The End of Data Leakage

About this episode

The episode discusses challenges in AI model leaderboards, focusing on data leakage and benchmarking innovations.

In this episode, we dive deep into the challenges facing time series AI model leaderboards, from hidden information leakage to the complexities of benchmarking foundation models. I sit down with Marcel Meyer to unpack why traditional approaches fall short and how our new TS Arena leaderboard is setting a new standard for fair, future-proof evaluation. We explore the pitfalls that plague current benchmarks, the surprising ways data contamination can skew results, and the innovative pre-registration protocol we've developed to keep evaluations honest. If you've ever wondered what it takes to build a truly trustworthy AI leaderboard—or why it matters for industry and research alike—this conversation is packed with insights you won't want to miss.

People in this episode

Hosts: Robert Weber, Peter Seeberg

Guest: Marcel Meyer

Topics covered

• data leakage
• AI model leaderboards
• benchmarking
• time series AI
• evaluation protocols

Keywords

• data leakage
• AI
• benchmarking
• time series
• evaluation
• trustworthy AI
• data contamination

Mentioned in this episode

Organizations: TS Arena

More episodes of Industrial AI Podcast

• AI for the Glass Industry · June 3, 2026 · 29 min
• NVIDIA´s Safety Initiative / The unique Volkswagen-Siemens joint venture · May 27, 2026 · 1h 11m
• Governed Autonomy: Why Edge AI Needs Guardrails · May 20, 2026 · 53 min
• Mistral acquires NXAI spin-off Emmi AI / API Calls · May 19, 2026 · 43 min
• Exclusive News Delay · May 13, 2026 · 1 min
• The CAD AI Hype!? · May 6, 2026 · 28 min