Publications – Peng Yifeng

Lastest Update:4th July 2024 中文版本 (Chinese Version)

Paper On Working

Data Reshaping Productivity - Exploration of Embedding Data Elements into Production Functions
This is my attempt to approach the human-AI alignment issue from the perspective of development economics. During the writing process, I have unlocked many new ideas. This article will be heavily theoretical, representing a bold attempt on my part.
Tainted Future of Language Models, Collapse and Degeneration
The proliferation of hallucinatory data, a byproduct of model generation, insidiously taints the digital landscape. In this study, we introduce an innovative approach that leverages an infinitesimal fraction of the training corpus as a guiding catalyst to elicit hallucinatory outputs from the model. However, the human capability to rectify such distortions is inherently finite. This intrinsic constraint may, in turn, circumscribe the developmental trajectory of future models, posing a formidable challenge to their evolution and refinement.
Bon gpt, unexpected food journey from AIGC
In the realm of culinary recommendation, do LLMs possess a profound grasp of the intricacies of cooking, as well as the notions of taste and health? Could there be regional biases in food recommendations—might the model exhibit favoritism or prejudice towards certain national or regional dishes? This research is currently in its nascent stages. Should you be intrigued by the prospect of joining this endeavor, do not hesitate to reach out to me directly for collaboration.