Have you seen footage of how quickly an unbelted person moves around a car when it crashes? If there's someone in the passenger compartment without a seatbelt they can cause serious damage to everyone else - especially children.

I don't get this take. Once a modern corporation starts making money, all the people in it diligently work to expand their influence by starting new projects and hiring as many people as possible. That seems to be human nature. Why will AI tools change that? Nobody is feeling important because they manage 50 AI agents. They feel important because they manage 50 people.

What percentage of the jobs in a modern office are truly necessary? If automation had the ability to kill jobs over the long term, we'd all have been idle since the industrial revolution. But instead we keep inventing new things that we need.

There is a possibility that the agents become better at managing the company than the people and businesses become as automated as farms did during the industrial revolution.

Yeah and you’re doing a lot of heavy lifting with the term agents.

Billions have been poured into agents and there’s no sign that they will get to a place where they on the path toward generating returns to justify more good money being invested into chasing bad.

I've found AI tools to be pretty awful for low level work. So much of it requires making small changes to poorly documented registers. AI is very good at confidently hallucinating what register value you should use, and often is wrong. There's often such a big develop -> test cycle in embedded, and AI really only solves a very small part of it.

Yeah, it's crazy to think an opaque chatbot will be preferable to a well designed UI for most users. People don't like badly designed UIs, but I'm pretty sure most people under 40 prefer a well designed UI to a customer service agent. We call customer service because the website doesn't do what we want, not because we don't want to use the website.

I got excited about that, until I actually tried to download a model and run it locally and ask it questions. A current gen local LLM which is small enough to live on disk and fit in my laptop's RAM is very prone to hallucination of facts. Which makes it kind of useless.

Ask your local model a verifiable question - for example a list of tallest buildings in Europe. I did it with Gemma on my laptop, and after the top 3 they were all fake. I just tried that again with Gemma-4 on my iphone, and it did even worse - the 3 tallest buildings in Europe are apparently the Burj Khalifa, the Torre Glories and the Shanghai Tower.

I wouldn't call that effective compression of information.

Yea, it's not an encyclopedia of facts. Language models store the FEELS of the data in vectors (or angles in Gemma4's case, it's a cool thing) not the exact string.

But what you can do with local models is give them actual data and tools to search it. Download a copy of Wikipedia locally, give the agent a way to search it and BOOM accurate information without an internet connection.

Also "small enough to live on disk" is a bit vague, especially when models get super stupid super fast when you get to the smaller size. At that point they're just basically 40k servitors that can use tools and nothing much.

I don't think any LLMs are good at accurately regurgitating arbitrary facts, unless they happen to be very common in their training, and certainly not good at making novel comparisons between them.

What you're describing is 2 or 3 sensors - effectively 2 or 3 pixels. Enough to discriminate when an aircraft launches a flare, but not really "imaging" in the modern sense.

Early heat seeking missiles would use a single IR sensor with mechanical scanning.

Thermal imaging and machine vision, of the kind you can now do cheaply, isn't 80s tech. It's probably late-90s tech for advanced western states. And now it's starting to be ubiquitous cheap tech too. You can buy a thermal imaging camera with 20k pixels for a few hundred dollars now. Combine that with some image processing and you've got a very robust target detection pipeline.

Reading between the AI induced hype of the article, I think the crucial development is that the missile is effectively using an infrared camera and image recognition rather than just "point at hot stuff" which is how earlier heat seeking missiles worked.

I'm pretty sure I could buy everything I'd need to build a thermal imaging tracker for a few hundred dollars. So perhaps not surprising that Iran did the same.

I realised after a few near misses that my voice is by far the lowest latency signal method I have. If a situation suddenly seems dangerous I'll yell. Perhaps not very polite, but far more polite than hitting someone who stepped out in front of me. A bike bell probably adds a second of latency to find the bell. I'd rather use that time to brake.

The bell can be useful as a more general "I'm here" warning. But if there's any actual risk of a collision, yelling and braking are far more effective.

There’s far more tourists in Nepal who are trekking than mountaineering. And those tourists are going to be much more price sensitive. It’s not just wealthy Americans, but people from all over the world - India, China, middle income countries etc. All those people are spending money in tea houses and hiring local guides.

If the author didn't abuse the fax, why would anyone notice the process was broken. It's only by abusing the existing process that change will be triggered.

You see this all the time in cybersecurity. Nobody cares until there's a breach. Nobody would care if he faxed 25 pages and mildly inconvenienced Karen, but by faxing 500 pages and inconveniencing the whole office, it's going to start something. Even if it takes them another 5 years to fix the process, it's a start.

Realistically, the change will probably be "no more than 25 pages of evidence required". But that's also a win for the person being asked for it.