UncertaiNLP 2025 Paper Index

#	Paper Title	Authors	Relevance	Links	Key Connection to Entropy Engine
23	ERGO: Entropy-guided Resetting for Generation Optimization in Multi-turn Language Models	Khalid, Jeyaganthan, Do, Fu, Sharma, O'Brien, Zhu	CRITICAL	ACL: .../main.23/	Independent convergence with Entropy Engine. Uses Shannon H over next-token distributions, ΔH̄(t) as drift signal, non-prescriptive input restructuring. 56.6% performance gain, 24.7% aptitude increase, 35.3% unreliability reduction.	https://aclanthology.org/2025.uncertainlp-main.23.pdf
				arXiv: 2510.14077
26	DeLTa: A Decoding Strategy based on Logit Trajectory Prediction Improves Factuality and Reasoning Ability	He, Takase, Ishibashi, Shimodaira	HIGH	ACL: .../main.26/	Logit trajectory analysis across transformer layers for factuality. Validates logit-level monitoring approach relevant to Engine's planned hardware implementation path.	https://aclanthology.org/2025.uncertainlp-main.26.pdf
				arXiv: 2503.02343
21	Read Your Own Mind: Reasoning Helps Surface Self-Confidence Signals in LLMs	Podolak, Verma	HIGH	ACL: .../main.21/	Shows reliable uncertainty estimation requires explicit exploration of generative space. Semantic entropy validated as reliable signal. Supports Engine's premise that distributional monitoring reveals genuine model state.	https://aclanthology.org/2025.uncertainlp-main.21.pdf
				arXiv: 2505.23845
16	Confidence-Based Response Abstinence: Improving LLM Trustworthiness via Activation-Based Uncertainty Estimation	Huang, Datla, Zhu, Samuel, Liu, Kumar, Soni (Capital One)	HIGH	ACL: .../main.16/	FFN activations carry more info than post-softmax probabilities. Financial industry RAG application. Abstention principle: cost of wrong answer > cost of no answer. Validates Engine's deeper-than-probability monitoring and high-stakes deployment context.	https://aclanthology.org/2025.uncertainlp-main.16.pdf
				arXiv: 2510.13750
2	Phases of Uncertainty: Confidence-Calibration Dynamics in Language Model Training	Durai	MODERATE	ACL: .../main.2/	Uncertainty as training-dependent property, not static. Relevant to Engine's temporal dynamics approach—uncertainty changes over time, not fixed characteristic.	https://aclanthology.org/2025.uncertainlp-main.2.pdf
6	Certain but not Probable? Differentiating Certainty from Probability in LLM Token Outputs for Probabilistic Scenarios	Toney, Wails	MODERATE	ACL: .../main.6/	Token-level probability diverges from theoretical distributions. Highlights gap between model confidence and actual correctness—the exact problem Engine monitors.	https://aclanthology.org/2025.uncertainlp-main.6.pdf
5	The Geometry of Creative Variability: How Credal Sets Expose Calibration Gaps in Language Models	Garces Arias, Rodemann, Heumann	MODERATE	ACL: .../main.5/	Credal sets (imprecise probability) for LLM uncertainty. Geometric approach to distributional analysis connects to Engine's distributional monitoring framework.	https://aclanthology.org/2025.uncertainlp-main.5.pdf