Regarding the timeline
by akarl (2023-03-25 15:32:53)

In reply to: Think about all the millions of people who have ...  posted by Ofcr. Tim McCarthy


I spent a lot of time a few years ago working on some applications of recurrent neural networks in Keras. These are a different architecture of neural network than used by GPT and DALLE, but they have a lot of similarities (as of 2017 it was the model structure used by Google translate). So I was able to get some insight into the potential of text and image analysis and generation. If you had asked me before I first looked at these programs a few weeks ago, I would have guessed that the kind of functionality I am already seeing when playing around would have been 10 years away. So I can't imagine what things will look like in 10 years.

There are still enough errors in the output to require expertise in sporting and fixing them, but even now this could represent a huge reduction in programming time just from getting the initial libraries and functions setup instantly while also minimizing typos and bugs.