There are instruction-based AI models and reason-based models you can use. The better ones use more memory and resources. I'll be trying both as lower-level models due to the limitations of my system.
Some of what it is saying is pretty interesting and could help with writing if I run into writer's block. Lots of inaccurate statements, but it seems to depend on how I word a question or give it instructions.
I'm using a reasoning AI "model" called DeepSeek that was created and developed in China. It can do a fair summary of written material, but when you ask it to come to a conclusion about a story or article, it can misunderstand what it is reading and come up with some crazy shit. It is almost like how a toddler understands words and creates sentences.
Edit to add: The AI program needs you to embed the data from your documents, and then it can search through large collections of documents to find what you ask it for. Giving it a command to find a topic in a collection works well, until you start asking it questions that require more than a summary.
At that point it only probes the embedded data enough until it can come up with a somewhat reasonable answer. It seems to pick up on no more than two or three files near the beginning of the collection to answer more complex questions. I am going to try using one big file at a time in the embedded collection so it can't just pull one or two near the beginning of 60 or more files.
Some of what it is saying is pretty interesting and could help with writing if I run into writer's block. Lots of inaccurate statements, but it seems to depend on how I word a question or give it instructions.
I'm using a reasoning AI "model" called DeepSeek that was created and developed in China. It can do a fair summary of written material, but when you ask it to come to a conclusion about a story or article, it can misunderstand what it is reading and come up with some crazy shit. It is almost like how a toddler understands words and creates sentences.
Edit to add: The AI program needs you to embed the data from your documents, and then it can search through large collections of documents to find what you ask it for. Giving it a command to find a topic in a collection works well, until you start asking it questions that require more than a summary.
At that point it only probes the embedded data enough until it can come up with a somewhat reasonable answer. It seems to pick up on no more than two or three files near the beginning of the collection to answer more complex questions. I am going to try using one big file at a time in the embedded collection so it can't just pull one or two near the beginning of 60 or more files.
A trail goes two ways and looks different in each direction - There is no such thing as a timid woodland creature - Whatever does not kill you leaves you a survivor - Jesus is NOT a bad word - MSB