Collinear AI COLM Social - The Future of Post-Training: Research Social
Post-training has emerged as the new frontier of model development, unlocking real performance gains, alignment breakthroughs, and data innovations.
This social brings together researchers, practitioners, builders, and engineers who are pushing the boundaries of fine-tuning, reward modeling, continual pretraining, and data curation to exchange ideas on what comes next.
We will explore how evaluation feedback, curated datasets, and adaptive improvement loops are reshaping how frontier models evolve and what new research directions this opens for the community.
Expect a fast, informal session with short provocations, open discussion, and networking among peers driving the post-training ecosystem across frontier labs and industry.
Who we are (Collinear AI)
Early-stage, seed-backed. Our mission is simple: make powerful AI amplify human potential. We blend alignment research with practical engineering so teams can ship models they trust.
Simulate: multi-turn, realistic user journeys to stress-test agents.
Analyze: 20+ performance/safety/reliability metrics with policy-aware judges.
Improve: targeted semi-synthetic data and reward signals for post-training.
Our COLM 2025 Paper: Cats Confuse Reasoning LLM: Query Agnostic Adversarial Triggers for Reasoning Models
Collinear at COLM 2025 Website
Location: Room 524B