Browse Comments

I wonder what would happen if Eg. Disney built their own AI, trained it only on their own IP/Copyright material, and used it to help create art for a new movie or show? Or go one step further, say they create a new character, and commission 100 pieces of art to be used solely to train the AI to continue to produce art of the new character? They own the art, and therefor the training data, th…

There are no arbitrary lines, technology changes but the fundamental truth that you are responsible for your actions doesn’t. This is akin to harassment and it has nothing to do with AI. If you used pen and paper to achieve the same result it would still be harassment. This doesn’t have to do anymore with freedom of speech because you are no more “free” to do this than you are free to draw nude…

Autonomy in this case is referring more to autonomous decision making, I believe. Whenever you fire off a missile the targeting parameters are decided beforehand, in a clear way: Harpoons lock based on radar signature, if memory serves, Tomahawks are GPS guidance with terminal radar. Even the AIM-9X or AMRAAM are given pretty specific parameters. The autonomy we’re talking about here is a step be…

Most people have absolutely no idea how LLMs work, who the top players are, how they are trained, actual usecases, probable dangers, etc. yet will very confidently speak about them =P

The idea that LLMs are stubborn and need to be coerced is actually a confusion of what’s really going on with the instructions here. It’s not that LLMs are so stubborn that you need to prompt like this to get them to obey. It’s that LLMs are so easy to manipulate and make obey, that the company providing access has a hard time keeping YOU, the user, from complete control. What you are seeing her…

I asked it to write a story about a snake person with color-changing eyes, like the snakes from the Robin Hood and Jungle Book cartoons, mind-controlling a woman. It told me no, that it wasn't allowed to make pedo content, and sent me the links for child abuse hotlines. This implies that Gemini thinks that women are a kind of child, which explains its over-protectiveness of them, I guess.

My worry is that the AI doesn't actually have to be *better*, doing an awful job for 1% of the cost might still be profitable for the company. The minimum character limit on comments forces people to inflate their comments with meaningless words, even if those words don't add anything of value.

> There's even external/operational threats like mass civil unrest when AI takes too many jobs and governments fail to implement social safety nets or some form of UBI. This is the one that way too many people ignore, we're already entering the beginning of the end of many service and skilled labor jobs, and much of the next level of work is already being contracted out in a race to the botto…

That’s actually more of a valid criticism than you might think. I’m currently in a Master’s program for a pretty technical field, and AI plagiarism is rampant but identifiable because of this.  Probably 80% of the students speak English as a second language, but the “discussion” portion of the assignments are all submitted with perfect grammar using the same tempo. The school is doing their best …

My wife does proposal writing, And she is terrified that she is going to be replaced by AI in the next couple of years. However, to replace your entire team with AI at this point when the AI isn't capable of high school level work, probably means you're company is run by a moron.

I never used AI image creator until the other day. Someone wrote a comment about Judge Thomas bathing in hotdog water. I thought it would be fun to see the image conjured up in my head. Just a few clicks and I had my choice of a few. Scary.

Have a few friends who work trade; if for whatever reason US OSHA did inspection at Korean job sites, they would be too exhausted to list all the violations happening

>Something even more malicious: once a user is hooked, can the company use the emotional attachment to the AI to persuade or coerce the user into doing something like vote differently? You say this like social media doesn't already do this.

My new phone came with Gemini, Google's AI assistant. I don't use the assistant feature often, but I've only used Gemini twice, and both times it's returned false answers. First asked which post office in my area was open after 3pm on Saturday. I knew I would have to drive across town but wanted to double check, and it returned several post offices that were already closed. The other day I asked …

Any CEO who got exposed to the many pipe dreams their latest overpriced consulting agency sold them is feeling as if they can already reduce their head count with AI and speak of that in a few more years they could layoff a half of their employees because „AI can do it faster and better than them“ Too much greed and overinflated confidence rots the brain

Same here. Literally does nothing useful for me except when I need some boilerplate and sometimes unit tests. But I’m also working with hard crap mostly, trying to ask AI will just make it invent some shit that probably means I’ll waste hours (trust me I’ve made that mistake before). Maybe front end-andys have a better time with AI, IDK.

Go to Riga, Tallinn, or Vilnius and try saying something like people are historically russian or they belonged to the ussr, and there’s a small chance you’ll get a swig of vodka from some gopnik or a smelly hobo but more likely than not you’ll get your teeth kicked in or arrested.

Me when I accidently auto-complete with AI: "FUCK!" Also me: "I'll turn that off later."

***”What does this neural howlround look like exactly? [My friends and I engaged with it in a non serious way, and after two prompts it was already encouraging us to write a manifesto or create a philosophy.”*** I just read the conversation. You literally led the AI where you wanted it to go. You told it to create you a philosophy and the AI based it on the previous context which is normal. The …

Legal is hitting this now, where ChatGPT 4.0 and 4.5 both pass the bar exam more consistently than law students do, and those models are also better outta the box at most legal tasks than someone with 0-2 years experience.