Cell contamination (idk what to call it: when you use a variable from another old cell by accident, especially when you thought you deleted or redefined it in your new cell) is such a huge footgun that it makes me incredibly nervous to use notebooks for anything non-trivial.

What is indirect annotation?

- human makes an annotation guideline
- LLMs used to annotate data
- fine-tune a model like BERT

Referencing "LLMaAA: Making Large Language Models as Active Annotators" which uses KNN-based few-shots, multiple training cycles, and active learning.

#chr2024

sicko behaviour but understandable i also love bibtex cleaning

the time / avg load graph also looks cool !