level 0

• zero shot this https://es.wikipedia.org/wiki/Falacia and prompt to use them to explain the falacies and remove them from the statement.
• make a small app/website to use it

level 1

• make a twitter bot to use it. https://developer.twitter.com/en/docs/tutorials/how-to-create-a-twitter-bot-with-twitter-api-v2

level 2

• make mixtral generate statements filled with each of the falacies, and then to explain why the statement is a falacy. This dataset can be statements and explanations per each
  falacy, or statements with many falacies. Then train a qlora on top of mixtral or llama2.
• use zero shot on top of qlora. Maybe compress the list of falacies before zero shoting them.
• check how the heck to host an os llm with a qlora (https://medium.com/@yuhongsun96/host-a-llama-2-api-on-gpu-for-free-a5311463c183 or exprimir some aws free credits or something else)