Research map / Large language models / Instruction and fine-tuning
ESG-Bench: Benchmarking Long-Context ESG Reports for Hallucination Mitigation
Siqi Sun, Ben Peng Wu, Mali Jin, Peizhen Bai, Hanpei Zhang, Xingyi Song
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add four-step CoT fine-tuning over plain SFT (ablation) | Gemma-2-2B-it · ESG-Bench | overall accuracy (%): 63.33 -> 92.00 | Improved |
| add four-step CoT fine-tuning over plain SFT (ablation) | Llama-3.2-3B-Instruct · ESG-Bench | overall accuracy (%): 90.67 -> 96.00 | Improved |
| add four-step CoT fine-tuning over plain SFT (ablation) | Mistral-7B-Instruct-v0.3 · ESG-Bench | overall accuracy (%): 80.00 -> 90.00 | Improved |
| four-step instead of two-step CoT fine-tuning (ablation) | Gemma-2-2B-it · ESG-Bench | overall accuracy (%): 72.67 -> 92.00 | Improved |
| add two-step CoT fine-tuning over plain SFT (ablation) | Llama-3.2-3B-Instruct · ESG-Bench | overall accuracy (%): 90.67 -> 92.33 | Improved |
| supervised fine-tuning instead of no fine-tuning (ablation) | Gemma-2-2B-it · ESG-Bench | overall accuracy (%): 71.00 -> 63.33 | Hurt |
| supervised fine-tuning instead of no fine-tuning (ablation) | Llama-3.2-3B-Instruct · ESG-Bench | overall accuracy (%): 76.00 -> 90.67 | Improved |
| supervised fine-tuning instead of no fine-tuning (ablation) | Gemma-2-2B-it · HaluEval | overall accuracy (%): 90.00 -> 67.00 | Hurt |
| supervised fine-tuning instead of no fine-tuning (ablation) | Gemma-2-2B-it · BioASQ | overall accuracy (%): 89.75 -> 74.50 | Hurt |
| add four-step CoT fine-tuning over plain SFT (ablation) | Gemma-2-2B-it · HaluEval | overall accuracy (%): 67.00 -> 94.00 | Improved |
| add two-step CoT fine-tuning over plain SFT (ablation) | Llama-3.2-3B-Instruct · BioASQ | overall accuracy (%): 81.25 -> 81.00 | No clear effect |
| add two-step CoT fine-tuning over plain SFT (ablation) | Gemma-2-2B-it · HaluEval | overall accuracy (%): 67.00 -> 69.25 | No clear effect |
Closest papers by meaning
- Benchmark Health Index: A Systematic Framework for Benchmarking the Benchmarks of LLMs (NeurIPS 2026 · similarity 0.84)
- The Labeling Problem in Hallucination Detection Benchmarks: An Empirical Evaluation (NeurIPS 2026 · similarity 0.84)
- OntoFact: Unveiling Fantastic Fact-Skeleton of LLMs via Ontology-Driven Reinforcement Learning (AAAI 2024 · similarity 0.82)
- Externalized CPDAG Summaries Improve LLM Causal Deduction (NeurIPS 2026 · similarity 0.82)
- INSIDE: LLMs' Internal States Retain the Power of Hallucination Detection (ICLR 2024 · similarity 0.82)
- CASE-Bench: Context-Aware SafEty Benchmark for Large Language Models (ICML 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
