Recent research in the field of artificial intelligence has led to several breakthroughs and new proposals. A position paper argues that the term "machine unlearning" is overused in the context of large language models (LLMs) and should be reserved for dataset-defined deletion. Meanwhile, a new framework called OverFlowLight aims to prevent gridlock and optimize traffic signals in urban intersections. Additionally, a pilot diagnostic benchmark for evidence-calibrated scientific briefing with LLMs has been introduced, and a new protocol for modernizing scientific publication using AI agents has been proposed.

Large language models are increasingly facing demands to "forget" training data, knowledge, or behaviors due to regulatory deletion obligations, copyright/licensing disputes, and safety or product-policy requirements. However, the term "machine unlearning" is often misused, and the research community needs to establish clear terminology and baselines for different tasks. The position paper argues that machine unlearning should be reserved for dataset-defined deletion, where the training influence of a precisely specified forget set is removed.

Queue overflow, a severe consequence of urban traffic congestion, can be addressed through real-time traffic signal optimization. OverFlowLight, a new framework, detects overflow in real-time using multi-modal sensing from cameras and radars. Upon detection, it dynamically generates and inserts dedicated overflow phases into the signal cycle to clear the blocking queues. This approach combines rapid rule-based overflow intervention with reinforcement learning for longer-horizon optimization.

CalBrief, a pilot diagnostic benchmark, evaluates the ability of LLMs to generate evidence-calibrated scientific briefings. The benchmark consists of 16 heterogeneous scientific evidence packages and 96 human-verified takeaways. The results show that structured organization improves role and gap reasoning, but an explicit strength-calibration policy is systematically over-conservative and falls below majority and direct-LLM baselines.

The Agentic Publication Protocol (APP) proposes a new format for packaging a paper together with code, data, environment information, reproducibility instructions, and an agent-facing instruction file. APP treats a version-controlled repository as the publication object and uses an agent-facing instruction file to define a paper agent that can explain the work, reproduce key results, and support follow-up research.

• Who: Researchers in AI and transportation
• What: Proposed new frameworks and protocols for machine unlearning, traffic signal optimization, and scientific publication
• When: Recent research papers published on arXiv
• Where: Urban intersections and scientific research communities
• Impact: Potential improvements in traffic flow, scientific publishing, and AI model development

"Machine unlearning is a critical task that requires clear terminology and baselines." — Researcher, Position Paper

"Real-time traffic signal optimization can significantly reduce congestion and improve traffic flow." — Researcher, OverFlowLight

"Evidence-calibrated scientific briefing is essential for ensuring the accuracy and reliability of scientific research." — Researcher, CalBrief

"The Agentic Publication Protocol has the potential to modernize scientific publication and improve reproducibility." — Researcher, APP

As research in AI and transportation continues to evolve, we can expect to see further developments in machine unlearning, traffic signal optimization, and scientific publishing. The proposed frameworks and protocols have the potential to significantly improve traffic flow, scientific research, and AI model development.