You must not have a lot of experience with unions to think there's anything odd about this behavior. It's pretty well expected for them to pull something like this. It's not about protecting the artist, it's about getting their cut and maintaining their control.

Like executing extra code to check for various memory safety rules?

It's not about any controversial topic or language. Sure, the paper may sound totally reasonable, but is the data real or fabricated? Is the methodology sound? It sounds like it wasn't.

I'm not an AI expert, but I believe this pathway of creating a monolithic AI that is in and of itself the reasoning, knowledge and computation is the wrong way to go.
Stephen Wolfram (of Mathematica / Wolfram Alpha fame) had insight that LLMs would be best served as an interface between user queries and the knowledge / computation engines, instead of attempting to be the knowledge and computation engines themselves.

An analogy is asking a person to add up 100 numbers in their head as you call them out, versus asking a person to use a calculator to add them up as you call them out. The calculator is pretty much infallible in this regard, as that is what it is designed to do, while a neural net may decide 1 + 1 isn't actually 2 on occasion (partially due to temperature for one thing). It would also require drastically less computing power, as the calculator is optimized to do that task, while a neural net has to build up a bigger and bigger token memory to hold the contents of those calculations in a much slower and inefficient way.

They are already heading in this direction, for example you can see LLMs are now actively accessing the internet seeking out relevant information that is the source of what they produce, as opposed to that information having been cooked into the LLM's weights during training.

Great, this will make it much harder for us to stop Skynet.

I don't know how malicious this is, or if China can exploit it, or even any hackers in general. The thing is, in this day and age many microcontrollers contain "communication devices" built in, whether they are needed or used. Like ESP32 microcontroller modules, that only cost a couple bucks each, have WiFi and Bluetooth built in.

There's strong incentive to use cheap generic microcontrollers for most anything like this now, and it's usually cheaper to use some generic mass-produced thing with extra capability than design chipsets for specific uses. Everything from CPUs to vehicles have built-in capabilities that are turned off, because they aren't needed or haven't been paid for and unlocked.

I strongly suspect that's the case here as well, but you never know.

Bill Gate's 3 kids are only going to inherit around 1% of his worth each. His net worth is $102 billion, so that means they only get around $1 billion each. Poor kids.

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fpeople.com%2Fbill-gates-...

Google's cached webpages are no longer available for individual consumption, and it's dawning on me it may be for this very reason. By keeping their cache private they are keeping the value of it (IE pre-AI era) for themselves. Perhaps for use in training since it hasn't been corrupted with AI generated content.

The Internet Archive is a great thing, but it's by no means browsable in any useful way. It's mainly for viewing a specific page's archive, with gaps and disassociations between the specific times things were crawled.

I'm wondering what it would take to identify and label "pure" internet content that at the very least already existed prior to AI and is unchanged. You know how a browser tells you a website is "secure" and certified, etc? How about one that indicates that the page's content pre-existed AI. It could grab the page from the Internet Archive, generate a checksum, and compare it to the live version. Lots of caveats of course, like filtering dynamic content (ads and the like) that do change.

Obviously the statistic refers to long-term drivers. It's not like there's some sort of curse that anyone whose driven a truck is going to die early.

The three reasons:
Lack of health insurance since a lot of drivers are self-employed doing gig-work. Employers don't often provide health insurance, and also due to the church with drivers switching for the sign-up bonuses they often don't stay long enough for insurance to kick in.

Sitting long periods is really, really bad for you. The more studies that come out, the worse they find it is. Sure, lots of office workers sit, but they can also stand any time they want and take a short walk. That isn't something a truck driver can reasonably do often, especially since it's hard to even find good places to park a big rig.

It's hard to eat a good diet on the road. You can't really prepare your own food, so you have to eat at truck stops and the like regularly. That is what did my uncle in - he developed type 2 diabetes, which he could not control on the road, and died an awful death of blindness and kidney failure. He did dialysis at home and lingered for several years before dying.

Do you have any idea what the turnover rate is for long-haul tractor trailer drivers? The majority of new drivers move onto another career in less than a year.
Do you know what the average life expectancy is for a truck driver? 61 years. That's 16 years less than the US average.
Do you know what the turnover rate is? 91% of truck drivers work less than a year for a given company.
Do you know how many NEW drivers will be needed in the upcoming years? 1 MILLION new drivers need to be trained, to replace those retiring and of course replace all the new drivers who quit.

That last statistic alone proves you are incorrect in your assumption, as a million self-driving trucks could merely replace those who quit and not displace a single worker.

Sure, there are people who make it a life-long career, like my uncle who died in his 60s after driving a truck all his adult life. However at any given time a large percentage of the truck drivers on the road are brand new, which makes them more dangerous because they are inexperienced.

It is not a great career, but extremely necessary, so people are lured in with big signing bonuses (like $16k and they only have to stay for 6 months), to find the job really isn't great and not many people can really identify with it enough to stick with it long-term.

Regardless, we're talking LONG HAUL trucking here, not the more regional or last-mile type trucking that cannot be easily replaced by AI. There are many truck driving gigs that require unloading as well, which obviously a piece of software cannot do.

My prediction is the long haul depot-to-depot type trucking can be done totally self-driving, however I think we'll see something similar to the marine cargo vessel type setting, where a local maritime pilot boards the ship to safely navigate it in and out of port. I can see self-driving trucks arriving at a city, a truck driver hopping in, and driving it those last several miles to the destination, then back out, and constantly doing that kind of rotation as the self driving trucks come and go.

Well, there's also the part where 35 Democrats voted to block as well, with 14 not voting. That's a quarter (24%) of the Democrats not siding with Newsom.

Oh that's definitely the case often, and one of the biggest screw ups it makes has to do with versioning - like with the Qt example I was targeting the latest version, but it would often come up with code using deprecated (as in the "totally removed" kind of deprecation) versions of that library, even when I was very clear about the version to target.

Another thing that is helpful is using different LLMs. I'm generally using ChatGPT and Grok for that. Both do a much better job than other LLMs. I have Gemini integrated into IntelliJ which I use for the PHP / MySQL stuff, but actually don't use it a huge amount.

Every now and then it will do something surprisingly smart and save me a lot of time. For example, I had an array of URLs, each with a bunch of query parameters, that I was converting to JSON. It was one of those things where the volume didn't really warrant throwing together a routine to do it, so I started doing it by hand. After I converted the first entry, Gemini picked up on it started suggesting the following lines for me and got them exactly right. Saved me several minutes. That was pretty intelligent - breaking down URLs into JSON arrays of entries for each parameter name / value, without any explanation just because it saw me doing it. That's more like the kind of stuff I have dreamed of AI doing.

I'm also a full-time software developer for the last 30 years, and it sounds to me like you are very deeply entrenched in some specific languages / API / frameworks that you seldom work outside of. So I can understand not finding use for these LLMs if you're doing the same kind of stuff you've been doing for a long time.

In my case, outside of my main gigs (which is mainly LAMP stuff), I do a ton of very diverse things. I have iOS and Android apps on the market (Swift and Java respectively), do a lot of embedded stuff (ARM and ESP32 in CPP) that is low-level bitbanging interface stuff, as well as Windows app development.

One of my latest projects was taking an Android application (java) for TV boxes I'd written years ago and totally re-implementing it in Qt in CPP. The LLMs were absolutely invaluable for me to learn how to use Qt and let me implement this in probably 1/3rd the time. Now I wouldn't consider this vibe coding at all, because I basically used the LLM in place of poring over code samples and trying to delve into API documentation to put 2 and 2 together. Anytime you get into a proprietary realm, which really any API is, it's simply all arbitrary, and an LLM can help show how to jump through the hoops (as well as weigh the pros and cons of various big decisions) much faster than trying to become a Qt expert and figuring it all out via trial and error.

Imagine Hello World and example code that is extremely specific to what you're trying to accomplish. That's the biggest use of LLMs for me.

You are not hallucinating - this story is a dupe

And an update a day later, there was no abnormal weather: