This is sad, actually, because this very technology is absolutely fantastic at identifying things in images. That’s how image generation works behind the scenes!
ChatGPT screwed this up so badly because it’s programmed to generate images instead of using reference images and then identifying the relevant parts. Which is something a tiny little microcontroller board can do.
If they just paid to license a data set of medical images… Oh wait! They already did that!
People should mostly be using a RAG retrieval system, not pure LLM slop like this, for reference. It just hasn’t really been made at scale because Google Search functioned as that well enough, and AI Bros seem to think everything should be done within LLM weights instead of proper databases.
I mean… WTF. What if human minds were not allowed to use references?
WolframAlpha was kinda trying to build this, but stalled.
Maybe it knows but doesn’t care. The general idea of meatbag anatomy is that they are full of soft squishy tubes and they die easy. Who cares about the details?
For images, it’s not even data collection because all the images that are used for these AI image generation tools are out on the internet for free for anyone to download right now. That’s how they’re obtained: A huge database of (highly categorized) image URLs (e.g. ImageNET) is crawled/downloaded.
That’s not even remotely the same thing as “data collection”. That’s when a company vacuums everything they can from your private shit. Not that photo of an interesting building you uploaded to flickr over a decade ago.
This is sad, actually, because this very technology is absolutely fantastic at identifying things in images. That’s how image generation works behind the scenes!
ChatGPT screwed this up so badly because it’s programmed to generate images instead of using reference images and then identifying the relevant parts. Which is something a tiny little microcontroller board can do.
If they just paid to license a data set of medical images… Oh wait! They already did that!
Sigh
Yeah, I mean, this can be done with text too.
People should mostly be using a RAG retrieval system, not pure LLM slop like this, for reference. It just hasn’t really been made at scale because Google Search functioned as that well enough, and AI Bros seem to think everything should be done within LLM weights instead of proper databases.
I mean… WTF. What if human minds were not allowed to use references?
WolframAlpha was kinda trying to build this, but stalled.
That “car” is a truck.
Truck ⊂ Car
source
That’s just the botanical definition of truck. Culinarily, a truck is a car.
Even this one?
Garnish should be edible, so yes, but culinarily it’s not very palatable. No flavor.
A bus is a truck?
A car with a stowable seats and 4wd is a truck?
A bus is certainly a truck.
According to this chart, that car with 4WD may be a truck depending on some math about it.
lmfao a bus is a truck, you gotta be kidding me
what kinda braindead US definition is that?
behold, a truck
or this?
Fair enough lmao TIL
That’s a car
It is in my opinion. Put this in perspective…
These are train cars.
These are road trains.
But you might call them something else.
Truck:
Car:
Maybe it knows but doesn’t care. The general idea of meatbag anatomy is that they are full of soft squishy tubes and they die easy. Who cares about the details?
It just took a bunch of results from a google search and slapped shit together
How much of the public outcry against data collection has resulted in us getting an inferior product publicly.
Alright guys, I found Sam Altman’s lemmy account!
For images, it’s not even data collection because all the images that are used for these AI image generation tools are out on the internet for free for anyone to download right now. That’s how they’re obtained: A huge database of (highly categorized) image URLs (e.g. ImageNET) is crawled/downloaded.
That’s not even remotely the same thing as “data collection”. That’s when a company vacuums everything they can from your private shit. Not that photo of an interesting building you uploaded to flickr over a decade ago.