From my own experience:

> Was it mostly boiler plate?

A good bit of it was, but speeding that up without having to dig through a bunch of templates saves a ton of time.

> How much of the generate code was correct?

A lot of it, but I generally only use it to complete a few lines of code at once, and do the high level thinking on my own. If you use AI that way, it's a great productivity tool.

> How much of the Copilot code makes into production?

A lot of it.

...unless you can point to the *specific work* that was taken from *without knowing how it was made*.

In other words, just because you train an AI on a bunch of works, it doesn't follow that most of what the AI makes is violating copyright or plagiarizing or whatever. A lot of people are fundamentally pissed off because they don't feel special, or else they wouldn't be yelling at AI users in hobbyist communities who aren't affecting anyone's livelihood.

I have a lot of code out on github, which LLMs have trained on. I have absolutely no right to tell them that the AI can't learn from my code, because my IP rights don't extend that far, and that goes for art as well. I also think it's great that AI enables people, on their local computers, to do something that until very recently was very hard to do.

No, you have a copyright, which allows you to control redistribution of your specific work. You do not, nor should you, have the right to prevent people from learning from it. If I want to disassemble Windows and learn from the code, I'm 100% allowed to do that.

> You can generate images with signatures from the works that were copied, without attribution.

I've never actually seen this happen. What actually happens is that the AI has generalized on what a signature is and figured out that the name of the artist often appears in cursive down in the bottom corner of the image, so it writes a name down there (an artist's name, if someone tells it to make an image "by so-and-so"), but I've never seen a case where the signature matches the signature of the artist to any degree.

The existence of signatures absolutely is not proof that it's copying anything, because the signatures themselves aren't even copies.

Copyright allows transformative works.

Also, you're using weasel-y wording here. It does train on entire works, but those entire works aren't saved in the AI itself (this is mathematically impossible giving how many works an AI trains on versus the size of the AI. When it trains on an entire work, it absolutely does generalize on style, concepts, and ideas. Only if the entire work is trained many times over does it memorize that work.

Castillo de San Marcos - Built at Sea Level in 1672.

Castillo de San Marcos - Still at Sea Level in 2025.

That's what the article is going for, anyway. They want readers to feel like AI is fundamentally insecure, when that's not what happened here.

Fuck making people go to the office. It's a waste of everyone's time and energy, and it causes pollution and more traffic and reduces productivity, all so micromanaging twats can feel good about themselves.

I'm lazy, but even I would do a daily backup if hundreds of millions of dollars were at stake.

Hundreds of millions were not at stake the day the hard drive was lost. He could have easily replaced it back then at a much lower cost.

The drive was lost in 2013. In 2013 the price of 8000 bitcoins ranged between $400,000 and $8,000,000US.

Even at the low end of that estimate, it's still more than the value of my house. Not doing basic due diligence like keeping good backups is massive negligence. Yes, his girlfriend shouldn't have thrown his hard drive away, but he was the one who put himself in a situation where he could lose that kind of money from a single mistake.

How is this of any significance to humanity? Seems like a weird way to waste both time and money.

You're throwing the baby out with the bath water. Most people don't even have landlines anymore, so you'd make 911 mostly useless. Not to mention that there are many valuable uses of 911 on mobile phones, like calling for help after being in a car accident. The big thing is not to let people spoof their call's origin. There are actual use cases for fooling caller ID, but the damage caused by spammers and scammers far outweighs the benefits. If we make caller ID a reliable indicator of the source of a call again, it would make people wary about using their phones for illegal activity like calling in bomb threats.

This gets both parts of the solution: don't have emergency services overreact, and find and prosecute people who make false calls. Overreaction- sending in the SWAT team with hair triggers- is a key reason this is worse in the US than elsewhere, and we need to stop letting our police shoot first and ask questions later.

We also need to do something about malicious calls. This is not a harmless prank. At the very least it's harassment and abuse of government resources; at the worst it's (attempted) murder. If we start devoting serious resources to finding the perpetrators and prosecuting them, people will mostly stop trying this stuff because they'll fear the consequences. Of course it would help if we made our phone system harder to spoof, since faking the source of the call is part of the way perpetrators think they'll get away with it. Making calls harder to spoof would have a huge number of other benefits, too.

Yes, we did need a formal study. In the absence of proper scientific studies, it's easy for people to confirm their preexisting beliefs. If they a true believers that AGI is just around the corner, they claim the success of LLM proves we're almost there. If they think AGI is a pipe dream, they dismiss the success of LLM as fooling people with a souped-up autocomplete model. A scientific study can't actually answer the question of whether AGI is coming soon, but it can answer questions like whether LLM have done subsidiary tasks like building a coherent model of the world. When we learn that it hasn't, it affects our beliefs in whether AGI is coming soon or not.

Now all my crappy code will crash much faster. ;)

I made my chat font Comic Sans because I think it's funny that the "Supreme Allied Linux Commander" for our organization (me) uses Comic Sans on Skype For Business. Keeps 'em on their toes.