Me: I finally made a profit from my game that took my 6 years of developing Unity: AND HIS NAME IS JOHHNNNN CENAAAA!! *DU DU DU DOOOOOOO DUE DUE DUE DYOOOOOO DA DA DA DYAHHAAAAAA BAM BAM BUM BHAAAAAAAAHH*
Java can do everything cleanly and has unlimited access to pre-built libraries. For higher level language, why would you need anything else? The only short fall in java is no primitives in generics, for which it takes a hit on performance if you want to keep the API clean.
I think one reason that ai art could possibly never be the same as real art, is that it's not like humans when they do art are just projecting a visualization in their head onto a paper. I mean, thats kinda the goal, but it's not how it works.
There's just not enough data, and there never will be, since we won't be able to get more data. Humans won't put more and more information online and it's simply impossible that robots can start autonomously exploring the world to gather more data...
At some point we have to be honest with what intelligence is. I love the "Draw a centaur" joke/example because There would definitely be a difference in asking a human artist who knows nothing about fantasy to draw a "centaur" vs. draw a "creature where the lower body is of a horse from the base of the neck down and the upper body of a human from the waist up." Is this prompt engineering or is it intelligence? Even in the existing LLM models, I believe there are things to discover if we can give meaning to it - but at the same time - if we asked the LLM to tell us (if this were somehow possible without brute-forcing the input tokens) all that it "knew", would there be any meaning to us as humans without it looking like books from the library of Babel?
Okay, but what if they let it collect the data? Like, give it access to thousands of robots of all kinds that would explore the real world? How about this way of data collecting? Roaming the world, asking questions about unknown, experimenting and figuring things out - train itself this way?
A version filed in every record? No... you know the whole data set was written with a specific version out of band. You don't need to encode it in every single record. If you are going to change versions, you will sync the existing table, fire up a new version, and tell it to start writing the new format here, and all of these historical files are in the old format. You need a version number in real time network protocols only to allow two different versions that don't otherwise know they are speaking with a client running a different version. With a database you can just put the version in your config file when telling the database to open an old table and you're done.
The leaders in this space are (saying that they are) discovering that theirs still a lot of room for improvement with more data and whilst I'm still willing to believe them, if you look at how easy it is for entities in this space tow discerption to grab market share or funding without the care of what the fall out will be if they're claims don't pan out then it would be foolish to not be skeptical. All that said though, you look at the various models that are less mature than LLMs and problems left to be solved on the way to AGI and you can't help but to feel like this stuff is still in the early days and there's a lot of room for improvement. Remember that 18 months ago, it seemed like a limit was much closer then than it is today and that was partially because there was issue with the amount of quality test data available for larger parameter models.
One use case for String constructor is if you want to store other properties against the string (eg isTranslated) let s = "hello"; s.isTranslated = true console.log(s, s.isTranslated) s = new String("hello") s.isTranslated = true console.log(s, s.isTranslated)
Your story about the horror of punch cards is true. It happened to everyone who every used them at some point. How about that business grad student who wrote thier compiler in COBOL? It was four BOXES of cards.
Other than the then-valid criticisms about Go's pitfalls, everything else basically came down to "I'm a code golfer, I should be able to write the most amount of logic in the smallest space so it looks like I'm way more clever than I might actually be"
Re: CS Lewis, funny thing. We took my toddler daughter to see the big blue whale at AMNH. We kept pointing at this giant whale and showing her, but she barely looked at it bc it was so vastly beyond her comprehension. Her 1 year old brain couldn't begin to parse that it was an animal.
recommendations account for like 20% of revenue for these companies, but they’re basically just market basket analysis classifications. Same with amazon “people also bought. . .”
AI is turning into rain man where it is so filled with information it can't make sense of it. Specialized AI will be far superior. Humans have to specialize. AI does, too. Questions are going to be routed to specialized AI's that are best able to answer a question or draw something.
"If you've ever seen something brand new, it's like you can't even see it." Why do people say this? I have never in my life experienced this. Yes, experiences change with familiarity -- you start noticing patterns or more subtleties in something. But to be totally unable to perceive something? No. Not once to my knowledge. And if I had been unable to perceive it, how would I know? This really sounds like some pseudophilosophical bullshit.