cs.CL papers | Gist.Science

The bond between carbon and chlorine, known as the C–Cl bond, is a fundamental building block in organic chemistry and materials science. This specific interaction influences everything from the stability of pharmaceuticals to the environmental behavior of industrial solvents. Understanding how these atoms connect and react helps scientists design safer chemicals and develop new methods for breaking down persistent pollutants in our ecosystem.

At Gist.Science, we process every new preprint in this category directly from arXiv to make these complex findings accessible to everyone. Whether you are a seasoned researcher or a curious learner, you can access both plain-language overviews and detailed technical summaries for each study. This dual approach ensures that the nuances of C–Cl chemistry are clear without losing the scientific rigor found in the original manuscripts.

Below are the latest papers on carbon-chlorine interactions, updated daily as they appear on arXiv.

💬 NLP

Simplicity Paradox: Debunking myths about prompting and datasets for LLM evaluation

This paper challenges the assumption that sophisticated prompting techniques enhance Large Language Model performance by demonstrating through a comprehensive empirical study that baseline prompting consistently outperforms complex methods, with only minimal expert or inductive framing yielding slight improvements, thereby suggesting the field should prioritize genuine model advancement over prompt engineering.

Inder Preet, Shuxin Lin, Dhaval Patel2026-07-17

💬 NLP

MAPS: Modeling Co-Existing Subjective Perspectives and Shared Meaning in Multi-Agent Cognitive Dialogue

The paper introduces MAPS, a novel multi-agent framework that balances individual subjective perspectives with shared meaning in dialogue by utilizing domain-weighted profiles, dynamic memory, and interpretable attention, thereby enabling cognitively distinct agents to achieve semantic alignment without sacrificing diversity.

Molood Arman, Clément Bonnafous2026-07-17✓ Author reviewed ⓘ

💬 NLP

Introspection Fine-Tuning (IFT): Training Small LLMs to Introspect

This paper introduces Introspection Fine-Tuning (IFT), a method that successfully trains small language models (including 1B parameter models) to reliably detect and report on internal activation perturbations, thereby unlocking latent self-monitoring capabilities that are not solely dependent on model scale.

Ely Hahami, Ishaan Sinha, Lavik Jain2026-07-17

💬 NLP

Information-Theoretic Limits of Reliability and Scaling in Language Models

This paper challenges the assumption that perfect reliability is achievable through scaling alone by establishing an information-theoretic framework that defines inherent reliability ceilings based on task ambiguity and inter-token dependencies, thereby deriving a unified scaling law that identifies the bottleneck between training data and model capacity while explaining phenomena like retrieval augmentation and catastrophic forgetting.

Subhabrata Majumdar2026-07-17

💬 NLP

Semantic Register Compression in Multi-Agent LLM Cascades

This paper identifies and quantifies "semantic register compression," a phenomenon where intermediate agents in multi-agent LLM cascades systematically reduce label separability in embedding space during semantic transformation, leading to measurable information loss across diverse high-stakes domains like fact-checking, sentiment analysis, and medical triage.

Manuele Tele Junior Fernandez2026-07-17

💬 NLP

Cross-Dataset Generalization in Urdu Fake News Detection: An Empirical Study with XLM-RoBERTa and a Length Confound Analysis

This paper presents the first cross-dataset generalization study for Urdu fake news detection using XLM-RoBERTa, revealing a severe performance collapse in one transfer direction caused by a systematic length confound in the Ax-to-Grind dataset that induces shortcut learning.

Muhammad Abdullah Haroon2026-07-17

💬 NLP

Breaking Refusal in the First Half: A Mechanistic Study of the Prefill Jailbreak

This study reveals that prefill-based jailbreaks bypass safety by exploiting a shallow, response-site computation where the model's harmful continuation is driven primarily by passive autoregressive conditioning on the initial "Sure, here is" prompt rather than by the suppression of an intact harm representation.

Alex Kwon2026-07-17

💬 NLP

MemoHarness: Agent Harnesses That Learn from Experience

MemoHarness is an adaptive agent optimization framework that enhances LLM performance across diverse tasks by decomposing the control layer into six dimensions and dynamically tailoring harness configurations for each case using a dual-layer experience bank derived from past executions, all without requiring test-time labels or additional search.

Yue Huang, Wenjie Wang, Han Bao, Yuchen Ma, Xiaonan Luo, Yi Nian, Haomin Zhuang, Zheyuan Liu, Yue Zhao, Xiangliang Zhang2026-07-17

🤖 machine learning

Privacy Leakage in Federated Learning in Radiology Reports: A Comparative Evaluation of Tokenizer-Driven Privacy Risks

This study demonstrates that federated learning on radiology reports is vulnerable to significant gradient-based privacy leakage, where sensitive clinical text can be reconstructed with high fidelity regardless of the tokenizer used, indicating that domain-specific tokenizers alone are insufficient to protect patient privacy without additional safeguards like secure aggregation or differential privacy.

Santhosh Parampottupadam, Andres Martinez, Dimitrios Bounias, Sinem Sav, Klaus Maier-Hein, Ralf Floca2026-07-17

💬 NLP

Implicit Reasoning Steering via Concept Chaining

This paper introduces "Concept Chaining," a method that exploits the reasoning fragility of large language models by continuing pretraining on short, natural-language paragraphs linking question entities to a target answer via intermediate concepts, thereby covertly steering model predictions without explicit instructions or direct cues.

Xiao Ye, Sanika Chavan, Yuxi Huang, Shahriar Kabir Nahin, Muhao Chen, Anshuman Chhabra, Ben Zhou2026-07-17

← Previous Next →