vrighter

21 ms·

541.
by vrighter 11mo ago
"RAM is not really something you explicitly ration." It most certainly is. My old pc ran on 8MB of ram. Modern ones need 16GB for a comfortable experience. They do not do much more than I needed back then. I think it's reason
542.
by vrighter 11mo ago
You can sideload, I mean INSTALL, software on any linux desktop. Yet there are still tons of people saying that desktop linux has gotten good enough for most of everyone's grandma to daily-drive.
543.
by vrighter 11mo ago
When programming stuff as a hobby, I do always plan to throw one away. The first one is where I learn my lessons and write enough spaghetti until I fully understand the problem. Then I delete the first one, and start over with the lessons l
544.
by vrighter 11mo ago
Modifying the hardware != running the software I want on the hardware I own. So yes, your last statement is correct, but utterly irrelevant to this discussion.
545.
by vrighter 11mo ago
But sony, microsoft and nintendo all heavily restrict what you can use their computer for, and what software you can run on it. This is not a free-market issue. Yes, I am still free to buy another device. But if all device makers heavily re
546.
by vrighter 11mo ago
So it's just a lot of hot air, simply because it is not the government that is restricting our right to compute. But the device makers, and software developers. Google deciding to monopolize app installations is a restriction on comput
547.
by vrighter 1y ago
Just like the 90s are more than a decade ago, and we all need to get to terms with that, so is the fact that crypto is not new. if it's still trying to find a valid use case, then that's a bad thing because cryptocurrencies are o
548.
by vrighter 1y ago
Why the fuck would I want an AI to play my games for me?????
549.
by vrighter 1y ago
Simple framework: You must not attempt to offload the actual thinking part to someone else. You should be able to answer any reasonable question about the code you submitted. If you can't you should be banned from any further contribut
550.
by vrighter 1y ago
They can't learn iterative algorithms if they cannot execute loops. And blurting out an output which we then feed back in does not count as a loop. That's a separate invocation with fresh inputs, as far as the system is concerned.
551.
by vrighter 1y ago
"Knowing what they don't know" is one of the big unsolved problems in llm training. There's a finite number of facts. There's an infinite number of non-facts. If it could do what you're suggesting, it wouldn&#x
552.
by vrighter 1y ago
Even if they had a magic wand, they still couldn't make them perfect. Because they are by nature, imperfect statistical machines. That imperfection IS their main feature.
553.
by vrighter 1y ago
I made a submission too! I suggested a "sausage of the day" section.
554.
by vrighter 1y ago
Implementing one also finally gave me a way to intuitively grasp (and remember) the chain-rule from calculus.
555.
by vrighter 1y ago
And where do the researchers supposed to come from, exactly?
556.
by vrighter 1y ago
You don't have to, but of that list at the beginning of your comment, I actually did study them a bit on my own just for shits and giggles. And I am grateful for that because even if I don't use them directly, they have directly i
557.
by vrighter 1y ago
The thing is that if you actually learn how they work, they lose all of their magic. This has happened to anyone I know who bothered studying them and is not selling them . So I'd rather people learned. Knowing how a drill works doesn
558.
by vrighter 1y ago
The problem is that not even the ones building them can understand it. Otherwise they wouldn't be breaking their "I promise AGI by the end of next year" promises for the Nth time. That or they are flat out lying. My money&#x2
559.
by vrighter 1y ago
I hate that argument too. The computer is there to serve ME, not to serve the software it runs. Each application should be using as little of the hardware as possible, to make it possible for ME to make use of all the software.
560.
by vrighter 1y ago
I had spent ages optimizing my maze solver ( https://github.com/VrIgHtEr/TheFarmerSolvedAMaze ). Meticulously going over which operations take how many ticks. But then the dev made a bunch of breaking changes and could
561.
by vrighter 1y ago
He says a lot of things. He hasn't delivered on any of them lately. Probably won't either.
562.
by vrighter 1y ago
There's a very simple method that's been found to be effective by OpenAI researchers. It only involves one step: pulling shit straight of of your ass.
563.
by vrighter 1y ago
14 hours drops them to at least 99.8% uptime, over a whole year. Where will that be advertised?
564.
by vrighter 1y ago
Who cares what you built it with? You should put the product front and center, not the tools. The fact that you felt the need to mention it's in rust immediately leads me to think it's just one of those "faster, but feature i
565.
by vrighter 1y ago
3 years in? How long had you been hibernating for 3 years ago?
566.
by vrighter 1y ago
So they released the product before they even developed it. The sad state of software nowadays
567.
by vrighter 1y ago
this. I was running a reinforcement learning training run. I could very clearly hear from the coil whine whether it was simulating or backpropagating
568.
by vrighter 1y ago
Don't look up!
569.
by vrighter 1y ago
why do they keep saying "IF it fails to live up to the hype"? It's been over a decade, it's years past the point it was supposed to put all of us out of a job. And the same problems they had on day 1 are still there, mat
570.
by vrighter 1y ago
a model calls another (not self) model, which in turn returns without calling anything else. What you'ce discovered is called a function call. It simply hopes two drunks are more coherent than one drunk.

More ›