NLP news and developer summaries

The DELEGATE-52 study evaluates LLM reliability in delegated workflows. Testing 19 models, researchers found that even frontier LLMs like Gemini 3.1 Pro, Claude 4.6 Opus, and GPT 5.4 silently corrupt approximately 25% of content in long-form document editing. The study concludes that current LLMs are unreliable for delegation due to cumulative errors exacerbated by task complexity.

Semble is a high-performance, CPU-based code search library designed for AI agents. It offers rapid indexing and low-latency natural language querying by returning precise code snippets, reducing token usage by approximately 98% compared to traditional grep methods. It integrates as an MCP server or CLI tool, enabling instant codebase navigation without external APIs or GPU requirements.

A developer analyzed two decades of digital messages to create a personal CRM, using LLMs to categorize relationship dynamics, emotional temperature, and life events. By quantifying social patterns, the author discovered that personal memory is often selective, revealing how friendship bandwidth, vocabulary, and communication styles evolve over time without necessarily indicating the end of a connection.

Papel is an academic research platform that simplifies paper discovery and analysis. It uses AI to provide grounded summaries and interactive quizzes directly from PDFs, operating locally for privacy. Users can engage with a community, share insights, and track their academic activity through a gamified experience, currently preparing for its App Store launch.

The Portuguese government funded AMÁLIA, an LLM project for European Portuguese, developed by top universities. While achieving impressive results on benchmarks, the project faces criticism for a lack of transparency regarding publicly available weights, data, and training logs. Experts suggest more focus on intrinsic Portuguese cultural knowledge and higher volumes of training data are needed.