That's literally not useful for the sort of applications people want to use transformers for.
As much as I love all these technologies, they don't have any direction that approaches the efficacy of a transformer
The answer to all ai worries is quite simple. Ai is a technology. While 'agents'may be deployed, the law simply needs to be clear as to who the human agent is who will be held responsible criminally.. Suddenly every…
Then he should resign
Forget capitalism. This is a call to end the basic right of people to multiply matrices
I specifically stick to monasticism because Christianity is the odd one out of all the abrahamic religions in advocating for monasticism, and monasticism started in Egypt which had significant enough presence of…
Christian and Buddhist monastic traditions are so uncannily similar it's difficult to really think they developed without influence. That's my opinion. Even the Europeans who first set foot in Asia during the age of…
I personally think thinking is basically variable but rate precision. If you are in a 4bit mode but need 2x as many tokens you're just doing fp8 with hoops( of course 4bit multiply is faster)
> while you yourself do the same about modern composers. I literally gave several examples of modern genres in which classical composers still produce good work. I just said they're unlikely to be found in academia
Pop music is also mostly trash too, but no one pretends otherwise. At least they have the decency to use auto tune.
> I know people who only listen to a single narrow genre or decade of pop music. How are they different? No one mistakes a pop artist or afficianado for some great connoisseur of culture.
Modern 'classical' music is a myth. In their day, classical composers were essentially pop performers. Women used to throw their undergarments at mozart... Today classical music is filled with people who want to pretend…
I mean at every layer in a transformer, the attention mechanism does a massive state transfer between tokens. In a recurrent transformer, instead of projecting from the latent space to token space after a fixed depth,…
The tokens inside the transformer are only projected into token space to train them. In reality they ought to be treated as their own thing. What's really gone on is you've trained the final projection to be sensible…
The hidden states of the tokens likely contain more semantic information than can be extracted by the final projection into token space.
I do the same thing in my head. There is no underlying logic to an llm. Logic is an external construct alien to human like forms of reasoning.
Why? I talk in my head and then enunciate only that which is relevant. My speech rate inside is incredibly fast.
Should have said it was post preschool psychosis
Same model. Same hardware... Depends. If you sacrifice speed then no. Ieee754 is pretty specifically specified, the issue is that it's not associative. If you get the associativity correct, then there's no issue.…
Llms are absolutely deterministic if you want them to be. Still a terrible idea for config
Whichever you want. There's a million text editors today because people have different approaches. I'm not criticizing having different approaches. I'm criticizing people who start writing their own text editor instead…
Exactly, this my point. If you want to make a great editor, the proper thing to do is start from a good base. Start from emacs and make something like doom emacs or spacemacs, etc. Or start from vim or whatever. Text…
It's just easy to install on Windows. Base emacs does exactly what you want it to do. And my emacs v vscode comparison was specifically because of infinite configurability and expandability. VSCode has the same thing…
My computer stays the same speed because I use software that just does the thing and not anything else. Native toolkits, single purpose, text based, etc. None of this is an issue if you just don't use the crap they push…
I am being pushed to use vs code right now by my team, but we already have a fully programmable and scriptable editor called emacs that is 100000x better. I don't understand why everyone just switches to these random…
That's literally not useful for the sort of applications people want to use transformers for.
As much as I love all these technologies, they don't have any direction that approaches the efficacy of a transformer
The answer to all ai worries is quite simple. Ai is a technology. While 'agents'may be deployed, the law simply needs to be clear as to who the human agent is who will be held responsible criminally.. Suddenly every…
Then he should resign
Forget capitalism. This is a call to end the basic right of people to multiply matrices
I specifically stick to monasticism because Christianity is the odd one out of all the abrahamic religions in advocating for monasticism, and monasticism started in Egypt which had significant enough presence of…
Christian and Buddhist monastic traditions are so uncannily similar it's difficult to really think they developed without influence. That's my opinion. Even the Europeans who first set foot in Asia during the age of…
I personally think thinking is basically variable but rate precision. If you are in a 4bit mode but need 2x as many tokens you're just doing fp8 with hoops( of course 4bit multiply is faster)
> while you yourself do the same about modern composers. I literally gave several examples of modern genres in which classical composers still produce good work. I just said they're unlikely to be found in academia
Pop music is also mostly trash too, but no one pretends otherwise. At least they have the decency to use auto tune.
> I know people who only listen to a single narrow genre or decade of pop music. How are they different? No one mistakes a pop artist or afficianado for some great connoisseur of culture.
Modern 'classical' music is a myth. In their day, classical composers were essentially pop performers. Women used to throw their undergarments at mozart... Today classical music is filled with people who want to pretend…
I mean at every layer in a transformer, the attention mechanism does a massive state transfer between tokens. In a recurrent transformer, instead of projecting from the latent space to token space after a fixed depth,…
The tokens inside the transformer are only projected into token space to train them. In reality they ought to be treated as their own thing. What's really gone on is you've trained the final projection to be sensible…
The hidden states of the tokens likely contain more semantic information than can be extracted by the final projection into token space.
I do the same thing in my head. There is no underlying logic to an llm. Logic is an external construct alien to human like forms of reasoning.
Why? I talk in my head and then enunciate only that which is relevant. My speech rate inside is incredibly fast.
Should have said it was post preschool psychosis
Same model. Same hardware... Depends. If you sacrifice speed then no. Ieee754 is pretty specifically specified, the issue is that it's not associative. If you get the associativity correct, then there's no issue.…
Llms are absolutely deterministic if you want them to be. Still a terrible idea for config
Whichever you want. There's a million text editors today because people have different approaches. I'm not criticizing having different approaches. I'm criticizing people who start writing their own text editor instead…
Exactly, this my point. If you want to make a great editor, the proper thing to do is start from a good base. Start from emacs and make something like doom emacs or spacemacs, etc. Or start from vim or whatever. Text…
It's just easy to install on Windows. Base emacs does exactly what you want it to do. And my emacs v vscode comparison was specifically because of infinite configurability and expandability. VSCode has the same thing…
My computer stays the same speed because I use software that just does the thing and not anything else. Native toolkits, single purpose, text based, etc. None of this is an issue if you just don't use the crap they push…
I am being pushed to use vs code right now by my team, but we already have a fully programmable and scriptable editor called emacs that is 100000x better. I don't understand why everyone just switches to these random…