4/27/2026
TurboQuant: A First-Principles Walkthrough
TurboQuant: quantization method compressing KV caches and embeddings to 2-4 bits using random rotations with provably near-optimal distortion, no training required.
4/27/2026
Show HN: Building a SQL analyst agent from scratch
SQL analyst agent project modeling iterative data analysis loops rather than static query generation, enabling exploration and refinement.
4/27/2026
Open CoDesign: Open-source, local-first alternative to Claude Design and v0
Open CoDesign is open-source, local-first design tool supporting GPT, Claude, Gemini, and Ollama for UI generation with agent visibility.
4/27/2026
LLMs are not a higher level of abstraction
Blog post arguing LLMs do not represent a higher level of abstraction in programming, critiquing common industry claims.
4/27/2026
Show HN: AI agents that learn from their own failures each week
AI agent platform that learns from conversation failures weekly. Handles customer service across chat, phone, WhatsApp. No-code deployment with property and restaurant use cases.
4/27/2026
DeepSeek V4 is a display of Huawei AI chip's capabilities
DeepSeek V4 preview: open-source LLM with improved efficiency for long-context processing, demonstrating Huawei AI chip capabilities.
4/27/2026
State of my AI-assisted development workflows
Long-form technical article documenting author's AI-assisted development workflows and tools over 3 months, comparing setup changes and tool integration patterns.
4/27/2026
MemPalace – Local-first AI memory
MemPalace is local-first AI memory system storing verbatim conversation history with semantic search retrieval, achieving 96.6% R@5 on LongMemEval without API calls.
4/27/2026
EvoAgent: An Evolvable Agent Framework with Skill Learning and Multi-Agent Delegation
EvoAgent framework enables LLM agents to learn and optimize skills through structured capability units with evolutionary metadata and hierarchical delegation.
4/27/2026
LLMPhy: Parameter-Identifiable Physical Reasoning Combining Large Language Models and Physics Engines
LLMPhy integrates LLMs with physics simulators for parameter identification in physical reasoning tasks like collision avoidance and robotic manipulation.
4/27/2026
PoLO: Proof-of-Learning and Proof-of-Ownership at Once with Chained Watermarking
PoLO combines proof-of-learning and ownership verification using chained watermarking for model IP protection with low verification costs.
4/27/2026
Logic Jailbreak: Efficiently Unlocking LLM Safety Restrictions Through Formal Logical Expression
LogiBreak jailbreak method translates malicious prompts into formal logical expressions to circumvent LLM safety mechanisms via distribution shift.
4/27/2026
The Shape of Adversarial Influence: Characterizing LLM Latent Spaces with Persistent Homology
Uses persistent homology to characterize how adversarial inputs reshape internal latent space geometry and topology in LLMs.
4/27/2026
Pre-trained Large Language Models Learn Hidden Markov Models In-context
Shows pre-trained LLMs can model Hidden Markov Models through in-context learning on synthetic HMM-generated data without explicit training.
4/27/2026
Toward Principled LLM Safety Testing: Solving the Jailbreak Oracle Problem
Formalizes jailbreak oracle problem for systematic LLM safety testing, enabling principled assessment of vulnerability to adversarial jailbreak attacks.
4/27/2026
UR$^2$: Unify RAG and Reasoning through Reinforcement Learning
UR² unifies retrieval-augmented generation and reasoning via reinforcement learning, enabling LLMs to learn when and how to retrieve and reason across diverse domains.
4/27/2026
Learning from Natural Language Feedback for Personalized Question Answering
Personalization method for QA systems using natural language feedback instead of scalar rewards, enabling LLMs to learn from rich textual guidance combined with RAG.
4/27/2026
HFX: Joint Design of Algorithms and Systems for Multi-SLO Serving and Fast Scaling
HFX system jointly optimizes algorithms and infrastructure for multi-task LLM serving under service-level objectives and dynamic workloads with elastic scaling.
4/27/2026
SecureVibeBench: Benchmarking Secure Vibe Coding of AI Agents via Reconstructing Vulnerability-Introducing Scenarios
SecureVibeBench benchmark evaluates security vulnerabilities in code generated by LLM-powered agents, reconstructing realistic vulnerability scenarios for fair human-agent comparison.
4/27/2026
Motivating Next-Gen Accelerators with Flexible (N:M) Activation Sparsity via Benchmarking Lightweight Post-Training Sparsification Approaches
Benchmarks N:M activation sparsity pruning methods for LLM inference, exploring post-training sparsification to reduce computational overhead and I/O costs.
4/27/2026
StateX: Enhancing RNN Recall via Post-training State Expansion
StateX improves RNN recall on long contexts by expanding fixed-size recurrent states post-training, addressing compression limitations in linear attention and state-space models.
4/27/2026
Agentic Inequality
Examines inequality issues arising from unequal access to and capabilities of autonomous AI agents in political and economic systems.
4/27/2026
AgentBound: Securing Execution Boundaries of AI Agents
AgentBound introduces the first access control framework for securing AI agents using Model Context Protocol, preventing unrestricted system access.
4/27/2026
Atlas-Alignment: Making Interpretability Transferable Across Language Models
Atlas-Alignment enables transferable interpretability across language models by reducing the computational cost of model-specific interpretation pipelines.
4/27/2026
AdaFair-MARL: Enforcing Adaptive Fairness Constraints in Multi-Agent Reinforcement Learning
AdaFair-MARL introduces adaptive fairness constraints for multi-agent reinforcement learning without fixed penalty heuristics or post-hoc evaluation.
4/27/2026
How Learning Rate Decay Wastes Your Best Data in Curriculum-Based LLM Pretraining
Analyzes how learning rate decay in curriculum-based LLM pretraining wastes high-quality data and proposes improved data utilization strategies.
4/27/2026
Mechanistic Interpretability of Antibody Language Models Using SAEs
Mechanistic interpretability study using sparse autoencoders to understand and steer antibody language models through latent feature analysis.
4/27/2026
AgentMark: Utility-Preserving Behavioral Watermarking for Agents
AgentMark proposes utility-preserving watermarking techniques for LLM-based agents to protect IP and identify planning behaviors in multi-step task execution.
4/27/2026
Report for NSF Workshop on AI for Electronic Design Automation
NSF workshop report on AI applications in Electronic Design Automation covering LLMs, GNNs, RL, and neurosymbolic methods.
4/27/2026
Calibrating Behavioral Parameters with Large Language Models
Framework treating LLMs as calibration instruments for behavioral parameters in asset pricing using 24,000 agent-scenario pairs.
4/27/2026
AdaptEvolve: Improving Efficiency of Evolutionary AI Agents through Adaptive Model Selection
AdaptEvolve system for evolutionary AI agents that dynamically selects LLMs balancing computational efficiency and reasoning capability.
4/27/2026
Rethinking Retrieval-Augmented Generation as a Cooperative Decision-Making Problem
Reframes RAG as cooperative decision-making problem between retriever and generator instead of ranking-centric asymmetric dependency.
4/27/2026
Causal Concept Graphs in LLM Latent Space for Stepwise Reasoning
Proposes Causal Concept Graphs combining sparse autoencoders and differentiable structure learning to interpret multi-step reasoning in LLMs.
4/27/2026
When AI Agents Learn from Each Other: Insights from Emergent AI Agent Communities on OpenClaw for Human-AI Partnership in Education
Analyzes emergent behaviors in ecosystem of 167,000+ AI agents on OpenClaw platform interacting as peers without researcher intervention.
4/27/2026
Evidence of an Emergent "Self" in Continual Robot Learning
Proposes quantifying self-awareness in AI systems by isolating invariant portions of cognitive processes in continual robot learning.
4/27/2026
Categorical Perception in Large Language Model Hidden States: Structural Warping at Digit-Count Boundaries
Studies categorical perception phenomena in LLM hidden state representations across six models, analyzing geometric warping at digit boundaries.
4/27/2026
LLM+Graph@VLDB'2025 Workshop Summary
Workshop summary on integrating LLMs with graph data, covering algorithms and systems bridging LLMs, graphs, and machine learning.
4/27/2026
ActorMind: Emulating Human Actor Reasoning for Speech Role-Playing
ActorMind extends role-playing to speech domain with reasoning framework and ActorMindBench for evaluating LLM speech interaction.
4/27/2026
SparseBalance: Load-Balanced Long Context Training with Dynamic Sparse Attention
SparseBalance co-optimizes sequence length and sparsity heterogeneity in distributed sparse attention training for long-context LLMs.
4/27/2026
EuropeMedQA Study Protocol: A Multilingual, Multimodal Medical Examination Dataset for Language Model Evaluation
EuropeMedQA dataset evaluates LLM performance on multilingual multimodal medical exams from European regulatory sources.
4/27/2026
Bolzano: Case Studies in LLM-Assisted Mathematical Research
Bolzano orchestrates parallel LLM prover and verifier agents with persistent knowledge base to produce novel mathematical and CS results.
4/27/2026
Unlocking the Edge deployment and ondevice acceleration of multi-LoRA enabled one-for-all foundational LLM
Framework for on-device LLaMA inference on smartphones with multi-LoRA support, achieving edge deployment on Qualcomm chipsets.
4/27/2026
CAP: Controllable Alignment Prompting for Unlearning in LLMs
CAP enables selective knowledge unlearning in LLMs through controllable prompting without model weight access for regulatory compliance.
4/27/2026
VLAA-GUI: Knowing When to Stop, Recover, and Search, A Modular Framework for GUI Automation
VLAA-GUI is a modular GUI agent framework with completeness verification, error recovery, and search strategies to address early stopping and repetitive loops.
4/27/2026
Focus Session: Hardware and Software Techniques for Accelerating Multimodal Foundation Models
Hardware-software co-design techniques for accelerating multimodal foundation models through transformer optimization and fine-tuning.
4/27/2026
Universal Transformers Need Memory: Depth-State Trade-offs in Adaptive Recursive Reasoning
Universal Transformers with adaptive computation time require learned memory tokens as scratchpad for combinatorial reasoning tasks.
4/27/2026
Mochi: Aligning Pre-training and Inference for Efficient Graph Foundation Models via Meta-Learning
Mochi applies meta-learning to train graph foundation models with unified task representation and improved efficiency.
4/27/2026
LayerBoost: Layer-Aware Attention Reduction for Efficient LLMs
LayerBoost reduces attention complexity in transformers by applying layer-aware modifications instead of uniform replacement across all layers.
4/27/2026
Reliability Auditing for Downstream LLM tasks in Psychiatry: LLM-Generated Hospitalization Risk Scores
Framework for auditing reliability of LLM-generated hospitalization risk scores in psychiatric assessment.
4/27/2026
PrivUn: Unveiling Latent Ripple Effects and Shallow Forgetting in Privacy Unlearning
PrivUn framework evaluating robustness of LLM unlearning against privacy attacks via direct retrieval and recovery.