I'm not on the engineering team at my company, but I do a little bit of coding as part of my job.

My company has a good AI coding policy: "You are responsible for all code you check in. You are encouraged to use AI responsibly, but you must check it for errors. If you turn in code that is bad, you can't blame the AI."

Plus, while we are trying out a few different AI platforms, we have token limits (I don't know what they are, because in my role I never come close to hitting them - but company-wide I know our budget is lower than one engineer salary.) And different platforms are allowed for different things. Any company-proprietary information is only allowed in certain AI platforms; and nobody is *EVER* allowed to put sensitive information (API keys, passwords, even user names) into any of our AI platforms.

In my role, I often interact with customer data (customer logs, customer integrations, etc,) we have exactly one AI platform we're allowed to put "customer data" in to, because it's the only one we have 100% control over the data for. That platform does have multiple models available, though.

As for what I personally use it for when coding? Debugging. "Crap, this shell script isn't passing through this environment variable properly over SSH, figure out what I did wrong."

Or "A customer wants to integrate our API with - write me a Python script that does " and have the AI look up that other company's API info and get me a starting point. I never directly use any of the code generated this way, I use it as a starting point for my own code. An "Oh, that's how that company's API works", which I then verify on the API documentation link the AI gives me, to make sure it was actually telling the truth.

Is this the same OpenClaw on which a major vulnerability was recently discovered that allowed websites to take over your machine? The same OpenClaw whose stack comes with a list of remote execution and access control vulnerabilities?

Typical for Microsoft to do a lousy job of vetting their tool chain.

I'll pass.

> some critics... safe to fly.

That includes me. A random internet nerd.

For columbia, they knew a piece of ice hit the wing, not that it had damaged the wing. What everybody tries to ignore is that during the two weeks in space, there was talk about an unplanned spacewalk to "have a look". They decided "not to have a look" and paid with 7 lives.

It may be a trite saying, but it's as true in education as it is in a gym. If you don't exercise your brain, it's not going to improve.

There's a reason weightlifters don't use a forklift or crane to pick up the barbells and do a dozen reps. The problem is not that the weights are in need of lifting. And that's the same problem with homework. The teacher doesn't need a stack of 5 page reports; what they need is for their students to practice using their brains.

Unfortunately the education system is designed to evaluate output instead of process. It's easier to grade a paper or a test, not evaluate a demonstration of knowledge. It's always been ripe for cheating, but now the cheat tools are everywhere and made legitimate by techbros demanding AI productivity. So either teaching will change, or we'll head straight for idiocracy and nobody will be left with the skills to wonder why it all went to hell.

This study found that *on average*, a majority of PHEV drivers in _Europe_ don't both plugging them in, making them no better than a "conventional non-plug-in" hybrid.

But as an individual PHEV owner, you can make it far better than this study says - simply by plugging in whenever possible.

I got a PHEV (the BMW i3, what BMW officially called "an electric vehicle with range extender) as my "entry into electric vehicles" - and in four years of ownership, I used maybe ten gallons of gasoline. And I'd say half of that was "burning it up just so it doesn't go stale". It prepared me to fully commit to battery-electric-only with my next vehicle.

Yep. They're always saying solid state batteries are 2-3 years away.

Sure, when solid state actually does happen (and I expect it will) it will be a game changer for EVs.

But Toyota just keeps postponing doing anything serious with EVs because they keep claiming this is right around the corner.

Wasn't the 2025 Prius supposed to use solid state batteries?

Specifically cross-platform, not vendor-app or vendor-cloud dependent.

I still have some Hue bulbs, and a few WiFi bulbs that are dependent on vendor lock-in that I'll be replacing when they go out.

Any newer devices are Matter/Thread compatible. Local control, no vendor lock-in.

Plenty of other companies have an OTA cadence on par with Tesla's now.

Rivian, Lucid, even Ford is at least in the ballpark.

Yes, Tesla's software is good, but they're not the only ones doing updates of substance with frequency any more.

> The "premium, authenticated digital identities" created by Hyperreal's system are "not replacing artists," says Hyperreal CEO/Chief Architect Remington Scott

Yeah, they're allowing the families of dead celebrities to wring a few more dollars out of their corpses...

The overwhelming majority of these charging stalls are 10kW AC, not 150kW+ DC.

And that's still less power than is used *JUST IN CALIFORNIA* for just *AI* datacenters.

You just bled tons of users over Jimmy Kimmel. Now you announce this *THE DAY JIMMY IS REINSTATED*?

That'll make some users just decide to stay off.

Seriously terrible timing. They should have waited at *LEAST* a month, to give the "Jimmy quitters" time to rejoin.

The big thing for me is that AI doesn't "write the code I put in production" - it provides guidance on techniques to use, or solves bugs I have written.

The same as StackOverflow for me. Just more personalized to my exact situation.

"I'm writing a shell script to ssh into a remote system and run some commands, I have to use some environment variables defined locally on the system I'm executing the script on, and other environment variables that are defined on the remote system I'm connecting to, and I can't remember how to escape things properly to pass through correctly." I can just feed an LLM my exact command that isn't working right, and ask it to rewrite it. It takes 2-3 further prompts ("That produced this error message, please try again") but it generally bug fixes it.

Or "I need a python script to integrate this company's API, as documented on this url with this other thing, and do this task, what would be a good sample?" I don't take it exactly as it spits it out, but use it as a basis for my own code.

I would say that in the last four years of using LLMs to assist, maybe 10% of my actual deployed code is "directly from an LLM, because it produced clearly functional code" - usually only short snippets. One short function in a Python script, for example. Maybe another 20% was "came from an LLM prompt, then heavily rewritten, because I didn't want to feed potentially proprietary data into the LLM."

They exist now. They're either small toys, or large horrendously expensive limited-purpose things.

The problem is that they're pointless. Anything a humanoid robot can do in an automated manner, a specialized non-humanoid robot could do much cheaper.

I don't need Rosie the Robot to use my regular stand-up vacuum cleaner. I have a Roomba.

I don't need a humanoid robot to sit in the driver's seat of a car to drive me around, Waymo exists.

I don't need a humanoid robot to stand in a factory using a spray can to paint a car, automated industrial robots that can do tasks like that (or welding) have existed for decades.

Chevy built one. In 2010. It was called the Volt.

Chrysler builds one. The Pacifica PHEV.

Heck, even BMW built one - the i3.