<aside> 👀 Details

⏰ Friday, 22th March, 18:00 - Friday, 24th March, 20:00

📍 EPFL

✅ Registration is required

</aside>

Are you ready to shape the destiny of artificial intelligence? Join us at EPFL for an electrifying weekend dedicated to designing innovative challenges that push the limits of large language models (LLMs). Our mission: to detect the pivotal risk of autonomous replication in AI systems. Picture tasks that would take a human expert hours to complete – from fine-tuning data collection to delegating work or even adding a backdoor to its own program.

hack.png

Why participate? Your contributions could set the stage for global AI safety strategies, impact international standards, and earn you bounties exceeding $4,000! Collaborate with METR for potential co-authorship on groundbreaking research. Seize the opportunity to leave an indelible mark on the future of AI.

Event Highlights: 🔍 Challenging Tasks: Craft tasks that stretch the boundaries, demanding 6-20+ hours of dedicated effort, tested by human professionals.

📊 Measurable Scoring: Ensure clear, automatic scoring from 0 to 1, allowing nuanced assessment of AI capabilities.

🌐 Realistic Scenarios: Ground your tasks in real-world situations, free from bugs or loopholes.

How to Participate:

  1. 📜 Agree to IP & Data Security: Commit to safeguarding tasks from inclusion in LLM training data. Sign our agreement.
  2. 🧠 Develop Ideas: Prepare your ideas before kickoff. Check existing ones or create your own under the "Ideas" tab with referral code "Code Red."
  3. 🚀 Keynote and Kickoff: Tune in on March 22nd for the livestreamed introduction by Beth Barnes, METR's CEO & Lead Researcher.
  4. 🛠️ Create Your Task: Find resources under the "Resources" tab for code, task specification, and reporting.
  5. 🧪 Test on Humans: Ensure realism by testing tasks on fellow participants with diverse professional backgrounds.
  6. 🏆 Finalize & Submit: Polish your task with feedback, then submit under the "Submission" tab for a chance to win!