Thought on working

In today’s efficient work environment, the main focus is to review the performance of jobs generated by AI.

I recently found in my work that tasks which used to take a long time can now be completed quickly using GPT or other AI tools. The main issue is that the content generated by it is not often satisfactory, and it always requires a lot of my time to review, which is actually quite tiring.

If I could precisely control the meaning it expresses and use a pipeline approach to help me read its output, that would be much better.

Additionally, regarding text, it becomes a bit uncomfortable after reading too much. Every day, I need to read a large amount of text, and I can’t use AI to summarize it because the summarized content doesn’t allow me to see the specific details.

My most used prompt in GPT4 is ‘fix errors only’ because most of the time, GPT will alter your original words and make them look like AI-generated, which is unacceptable.

Today, it took me only about $400 to build a desktop computer and use LLAMA 2 variant models to automatically generate content continuously, which is very meaningful for a baseline estimation: you can spend much less and build a working machine endlessly working for you, only consuming electricity. Currently, the mistral type of models works as well as GPT-4, only using your home computer, which is horrifying.