they know no-one will try to stop them.

Shops also know that our court system is completely overwhelmed at the moment and there is little chance of bringing any successful civil action to recover damages from shoplifters who aren't even being arrested by the police and put through the criminal justice system.

So what do they have left? Basically, they're trying to spot the "usual suspects" and refuse them entry in the first place, which is a less risky proposition in terms of potential for physical violence and quite effective because a lot of shoplifting is done by the same people, or they're trying to make it more physically difficult to steal goods, for example by using locking systems or keeping high value items behind counters where customers can't just pick them up. But these systems aren't perfect, far from it, and what we're discussing here today is one of the significant problems.

Obviously this kind of technology can have very bad consequences, as we see here. I don't think it should be used if it's not highly reliable, given the potential harm done in cases where it gets something wrong. But I do have some sympathy with the shops who are looking for something -- anything -- to reduce the threat they're facing because the rule of law has failed in retail settings.

That's fine, and I use Claude for software development. I check what it creates, and my goal is just to create the lines of code that I would have created, but faster. It can watermark those lines of code if it wants. My employer knows I use the AI coding assistant, pays the monthly fee, and wants me to make use of it to produce the same code faster.

But I suppose this is mostly about kids using it to do their homework, or people using it to try to defraud others. 100% good idea.

On the contrary, I think most of your comment is as relevant as ever. A carefully designed UI is still better for almost any specific task than what "agents" will produce. It will probably be more efficient to describe what you want precisely enough to get it right. It will not be undermined by problems like non-repeatability, non-determinism and hallucinations that are inherent in LLM-backed agents.

Agents can be quite effective at getting something that seems about right, as long as you don't care too much about the details. And of course they get there much quicker than doing it manually as a human. This is useful for some tasks, and why they have become popular.

Unfortunately, it turns out that for many of the things we really care about, "about right" doesn't cut it. Right now, today, the software industry is already being overwhelmed by AI slop. It's hard to believe, but it's still only been a few months since agentic development processes arrived on the scene. And yet after just those few months of management trying to replace real developers with agents, we can see the slow-motion train wreck happening right before our eyes. And in return, where are the new applications that radically improve on what we had before thanks to all this added "productivity"? If this tech is going to change the world as the AI tech bros and the sycophantic CxOs kneeling at their feet would have us believe, where's my flying car?

Imagine a Beowulf cluster of AI-generated cybersecurity nightmares...

It is. The fact that our dominant desktop operating systems are still using security models from last century is probably the #2 disaster of modern tech security, coming in just behind so much software with potential security implications still being written in languages like C and C++ at #1. And the understandable yet still deeply regrettable reason for both disasters is the same: momentum.