Post
11
š¤ We train LLMs ā but how do LLMs reshape AI research itself?
Sharing our new survey, AI4AIR: A Comprehensive Survey on Large Language Models for AI Research š
Rather than treating LLMs as outside helpers for writing or literature review, we map how they enter the ML pipeline itself ā through a 2D taxonomy (research domain Ć pipeline stage) and 5 recurring roles:
š·ļø Annotator Ā· š§Ŗ Synthesizer Ā· āļø Optimizer Ā· š Evaluator Ā· šļø Orchestrator
Our key takeaway: how deeply an LLM can empower a stage is bounded by its validation cost ā which is exactly why data annotation & evaluation are already mature, while model design & full-loop orchestration remain hard.
š Paper (PDF): https://github.com/ICT-FinD-Lab/Awesome-LLMs-for-AI-Research/blob/master/PDF/AI4AIR_Survey_v260601.pdf
š Project: https://ict-find-lab.github.io/Awesome-LLMs-for-AI-Research
ā GitHub: https://github.com/ICT-FinD-Lab/Awesome-LLMs-for-AI-Research
Stars & feedback warmly welcome š
Sharing our new survey, AI4AIR: A Comprehensive Survey on Large Language Models for AI Research š
Rather than treating LLMs as outside helpers for writing or literature review, we map how they enter the ML pipeline itself ā through a 2D taxonomy (research domain Ć pipeline stage) and 5 recurring roles:
š·ļø Annotator Ā· š§Ŗ Synthesizer Ā· āļø Optimizer Ā· š Evaluator Ā· šļø Orchestrator
Our key takeaway: how deeply an LLM can empower a stage is bounded by its validation cost ā which is exactly why data annotation & evaluation are already mature, while model design & full-loop orchestration remain hard.
š Paper (PDF): https://github.com/ICT-FinD-Lab/Awesome-LLMs-for-AI-Research/blob/master/PDF/AI4AIR_Survey_v260601.pdf
š Project: https://ict-find-lab.github.io/Awesome-LLMs-for-AI-Research
ā GitHub: https://github.com/ICT-FinD-Lab/Awesome-LLMs-for-AI-Research
Stars & feedback warmly welcome š