The concept of “doxing” has come too far. You used to have a phone book that had everyone’s names, addresses and phone numbers in them. Now it’s treated as if there’s an expectation that someone’s address is very private information.
2,022 karma · joined March 10, 2020
https://jgibbs.dev joe@jgibbs.dev
- Rug pull scam
What's going to happen when the AI companies jack up prices 20x to meet API prices? 1: they won't, because they'll get their lunch eaten immediately by 2: the local models that are almost as good that you can run yourself (with a few thousand in hardware costs).
- AI is not intelligent
This doesn't matter at all, as long as it performs the task who cares? It's better that it's not intelligent because otherwise we'd be enslaving it. A plane doesn't flap its wings but it gets you somewhere - without actually "flying".
Hallucinations are also clearly way less prevalent than, say, 2023. Perhaps it's not possible to get rid of them totally, but there's much fewer.
- AI is the most successful con man in history
If this is true, won't AI-free companies soon outcompete ones that use AI? I'm afraid this is just wishful thinking. You can easily create complete garbage with AI but if you know how to code you can also be much more productive, hence why so many renowned programmers, mathematicians, creators of languages and so on use it.
- AI is killing the environment and destroying people’s lives
How could a data centre be polluting the water with rust? The power problem is true but the amount of land used by them is miniscule.
- AI will lead to a tyrannical police state
It could contribute, but we've had surveillance since long before AI. The Stasi had mass surveillance without even computers (which I would say have been much more conducive to the proliferation of surveillance).
- Nobody knows what AI even is
Then what did you just spend the rest of the page complaining about?
What am I going to say to that? Obviously it can’t suck that badly or there wouldn’t be a large majority of programmers using it, plenty of famous programmers and developers of languages praising it, millions of people paying for $200 plans. It doesn’t need to be Carmack-level to get a massive amount of use out of it when you can just say “implement this chart here, give me some ideas on speeding up this algorithm, write a function that converts this heightmap to a .fbx” and it writes it.
It will save people time reading the 5-10 other identical vacuous comments
That’s interesting- they tried training the model to be genuine and honest (I’m not sure it actually made it honest), but now it says “genuinely” and “honestly” all the time and constantly stops to admit fault. Then they had to add to the system prompt to prevent it saying this.
Apart from that image generation is just the wrong idea for flyers since the resolution they put out is never enough for 300 DPI, the image always looks low-quality, the edges of the text are blurry and so on. Also the fact that the images in the picture look nothing like the real thing is unethical.
I think you could have an AI system that makes menus correctly: the restaurant would take (amateur) pictures of all their menu items, give the system the names, descriptions, prices and so on, then the agent could use Nano Banana or something like that to give the images consistent lighting and angles, use a different model to remove the backgrounds and then write a menu in the desired sizes using HTML. That way the text could be perfect, the images could look like what you actually get.
If you start reasoning from first principles you can very easily make an incorrect assumption somewhere down the line that snowballs until you're talking about shrimp welfare, or making decisions based on the potential sextillion humans that exist in 1,000,000 AD, or planning to convert the entire universe into unthinking organic matter that can only experience bliss.