Applied AI Journal Club Bern - April

We’re excited to invite you to the april edition of the Applied AI Journal Club Bern 🎉

🕓 Agenda
17:25 Arrival and get-together
17:30 Welcome and short introduction
17:35 Paper discussion
18:25 Wrap-up and open exchange
18:30 End of event

📘 Preparation
Please read the selected paper in advance so that you’re familiar with the main ideas and key concepts. We won’t present the paper in detail — instead, we’ll focus our time on discussing its content, insights, and real-world implications.

Paper:
AutoHarness: Improving LLM Agents by Automatically Synthesizing a Code Harness
Published by researchers from Google DeepMind.

The study introduces a framework where LLMs automatically synthesize Python code "harnesses" to interact with environments and filter illegal actions, enabling a smaller Gemini-2.5-Flash model to outperform flagship models like GPT-5.2-High at nearly zero test-time cost. We will also discuss the general topic of LLM harnesses—the use of external, verifiable code wrappers to transform LLMs into robust and cost-effective agents.

📍 Location & Registration

• We are welcomed at the inpeek AG office (Bogenschützenstrasse 9) in the PostParc building next to the train station.
• Due to the limitations of meetup.com (we don’t pay for additional capacity), the event is limited to 10 participants. If you don’t secure an ordinary spot, just join the waitlist—you’ll still be very welcome! This helps us plan and know how many people to expect.
• The event will be held in english

Questions? Please reach out to Fabian Hüni.