Ah yeah, anything under 7B gets kind of dicey imo. Especially if quantized.
Ah yeah, anything under 7B gets kind of dicey imo. Especially if quantized.
Interesting conclusion, LLMs are inherently 1D in nature, and ARC is a 2D task. LLMs are able to emulate 2D reasoning for sufficiently small tasks, but suffer greatly as the size of the task increases. This is like asking humans to solve 4D problems.
This is probably a fundamental limitation in LLM architecture and will need to be solved someday, presumably by something completely different.
Wasting time is the point. Presumably if you can waste four years most of the damage will have been avoided.
And more importantly, rich celebrities will sue you for defamation even if you’re right. That can cost a lot of money to defend against.
accounting for 15% of all cancer diagnoses in the US
Wow, that’s a significant portion of cancer this would be effective against. Assuming it still works for humans.
It didn’t make anything better?
Trump sees Russia and Israel both making land grabs and feels left out.
Presumably someone with better aim than the last shooter.
I’d be very careful as a wannabe dictator in a land with this many guns. The next four years are going to be tense for everyone.
They could still win (and lose) in three days. It all depends on who gets the first nuclear strike.
I think the explanation might be even simpler - right wing content is the lowest common denominator, and mindlessly watching every recommended short drives you downward in quality.
There is nothing wrong with having family ties to Nazis. You can’t choose your family.
There is everything wrong with doing Nazi things, regardless of who you are related to.
Works on some, but a lot of AI stuff uses a Speech to Text process to create the annotations themselves instead of trusting the provided subtitles.
Out of curiosity, what’s to hate about self checkout? It’s faster and you don’t have to talk to anybody.
Obviously preference matters so I’m not going to judge you for it!
Many camera for make many photo
I mean I’m personally happy I don’t need to have someone help me fill a fuel tank. I’m happy to do self checkout. Hell, I’d be happy to make my own sandwich at a restaurant if they’d let me.
I suspect we’ll have another Rosa Parks moment for the history books sometime in the next four years or so.
I mean it’s not crazy to say “you fix it, you assume liability” right?
Hmmm, I don’t think I will.
Pubkit.net looks quite useful. I may be testing that out soon.
I think so, with reasonable quantization. I’ve used 34B LLMs on my GPU. That being said, most models can run on CPU okay - they’re more memory constrained than compute constrained.