AINewsnow

Breaking the Subword Boundary: Probing Architectural Vulnerabilities in Modern LLMs (ASTRAL-Bench on Kaggle)

This is a submission for the Kaggle Benchmarking Challenge When we look at standard LLM leaderboards, everything seems solved. Models boast 90%+ on MMLU and 85%+ on GSM8K. We are told these models possess advanced reasoning, can act as autonomous enterprise agents, and are ready to manage multi-mil…

Read the full story at DEV Community — Machine Learning ↗

Timeline · 1 report

  1. 2026-10-08 16:27 · DEV Community — Machine Learning
    Breaking the Subword Boundary: Probing Architectural Vulnerabilities in Modern LLMs (ASTRAL-Bench on Kaggle)

More stories

  1. GPT-6 and Intelligent UI for everyone — OpenAI News
  2. Introducing Mistral Large 4 — Mistral AI News
  3. Introducing Claude Haiku 5.5 on AWS — AWS Machine Learning Blog
  4. Mistral Says Its New AI Model ‘Le Chonk’ Is the Best Open-Weight Offering Outside of China — Wired AI
  5. Sharing AI progress in mathematics — OpenAI News
  6. Introducing Playground: Create and play custom games — Google AI Blog
  7. OpenAI Decisions API now available on AI Gateway — Vercel Blog
  8. Everything announced at Microsoft's Surface Laptop Ultra event — The Verge AI

Get the daily brief of stories like this at 6:30 every morning →