zurichrain / agi-hallucination

A Survey of MultiModel LLM Hallucination

AGI-Hallucination

Please check recent paper index list in PaperList.

We also present a bird’s eye view of hallucinations in AGI and compile a comprehensive survey. LightHouse: A Survey of AGI Hallucination

📖 Once this problem is solved, the path to AGI unfolds

We called it LightHouse for AGI

✨ Nowadays, we defined hallucinations as: Model outputs that do not align with the contemporary empirical realities of our current world.

• Definition for AGI Hallucination

  • Conflict in Intrinsic Knowledge of Models
  • Factual Conflict in Information Forgetting and Updating
  • Conflict in Multimodal Fusion

• Emergence for AGI Hallucination

  • Training Data Distribution
  • Timeliness of Information
  • Ambiguity in Different Modalities

• Mitigation for AGI Hallucination

  • Data
  • Train & SFT & RLHF
  • Inference & Post-hoc

• Evaluation for AGI Hallucination

  • Benchmark
  • Rule-Based
  • Large Model-Based
  • Human-Based

• Talk about Future

  • Dataset for 3D, Ego-Centric, Video, and Agent
  • Mutimodel Fusion, RLHF Refine, Objective Refine, Inference Refine
  • Knowledge Update, Catastrophic Forgetting
  • Balance Creation and Hallucintion

🔍 Case Study

🔮 Talking about Future

📲 Contact

We warmly welcome any kinds of useful suggestions or contributions. Feel free to drop us an issue or contact Rain with this e-mail.