Crypto & Blockchain · Wed, Oct 28, 2026

Bitcoin, Ethereum, stablecoins, DeFi, smart contracts, scaling and consensus.

RSS feed
arXiv cs.CV · cs.AI

MEV: A Multi-Event Video Dataset for Long-Take Generation

In briefMEV: A multi-event video dataset for generating long, continuous video takes

SummaryThis paper introduces MEV, a dataset of real-world videos in which subjects perform a sequence of distinct, non-overlapping events, with precise temporal boundaries and fine-grained procedural annotations. The authors also present an annotation pipeline that segments videos into temporally grounded events and produces structured captions at both global and event levels. Under an event-centric evaluation protocol, fine-tuning on MEV with temporally grounded event conditioning is reported to improve event transitions and content coherence in long-take generation.

Peiyuan Zhu, Shaoan Xie, Yifan Shen et al.

★★★★☆ Notable
arXiv cs.AI · cs.HC · cs.IR Open resources

Measuring Where Intent Gets Lost in Agentic Smart Contract Synthesis

In briefMeasuring Where Intent Is Lost When AI Agents Write Smart Contracts

SummaryThis paper examines whether AI agents that write smart contracts from plain-English descriptions actually produce code that behaves as requested. The authors compare an AI-written contract with the original on the same transactions and find that common AI judges and standard analysis tools miss obvious mistakes. They propose asking the client clarifying questions before code generation and checking the result against an independently written executable answer key, which they report works far better than reviewing or patching afterwards.

WhyThe authors propose asking clarifying questions before generation and checking against an independent executable answer key, reporting results on 600 tasks across three models and releasing tools.

Abhinav Goel, Agostino Capponi, Alfio Gliozzo et al.