Twitter bot for Art of War -- Part 2
This is the second blog in the series see the other parts here:
Summary so far:
Tried out aitextgen.io on Art of War and it is generating incomprehensible shit!
AITextGen with GPT-Neo
Here’s the Google Colab notebook, basically trained the AITextGen code on GPT-Neo, and it gave shit like this
China: not coming, but are five primary colors (1) Which of war the Moral Law causes, to Earth; (2) insubordination; (3) collapse; (4) ruin; (4) ruin; (5) ground; (5) disorganization; (4) ground; (5) disorganization; (4) rout.
Other conditions being equal, if one force is hurled against another ten times its size, the former.
When the former.
When the common soldiers are too strong and the result will be bad. When the officers of a feeling of resentment, the officers too weak, the honor which is prepared for the soldiers too strong and the soldiers too weak, the result is insubordination.
When the officers are likely to defeat.
When the officers are able to defeat, and their officers are angry and insubordinate, and you fail to defeat, and their officers and they will fail to defeat, and their officers and men, and you fail to defeat, and their lives are not coming, and their kingdoms.
When you fail to defeat, we may angry and do not go after you are angry and the enemy fail.
If you can do notSide notes:
It 40 minute to train the model on a paid Google Colab GPU
Next steps:
Split the sentences and train again
If doesn’t work, try a larger model
Also try for different authors as well, maybe Art of War is too obscure to produce sane texts on