Topic
code
Technology Flock Safety's AI Tool for Police Can Identify Drivers by Movement Alone, WIRED Found
WIRED obtained code from Flock Safety's website revealing OS Investigate, an AI tool that identifies drivers and tracks vehicles by movement patterns across more than 6,000 communities. The leaked files describe 69 prewritten prompts and 45 tools accessing plate scans, arrest records, dispatch logs, and commercial databases, raising constitutional concerns.
Technology Study Finds Mobile Apps Marketed to US Troops Contain Chinese and Russian Code
A study by Purdue University, West Point, and Florida International University examined 220+ mobile apps marketed to US troops, finding that more than one in eight contained software from foreign adversaries like China and Russia. 64% had third-party SDKs, 40% collected more data than disclosed, and 12 apps included Huawei's HMS Core, posing risks for troop location tracking.
Research Shows Code Execution Outperforms Natural Language for AI Algorithmic Reasoning
A new research paper from arXiv investigates whether code or natural language is more effective for tool-augmented language models performing algorithmic reasoning. By separating intermediate representation from execution mechanism, the study finds that deterministic code execution outperforms natural-language reasoning by 31.6 percentage points, while changing the intermediate representation alone yields only a 0.15pp difference. Results suggest performance gains require reliable external execution.
Semi-Supervised Framework Scales LLM Reasoning Using 10-15x Fewer Labels Than Traditional Methods
A new semi-supervised framework for training LLM reasoning uses a lightweight verifier to judge reasoning quality, requiring only a few labeled samples. Experiments on math problems and visual question answering show accuracy comparable to 10-15x more labeled data. The method could reduce the cost of building large-scale reasoning datasets.
New Agentic Programming Framework Shifts Control from LLMs to Deterministic Code for Greater Reliability
A new paper argues that current LLM agent frameworks have architectural flaws leading to token explosion, control-flow hallucination, and unreliable completion. The authors propose Agentic Programming, where the program governs all control flow and the LLM is an adaptive component called LLM-as-Code, invoked only for reasoning or generation. A case study on computer-use agents shows improved stability in long visual operation sequences.