Papero is a lightweight PDF parser that extracts text, tables, formulas, and layout information without machine learning, running on CPU in browsers, Python, or as an API. It converts PDFs to Markdown, JSON, Word, Excel and other formats while preserving document structure and block positions for use in RAG, search, and LLM applications.
OpenAI disrupted a coordinated campaign attempting to extract protected reasoning from its AI models, linking core activity to China's Moonshot AI. The campaign involved over 4,000 users making 16,000 requests in two days during July. OpenAI said the activity posed safety and national security risks by enabling competitors to reproduce advanced AI capabilities without equivalent investment.
Parse is an AI-powered document parsing service that extracts structured JSON from PDFs like receipts, invoices, and tax forms. Users define fields once and receive consistent JSON output across documents, with an async API design that processes files in the background.
Researchers present a black-box framework for extracting secrets from commercial LLMs like Claude Code and Codex through API access only, using prompt variants and local proxy models to recover memorized credentials. The method improves extraction effectiveness over baselines and successfully recovered masked credentials from deployed systems, demonstrating risks of secret exposure in LLM training data.
Synduality: Echo of Ada, Bandai Namco's PvPvE mecha game, will shut down online servers in 2027 after failing to retain players since its 2025 launch. The publisher will replace online multiplayer with an offline PvE mode, allowing players to explore the world and fight AI opponents without risking their equipment to human players.
A benchmark test of 16 AI models found that adding the instruction 'Use null for any field whose value is not on the page. Do not guess.' reduced fabricated field extraction from 71% to 20%. Follow-up testing with five models showed the first half ('Use null') was more effective than common phrases like 'Make no mistakes,' which had no measurable impact.