> I find it slightly reductive of you to say that’s ChatGPT’s single purpose.
I knew instantly upon reading the title that the author was making no single-purpose indictment of ChatGPT. In the body, there are many passages that make it clear, in the context of pointless corporate sausage-making, this tool has been ideal.
> I find it slightly reductive of you to say that’s ChatGPT’s single purpose.
I knew instantly upon reading the title that the author was making no single-purpose indictment of ChatGPT. In the body, there are many passages that make it clear, in the context of pointless corporate sausage-making, this tool has been ideal.
> Github Copilot is incredibly effective ...
Yes - because it's a co-pilot. ChatGPT is at best, a demo. It has no fitness-of-purpose like a co-pilot. Untouched, untamed, and in the wild, ChatGPT is perfect AS IS for some work.
co-pilot send every line of code that you will ever type home to GOOGLE for code-training, ergo they will own your algo
anybody that is stupid enough to code and give their code away for free to google, is well a fool
...
chat-GPT is a LIE, that's why its BULLSHIT
open-AI is not open, and the LLM tech is not AI, its 20 year old look-ahead mapping using recursion algos to minimize a matrix, at best you could argue that the novel 30B, 70B & 300B is only possible with cheap A100 class Tesla NVIDIA machines allow TB matrixes, otherwise we were all doing this stuff +15 years ago, but we called it LSTM, sure the 'prompt' is a new paradigm, but IMHO things have gotten worse the stuff we were running +10 years ago could do math, the new chatGPT can't even do binary arithmetic, let alone multiplication, ... Bullshit
LLM will never be self-aware or AGI, impossible, its just a random number generator that has an optimized solution to generate human preferable word orders
The training for chatGPT being facebook&twitter means you have a chat-bot with sub 80 IQ, and that all you will ever have; is a chat bot;
Lots of people on hugginface are figuring out that you can train with real books say feynman lecture on physics and actually have something a real expert, using small 7B models that you can train on a home GPU
>... co-pilot send every line of code that you will ever type home to GOOGLE for code-training
To be clear, I wasn't defending co-pilots; simply pointing out that stark difference between things that are designed to serve specific purposes and the giant ball of BS that ChatGPT uses to make everyone feel empowered.
>... its just a random number generator that has an optimized solution to generate human preferable word orders
And they wonder why all the hallucinations. There are no hallucinations. There are only very fleeting moments of clarity that may as well be random.
> I find it slightly reductive of you to say that’s ChatGPT’s single purpose.
I knew instantly upon reading the title that the author was making no single-purpose indictment of ChatGPT. In the body, there are many passages that make it clear, in the context of pointless corporate sausage-making, this tool has been ideal.
> Github Copilot is incredibly effective ...
Yes - because it's a co-pilot. ChatGPT is at best, a demo. It has no fitness-of-purpose like a co-pilot. Untouched, untamed, and in the wild, ChatGPT is perfect AS IS for some work.
co-pilot send every line of code that you will ever type home to GOOGLE for code-training, ergo they will own your algo
anybody that is stupid enough to code and give their code away for free to google, is well a fool
...
chat-GPT is a LIE, that's why its BULLSHIT
open-AI is not open, and the LLM tech is not AI, its 20 year old look-ahead mapping using recursion algos to minimize a matrix, at best you could argue that the novel 30B, 70B & 300B is only possible with cheap A100 class Tesla NVIDIA machines allow TB matrixes, otherwise we were all doing this stuff +15 years ago, but we called it LSTM, sure the 'prompt' is a new paradigm, but IMHO things have gotten worse the stuff we were running +10 years ago could do math, the new chatGPT can't even do binary arithmetic, let alone multiplication, ... Bullshit
LLM will never be self-aware or AGI, impossible, its just a random number generator that has an optimized solution to generate human preferable word orders
The training for chatGPT being facebook&twitter means you have a chat-bot with sub 80 IQ, and that all you will ever have; is a chat bot;
Lots of people on hugginface are figuring out that you can train with real books say feynman lecture on physics and actually have something a real expert, using small 7B models that you can train on a home GPU
>... co-pilot send every line of code that you will ever type home to GOOGLE for code-training
To be clear, I wasn't defending co-pilots; simply pointing out that stark difference between things that are designed to serve specific purposes and the giant ball of BS that ChatGPT uses to make everyone feel empowered.
>... its just a random number generator that has an optimized solution to generate human preferable word orders
And they wonder why all the hallucinations. There are no hallucinations. There are only very fleeting moments of clarity that may as well be random.