It's everywhere right now.
Projects are rejecting AI contributions in needlessly cruel and unnecessary ways.
All they need to do is post an agents md file.
The theatrics are needless.

This kind of thing makes me angry.
Here's a fork I made that explicitly removes the malicious prompt injection.
https://github.com/LynnColeArt...

Side mirrors almost always leave a large blind spot directly behind and close to the vehicle. There's a reason that when firefighters are reversing their appliances they always have at least one of the crew physically get out and watch the area behind the vehicle.

Even a rear window and rear view mirror almost always leave a significant blind spot low and close behind the vehicle, which is why reversing cameras became a thing. When they're done well, they really are significantly safer, as well as sometimes making it a lot more reliable for most people to park the vehicle in difficult spaces.

One of the modern innovations I really would like to have is full AR on my windscreen. I want unexpected hazards highlighted in real time, particularly those that are more easily detectable by non-visual sensors, like big potholes or animals obscured by vegetation near the side of a country road. I want the actual driving line I need to take to follow my planned route through complex junctions overlaid slightly on my view of the road ahead. I want light amplification for night driving, ideally combined with some other technology that can reduce the glare from oncoming headlights to prevent dazzle.

Although I only want all of this if (a) it's implemented well and (b) any additional data it uses is reliably up-to-date and (c) there's an emergency shut-off that instantly clears everything off the windscreen in case anything goes wrong.

Don't worry. You probably have funky modern door handles that don't work when the power goes out anyway. Not that the power in an EV is likely to go out if it's underwater or on fire or anything.

Oh, don't be silly. Next you'll be making even more absurd claims, like that car theft was already a solved problem 20 years ago thanks to immobilisers, or that having separate physical controls for essential functions that you can find and use without taking your eyes off the road for several seconds to mess around with a touchscreen is safer, or that no-one ever hacked 100,000 cars at once from 1,000 miles away back when they didn't have always-on remote connectivity and allow OTA updates to their essential control systems.

Yes, as long as you're the one in the big, heavy car, it's great. Shame if you're the kid it's reversing over though.

Do you ever use reverse gear? What's behind you is pretty important when you're going backwards...

Yes, Apple's shared RAM model really works for them in the context of running LLMs locally. It's a huge advantage. As you say, not much use for those running other platforms, though.

I'm impressed that anyone can afford a new development PC in 2026. I'd need a second mortgage with the price of the RAM, SSD and GPU these days!

Google learned to embrace, extend and extinguish right out of Microsoft's playbook. They were excellent students and you can see the results in how email and web "standards" work today.

The difference is that when Microsoft did it the authorities eventually started getting in their way to promote more openness and competition again. So far there is little sign that anyone intends to challenge the way a few tech giants have recently been capturing long-established standards that we rely on for what have become vital services and effectively taking ownership for their own purposes. The governments and their regulators are either asleep at the wheel or, if you're a bit less trusting, bought and paid for.

Do you often use VeraCrypt on a company-managed device? I'm sure if you do then it's with the knowledge and consent of your IT department and they'll be responsible for managing any consequences of the VeraCrypt issue according to their official policy as well.

Concerned that the reason we keep doing open source is because we believe in access.
The false tradeoff there, is believing that access and exploitation are necessary corollaries. And I don't think they are.
It's a tough balance, and open source licenses have clearly failed us here.
But I'm not sure where to go with it. Shared source might be better, like the Mongo license, or something like it. The Kimi2 license had the right idea.
On the other hand, when you leave the open source path, you pay by losing access.

Let us not forget that we've spent the last 30 years trying to make ads less invasive. This is a fact. There is what is now an entire category of software that revolves stealthy ways to block them. This was always a weak, ineffective, and arguably immoral stream of revenue, with more than trivial privacy concerns.

If you're still depending on ad revenue to run your website, please think of something else.

Next up, this isn't the first time the google algorithm has changed. Louis Rossman did a great video on this. Where he discussed the ongoing troubles he was having getting his website ranked in Google. TLDR there was that he ended up using Gemini to reword his pages in the particular way that Gemini wanted him to, and he was fine.

But the bigger question is: Why are you still depending on Google?

AI porn is avoidable. It's illegal in fifteen states. Why are you running into so much of it?
I'm actively on social media, all the time, and I intentionally follow the topic, but rarely see it.

What are you doing that's inundating your feed with AI porn? No judgement, just curious.

Yes. So far, the LLM tools seem to be much more useful for general research purposes, analysing existing code, or producing example/prototype code to illustrate a specific point. I haven't found them very useful for much of my serious work writing production code yet. At best, they are hit and miss with the easy stuff, and by the time you've reviewed everything with sufficient care to have confidence in it, the potential productivity benefits have been reduced considerably. Meanwhile even the current state of the art models are worse than useless for the more research-level stuff we do. We try them out fairly regularly but they make many bad assumptions and then completely fail to generate acceptable quality code when told no, those are not acceptable and they really do need to produce a complete and robust solution of the original problem that is suitable for professional use.