Idk, to me there is this weird fringe of people that want computers to be this friendly “machine turns on and smiles and says hello and you talk to it”. But that dream (as far as I’ve always understood it) stems from Steve Jobs quotes about selling computers and products.
I’ve always used the computers we evolved today and have never had the desire to seek a friendship where the other “being” was the computer. It’s always been a tool and writing full sentences/paragraphs to get back correct information doesn’t feel like the next evolution of computers where before I was dropping keywords and filtering myself.
The examples in the article are designed to support the points made but are not remotely accurate.
> Consider the difference:
> GUI era: “Open Photoshop → Create new file → Set dimensions to 1200x628 → Select rectangle tool → Draw rectangle from coordinates (0,0) to (1200,628) → Fill with color #3B5998…”
> AI era: “Create a Facebook cover image with our company logo and a modern blue background.”
The LLM prompt in the article looks simpler but really there is a lot of hidden prompt describing the output which is probably the same info as the GUI era example. Which blue will my LLM pick and how will it know without me telling. How will it know how big to scale the logo and to tilt it 3 degrees left? How does it even know what the “company logo” is?
LLMs might also collect the wrong data. Does my LLM use Facebook header image size specs from 2015 or 2022? Most “blogs” online might be how-to blogspam with outdated answers.
IMO LLMs are an attempt to filter blogspam from search and make knowledge gathering and scraping of walled-gardens easier, less how everyone will be using the computer.
I’ve always used the computers we evolved today and have never had the desire to seek a friendship where the other “being” was the computer. It’s always been a tool and writing full sentences/paragraphs to get back correct information doesn’t feel like the next evolution of computers where before I was dropping keywords and filtering myself.
The examples in the article are designed to support the points made but are not remotely accurate.
> Consider the difference:
> GUI era: “Open Photoshop → Create new file → Set dimensions to 1200x628 → Select rectangle tool → Draw rectangle from coordinates (0,0) to (1200,628) → Fill with color #3B5998…”
> AI era: “Create a Facebook cover image with our company logo and a modern blue background.”
The LLM prompt in the article looks simpler but really there is a lot of hidden prompt describing the output which is probably the same info as the GUI era example. Which blue will my LLM pick and how will it know without me telling. How will it know how big to scale the logo and to tilt it 3 degrees left? How does it even know what the “company logo” is?
LLMs might also collect the wrong data. Does my LLM use Facebook header image size specs from 2015 or 2022? Most “blogs” online might be how-to blogspam with outdated answers.
IMO LLMs are an attempt to filter blogspam from search and make knowledge gathering and scraping of walled-gardens easier, less how everyone will be using the computer.