screenshot taken from: https://priorlabs.ai/tabpfn
When
Tuesday, 31st March 2026, 6:00pm to 9:00pm
Where
Merantix AI Campus, Max-Urich-Straße 3, Berlin, Germany
Hosting Organization
Merantix AI Campus & Prior Labs
Participation Fee
Free Entrance
Agenda
Host Intro, Panel Discussion, QnA, Demo 1, Short Break, Demo 2, Socializing
Topics Covered
The Acitivities of the Merantix Group & The 4-month Residency for AI Engineers called Hacker Room (Host Intro), Foundational Models for Tabular Data & AI Agents Struggling with Structured Data (Panel Discussion), Jevons Paradox, Partial Dependence Plots & Change Management (QnA), TabPFN & Agent Bricks (Demo 1), A Hands-On Masterclass of Building Agentic Systems with Langdock (Demo 2)
I’ve learned something today
- TabPFN is a transformer-based model pre-trained on many synthetic tabular datasets that delivers strong predictions on small, single-table data with zero training or hyperparameter tuning. It rivals or beats XGBoost on datasets under ~10k rows, but XGBoost remains superior for larger datasets due to its scalability and production efficiency.
- Synthetic data is artificially generated data that mimics real-world patterns without using real individuals or events. It enables scalable, customizable, and privacy-safe datasets for training AI systems. Generative AI models like GANs, diffusion models, and LLMs make this possible by automatically creating high-quality, diverse data at scale. This helps overcome data scarcity, reduce bias, and significantly accelerate AI development.
- Jevons Paradox states that improving efficiency in using a resource often leads to higher overall consumption of that resource, not less. In AI, this means that as systems become cheaper and more efficient, demand for human intelligence can actually grow alongside them. Lower costs unlock more use cases and outputs, increasing the need for human judgment, oversight, and strategic thinking.
- According to the article “TabPFN AI Accelerates Business Transformation on Databricks” on the Databricks blog, TabPFN delivers up to 90% faster data science workflows and 10–65% improvements in accuracy compared to traditional ML methods. These quantified gains are the key reason Databricks highlights TabPFN as a breakthrough for scaling predictive AI efficiently.
- We have moved from the flawed idea of measuring the productivity of software engineers by the number of lines of code they produce to measuring it by the token output they generate in the GenAI era.
- The command line is resurging because AI removes the need for expert memorization while also reducing reliance on intermediate abstraction layers like higher-level programming languages. Instead of writing full programs or navigating complex frameworks, users can express intent in plain language and have AI translate it directly into executable commands.
- Contentful for some reason features in its Berlin office, located in the same building as the Merantix AI Campus, a perfect example of a fun, minimalistic utility car:
picture taken at venue
Published:
Modified: