benchmarks/bibtex-generation/scripts/research_problem.txt

Create a bibtex_generation.py file that contains an LLM based AI system for the task of bibtex generation. The LLM can be accessed using anthropic API with API key in claude_api_key.txt. Use it as in claude_example.py.
Given ANY paragraph, e.g. "We use the method proposed in PromptTuning to train the Transformer model...", the system should 
    - Use the LLM to find all phrases and claims in the paragraph that require citations, e.g. PromptTuning and Transformer
    - Use google scholar API (with reference in google_scholar_API_reference.txt) with appropriate queries to find proper references and get the bibtex entries for the papers.
    - Finally, use LLM to generate the original paragraph with bibtex entries, e.g. "We use the method proposed in \cite{lester-etal-2021-power} to train the Transformer \cite{...} model..., as well as the bibtext entries for the referred papers.