Constructor Labs
Bremen, DE · On-site · Internship
Constructor Student Labs
NewPosted 2 hours ago
- Location
- Bremen, DE
- Work mode
- On-site
- Employment
- Internship
- Experience
- Mid-level
About the role
Student Labs: Constructor Fabric Research Edition
Curious how AI is changing the way software is built? Want to explore it hands-on, guided by the researchers who work on it every day?
Constructor Tech and Constructor Labs, together with Constructor University, invite you to Student Labs: a 2-month educational research program for BSc and MSc students.
This edition is dedicated to Constructor Fabric, a shared AI and software platform. Fabric brings together reusable building blocks, common infrastructure and AI agents that help design, build and test software.
🚀 Why Constructor Fabric?
AI-native software platforms are one of the fastest-growing fields in tech, and few students get to study one from the inside. Student Labs gives you that chance. You explore open research questions on a platform in active development, learn current research methods, and build skills that are in high demand in industry and academia.
🔬 What you can research
AI that builds software
• Domain-specific languages (DSL): how a formal language can describe software components so that AI can generate and regenerate them from specifications
• Neurosymbolic models: combining neural networks with logic and rules, so that AI-generated code can be checked, not just produced
• Rule extraction and fine-tuning: how to extract rules and knowledge from existing code and use them to adapt language models
• World models: how AI systems build an internal model of their environment to understand, plan and predict
• Formal verification: mathematical methods that prove software behaves as specified
• Geometric models: using geometry to represent data and help AI models learn from it
• Benchmarking: how to measure fairly whether one AI model or method is really better than another
Next-generation language models
• Diffusion models for text: today’s LLMs write one token at a time, which makes them slow and prone to hallucinations. Diffusion models generate whole passages at once. Explore how to make them match and beat today’s models.
• LLM security: how to host large language models securely when many users and organisations share the sa
Constructor Labs
Posted via Arbeitnow
Apply Now takes you to Rozgoo, where auto-apply can submit your application for this role. Updated today.
Apply Now