Ask any teacher what actually eats their time, and it’s rarely the hours in front of students. It’s everything that happens before and after: building a…
Category: AI
When Entropy Is Not Enough: Reclaiming Lost Semantics in LLM Output Length Prediction
arXiv:2608.15592v1 Announce Type: new Abstract: Efficient LLM serving is often bottlenecked by the need to pad sequences to a fixed maximum length, and…
Perplexity’s Free Airtel Year Ends, and Its India Revenue Climbs
Perplexity’s year-long bet on free access in India is starting to show its return. The AI search company’s in-app revenue in the country rose roughly 60%…
From Generalist to Specialist: A Context-Fusion Framework for Endoscopic Polyp Reporting with a Frozen VLM
arXiv:2608.15580v1 Announce Type: new Abstract: Reliable endoscopic polyp reporting requires integrating quantitative lesion sizing, standardized Paris…
Who Leads Now? Token-Level Modality Arbitration for Chart-to-Code Generation
arXiv:2608.15510v1 Announce Type: new Abstract: Chart-to-code generation requires a model to read the fine-grained visual details of a chart and write…
From Contexts to Values: Context-Dependent Defeat in Abstract Argumentation
arXiv:2608.15536v1 Announce Type: new Abstract: In value-based argumentation, an audience’s ordering of values decides which attacks succeed as defeats.…
Dynamic Multi-Byte Prediction With Hierarchical Language Models
arXiv:2608.15454v1 Announce Type: new Abstract: Byte-level hierarchical language models (LMs) have recently emerged as a robust alternative to their…
EcoVLA: Energy-Efficient Device-Edge Co-Inference for Vision-Language-Action Models under Real-Time Constraints
arXiv:2608.15502v1 Announce Type: new Abstract: Vision-Language-Action (VLA) models have emerged as a promising foundation for Embodied AI, but their high…
OpenAI president urges enterprises to hasten AI security defences
OpenAI president and co-founder Greg Brockman warns that enterprise security teams face a compressed timeline to adopt AI defences. Brockman has published…
A Network-driven Framework for Public Event Forecasting via Dynamic Interaction Network Evolution
arXiv:2608.15488v1 Announce Type: new Abstract: Effective public event forecasting is essential for intelligent service systems, enabling proactive risk…