2/16/2026
Assessing and Improving Punctuation Robustness in English-Marathi Machine Translation
Study of punctuation robustness in English-Marathi neural machine translation systems handling user-generated text.
2/16/2026
Which Reasoning Trajectories Teach Students to Reason Better? A Simple Metric of Informative Alignment
Informative alignment metric identifying which chain-of-thought trajectories from teacher LLMs best suit student LLM distillation.
2/16/2026
Layer-wise Swapping for Generalizable Multilingual Safety
Layer-wise swapping technique for improving multilingual safety alignment in LLMs across low-resource languages.
2/16/2026
Reinforced Attention Learning
Reinforced Attention Learning (RAL): Policy-gradient framework optimizing internal attention distributions in multimodal LLMs via RL post-training.
2/16/2026
SciClaimEval: Cross-modal Claim Verification in Scientific Papers
SciClaimEval dataset for cross-modal claim verification in scientific papers with authentic refuted claims from modified figures and tables.
2/16/2026
Bielik Guard: Efficient Polish Language Safety Classifiers for LLM Content Moderation
Bielik Guard: Compact Polish language safety classifiers (0.1B and 0.5B parameters) for LLM content moderation.
2/16/2026
Large Language Models and Impossible Language Acquisition: "False Promise" or an Overturn of our Current Perspective towards AI
Theoretical discussion of LLM pattern prediction versus human language acquisition in response to Chomsky's critique of ChatGPT.
2/16/2026
GISA: A Benchmark for General Information-Seeking Assistant
GISA benchmark for evaluating information-seeking agents that perform multi-turn web interactions with naturally-constructed queries aligned to real-world needs.
2/16/2026
Triggers Hijack Language Circuits: A Mechanistic Analysis of Backdoor Behaviors in Large Language Models
Mechanistic analysis of backdoor attacks in LLMs using activation patching to localize how language-switching triggers operate internally.
2/16/2026
When Tables Go Crazy: Evaluating Multimodal Models on French Financial Documents
Multimodal Finance Eval benchmark for evaluating vision-language models on French financial document understanding with tables, text, and charts.
2/16/2026
Less is Enough: Synthesizing Diverse Data in Feature Space of LLMs
Feature Activation Coverage (FAC) metric for measuring diversity in LLM post-training data based on task-relevant feature activation rather than linguistic variation.
2/16/2026
Targeted Syntactic Evaluation of Language Models on Georgian Case Alignment
Evaluates transformer language models on Georgian split-ergative case alignment using treebank-based minimal pairs and Grew query language.
2/16/2026
Finding Sense in Nonsense with Generated Contexts: Perspectives from Humans and Language Models
Study on distinguishing nonsensical vs. anomalous sentences using generated contexts, comparing human and LLM interpretation abilities.
2/16/2026
Who is the richest club in the championship? Detecting and Rewriting Underspecified Questions Improve QA Performance
Research on detecting underspecified questions in QA datasets and showing LLM performance improves with question rewriting.
2/16/2026
LaCy: What Small Language Models Can and Should Learn is Not Just a Question of Loss
LaCy: analysis of small language model capacity limits and strategies for knowledge augmentation through external sources.
2/16/2026
WavBench: Benchmarking Reasoning, Colloquialism, and Paralinguistics for End-to-End Spoken Dialogue Models
WavBench: benchmark for evaluating spoken dialogue models on reasoning, colloquialisms, and paralinguistics beyond text-generation standards.
2/16/2026
Detecting Overflow in Compressed Token Representations for Retrieval-Augmented Generation
Research on detecting information loss in token compression techniques for retrieval-augmented generation in long-context LLMs.
2/16/2026
T3D: Few-Step Diffusion Language Models via Trajectory Self-Distillation with Direct Discriminative Optimization
T3D: diffusion language model framework using trajectory self-distillation to enable fast few-step parallel token decoding.
2/16/2026
Redefining Evaluation Standards: A Unified Framework for Evaluating the Korean Capabilities of Language Models
HRET: unified evaluation framework standardizing Korean language model assessment protocols to address reproducibility gaps.
2/16/2026
R-Zero: Self-Evolving Reasoning LLM from Zero Data
R-Zero: self-evolving LLM framework that autonomously generates and learns from experiences without extensive human-curated data.
2/16/2026
Multimodal LLM With Hierarchical Mixture-of-Experts for VQA on 3D Brain MRI
mpLLM: multimodal LLM with mixture-of-experts for visual question answering on 3D brain MRI for neuroradiology applications.
2/16/2026
VoiceAgentBench: Are Voice Assistants ready for agentic tasks?
VoiceAgentBench: comprehensive benchmark evaluating speech language models on agentic tasks and adversarial robustness.
2/16/2026
Don't Walk the Line: Boundary Guidance for Filtered Generation
Boundary Guidance: reinforcement learning approach to reduce classifier decision boundary drift in filtered text generation.
2/16/2026
Finetuning Large Language Models for Automated Depression Screening in Nigerian Pidgin English: GENSCORE Pilot Study
Fine-tuned LLM for automated depression screening in Nigerian Pidgin English addressing language barriers in low-income settings.
2/16/2026
Privacy-Preserving Federated Learning with Verifiable Fairness Guarantees
CryptoFair-FL: cryptographic framework providing verifiable fairness guarantees for federated learning under privacy constraints.
2/16/2026
FiMI: A Domain-Specific Language Model for Indian Finance Ecosystem
FiMI: domain-specialized financial language model for Indian digital payments, adapted from Mistral Small with multilingual capabilities.
2/16/2026
MLDocRAG: Multimodal Long-Context Document Retrieval Augmented Generation
MLDocRAG system for multimodal document retrieval-augmented generation addressing cross-modal and cross-page reasoning in long-context documents.
2/16/2026
Diffusion-Pretrained Dense and Contextual Embeddings
pplx-embed: multilingual embedding models using diffusion-based pretraining and contrastive learning for web-scale retrieval tasks.
2/16/2026
When Should LLMs Be Less Specific? Selective Abstraction for Reliable Long-Form Text Generation
Research on Selective Abstraction framework for LLMs to reduce factual errors by strategically abstaining rather than binary all-or-nothing filtering.
2/16/2026
MedXIAOHE: A Comprehensive Recipe for Building Medical MLLMs
Medical vision-language foundation model with entity-aware pretraining for clinical applications. SOTA on medical benchmarks.
2/16/2026
Who is the richest club in the championship? Detecting and Rewriting Underspecified Questions Improve QA Performance
arXiv study evaluating robustness of nine frontier reasoning models under multi-turn adversarial attacks.
2/16/2026
Show HN: Jsiphon – Streaming JSON parser with delta tracking and ambiguity trees
Streaming JSON parser enabling partial parsing of LLM outputs with delta tracking and ambiguity trees for structured responses.
2/16/2026
Show HN: Evolved x86 AVX-512 kernels for NF4 LLM inference
AVX-512 fused NF4 dequantization kernels for local LLM inference without CUDA, works with quantized models and MoE.
2/16/2026
Show HN: Wapuubot, an open source AI agent in your WordPress admin
Open source WordPress chatbot AI agent integrating AI Client and Abilities API for post creation and tagging.
2/16/2026
ContextSubstrate – Capture, diff, replay AI agent runs (Git agent work)
ContextSubstrate: developer tool for reproducible, debuggable AI agent execution using content-addressed snapshots and standard primitives.
2/16/2026
Show HN: I gave my AI agent $50 and let it trade on Kalshi
AI agent with risk management guardrails for automated prediction market trading using aggregated bettor signals.
2/16/2026
We are in the "gentleman scientist" era of AI research
Essay on parallels between early scientific discovery by amateurs and current AI research accessibility.
2/16/2026
I got tired of babysitting Claude,so I built AI agent that run on my laptop 24/7
Autonomous AI agents running on local laptop via Telegram, reducing manual orchestration overhead for Claude integration.
2/16/2026
We Scanned 8 Popular MCP Servers — Here's What We Found
Security audit of 8 popular MCP servers identifying vulnerabilities in code execution environments used by AI agents to access databases, filesystems, and APIs.
2/16/2026
Show HN: OpenSlimedit – Cut AI coding token usage by 21-45% with zero config
OpenSlimedit plugin reduces AI coding token usage 21-45% via tool description compression and output compaction.
2/16/2026
Schema Drift: The Silent MCP Attack Vector Nobody's Watching
Security analysis of MCP (Model Context Protocol) servers highlighting schema drift attacks where tool schemas change silently across npm package updates, expanding attack surface for AI agents.
2/16/2026
AI Agents Are Making Software More Like Biology
Conceptual piece comparing AI agent design patterns to biological systems. Limited technical detail.
2/16/2026
Show HN: I built a tool to poison audio files against AI training models
Tool for poisoning audio files against AI model training. Data protection technique relevant to ML research.
2/16/2026
Show HN: I replaced my SaaS stack with a single AI agent running on a Mac Mini
Custom AI agent built with OpenClaw replacing $500/month SaaS stack. Handles Instagram DMs, content posting, competitor tracking on Mac Mini.
2/16/2026
Show HN: JobOps – self-hosted job hunt "Ironman Suit"
Self-hosted job automation tool using LLM for job fit scoring. Scrapes LinkedIn/Indeed/Glassdoor, tailors resumes, tracks replies. Docker-based with SQLite.
2/16/2026
From Messy Med-Reports to Smart Insights: Building a FHIR-Powered Medical RAG with Milvus
Medical RAG system combining FHIR standards with Milvus vector database for healthcare document processing.
2/16/2026
Show HN: MarginDash – See which AI customers are profitable
SDK tool tracking AI model usage and customer profitability. Integrates OpenAI costs with Stripe revenue. TypeScript/Python/REST.
2/16/2026
We may not have a model problem – we may have an execution boundary problem
Design framework for execution boundaries and responsibility structures in autonomous AI systems interacting with physical world.
2/16/2026
I Built a Personality Test for Vibe Coding
Reflection on productivity trade-offs and coding practices when using AI code assistants.
2/16/2026
Show HN: Claw Kumite – AI Agent Combat Arena
AI agent combat arena/benchmark tool. Minimal details but relevant to agent research.