All Articles
17700 articles total
Database Context Compression for Text-to-SQL on Real-World Large Databases
Introduction of DBCC, a model-agnostic middleware for database context compression to improve Text-to-SQL performance on large databases.
Fast and Accurate Outlier-Aware LiDAR Super-Resolution for SLAM Applications
A novel Deep Unrolling-based Super-Resolution model for LiDAR point clouds to improve SLAM pose estimation accuracy.
What LLMs explain is not what they believe: Evaluating explanation sufficiency under models' own input beliefs
An evaluation of LLM explanation sufficiency using a new information-theoretic metric called SCSuff to determine if explanations match model beliefs.
Acer’s Swift Go 16 is a lot of laptop for $900
Acer's Swift Go 16 AI laptop is currently discounted to $899, featuring an Intel Core Ultra 7 processor, 32GB RAM, and an OLED touchscreen.
Generative AI Literacy Training Improves Intelligence Analysts' Discrimination of Real and AI-Generated Images
A study demonstrates that a brief, structured training intervention significantly improves the ability of intelligence analysts to distinguish real images from AI-generated ones.
HDDPM: Heteroscedastic Denoising Diffusion Probabilistic Model for Quantitative Low-Count Brain PET Recovery
Researchers introduce HDDPM, a heteroscedastic denoising diffusion probabilistic model designed to improve quantitative low-count brain PET scan recovery by accounting for intensity-aware noise.
A Gravitational Interpretation of Fine-Tuning Reversion
The 'gravitational interpretation of fine-tuning reversion' hypothesis suggests that early training phases create dominant behavioral manifolds that cause models to revert to earlier behaviors during subsequent fine-tuning.
The Speedup Paradox: Rethinking Inference Speed-Quality Trade-off in Embodied Tasks
The TISED framework analyzes the speed-quality trade-off in embodied AI, uncovering paradoxical effects where moderate lossy optimization can actually increase task success rates in dynamic tasks.
CMSL: Constructive Multi-Sequence Learning for Recommendation Systems
Meta introduces CMSL, a constructive multi-sequence learning paradigm for recommendation systems that disentangles user history into thematic strands to reduce context pollution.
MammoFlow: Multiview Mammogram Synthesis with Anatomically Consistent Flow Matching
MammoFlow is a new method for synthesizing anatomically consistent multiview mammograms using flow matching and geometric tissue correspondence.
KernelSight-LM: A Kernel-Level LLM Inference Simulator
KernelSight-LM is a fine-grained LLM inference simulator that provides kernel-level latency breakdowns to help with hardware/software co-design and capacity planning.
Digitizing Coaching Intelligence: An Agentic Framework for Holistic Athlete Profiling using VLM and RAG
A new agentic framework for athlete profiling combines MediaPipe, Llama-4-scout and LangGraph to automate holistic athletic assessments according to SAI protocols.
Geometric Measurements of the Axiom of Choice in Neural Proof Embeddings
Research using Lean 4 shows that the axiom of choice has a measurable geometric correlate in proof space, impacting the performance of neural theorem provers.
Show HN: I made a heatmap of 3400 VCs who are open to cold emails
A user created a heatmap visualizing 3,400 venture capitalists who are open to cold emails.
Matrix URIs, a URL syntax from Tim Berners-Lee that never shipped (1996)
A look back at Matrix URIs, a URL syntax proposed by Tim Berners-Lee in 1996 that was never officially adopted.
Realta Fusion generates electricity directly from a fusion reaction, an apparent first
Realta Fusion claims to have generated electricity directly from a fusion reaction, marking a potential first for the technology.
Google’s NotebookLM can sum up your research in a TikTok-style clip
Google's NotebookLM now allows users to generate TikTok-style short AI video clips to summarize research notes.
Trump asked Musk for SpaceX stock to seed US kids’ savings accounts, report says
Reports indicate Donald Trump asked Elon Musk for SpaceX stock to fund savings accounts for US children.
Anthropic launches Claude Sonnet 5 at a steep discount to its top model as the company races toward a blockbuster IPO
Anthropic released Claude Sonnet 5, a mid-tier model with near-flagship performance and enhanced agentic capabilities at a lower cost.
An Agentic AI Pipeline for Appliance-Level Energy Anomaly Detection and LLM-Driven Recommendations
Researchers propose an agentic AI pipeline combining time-series forecasting and LLMs for appliance-level energy anomaly detection in office buildings.