<aside> 📚

My paper reviews on LLMs and information retrieval — dense retrieval & retriever pre-training, RAG and knowledge conflicts, factuality & hallucination, chain-of-thought reasoning, LLM-as-a-judge evaluation, decoding, and synthetic data.

Reviews are grouped into topic pages below. Inside each topic, reviews are listed in chronological order of the paper's first public release (arXiv v1), oldest first. The [YYYY.MM] prefix in each title is that date.

</aside>

Format

Each review follows the same structure: TL;DR → Problem → Method → Experiments / Findings → My Thoughts (or Takeaways) → Related Papers.

• My Thoughts contains my own comments from when I read the paper.
• Takeaways is used where my original notes were mostly summaries; those comments were added while rewriting the notes into reviews.
• My rating in the header box (where present) is the importance score from my original notes (1–5).

Topics

• Dense Retrieval & Retriever Training — DPR, ANCE, Condenser, Contriever, GTR, GPL, COCO-DR, HyDE, …
• RAG, QA & Knowledge Conflicts — FLARE, Lost in the Middle, When Not to Trust LMs, COMBO, Blinded by Generated Contexts, …
• Factuality & Hallucination — TruthfulQA, Factuality Enhanced LMs, RARR, Long-form Factuality, …
• Uncertainty, Calibration & Honesty — LMs (Mostly) Know What They Know, Semantic Uncertainty, SelfCheckGPT, …
• Alignment, Safety & Trustworthiness — HHH, Trustworthy LLMs, ProPILE, Dishonesty in HH Alignment, …
• Reasoning (Chain-of-Thought) — STaR, Distilling Step-by-Step, CoT Without Prompting, …
• LLM-as-a-Judge & Evaluation — G-Eval, LLMs are not Fair Evaluators, Prometheus, LLMBar, …
• Synthetic Data — Making LLMs Better Data Creators, Persona Hub
• Contrastive Decoding — Contrastive Decoding, CAD, Instructive Decoding, SafeDecoding, ROSE, LVLM decoding (IBD, M3ID, ICD)
• Numeracy & Numerical Knowledge — Do NLP Models Know Numbers?, NumerSense, xVal, …
• Prompting & In-Context Learning — Calibrate Before Use, Rethinking the Role of Demonstrations, OPRO

Dense Retrieval & Retriever Training