Freelance Catalyst Game Labs Jobs (NOW HIRING)

Good Start Labs builds games that make AI models better, for us. We're defining the category of alignment through entertainment.

We need a Research Scientist to help design and execute research that improves language models through game-based training. You'll work on improving a 260B+ parameter model we have under contract by EOY, using RL environments and game-generated data to demonstrate measurable capability improvements.

Our agents are integrated into games with millions of players, we have live contracts with model training labs, and our research was just accepted to the NeurIPS workshop on multi-turn interactions in LLMs (submitted to AAAI). We're backed by General Catalyst and Inovia, with advisors including Kelly Clancy (ex-DeepMind, author of Playing With Reality) and researchers from major AI labs like Anthropic.

• Help design research experiments exploring how games improve model capabilities—initially focused on agents and computer use
• Build and optimize RL environments, curate pre-training data, implement training algorithms, run large-scale experiments
• Work on improving a 260B+ parameter model using game-generated data and RL techniques (pre-training and post-training)
• Collaborate with university research partners at Princeton, Rice, UCSD, HKUST, UMD, and NYU on experiments and publications
• Demonstrate measurable model improvements to our lab partners

Required:

• PhD in CS, ML, or related field (or equivalent research experience at top labs)
• Published research in RL, multi-agent systems, or LLM training
• Strong Python and PyTorch/JAX skills—comfortable implementing novel research ideas
• Genuine passion for games and what they teach about intelligence
• Comfortable at early-stage startup (we're 2 years funded, moving fast)

Preferred:

• Experience at DeepMind, FAIR, OpenAI, Anthropic, or top academic groups
• Background in game-playing AI (AlphaGo lineage, Cicero, game engines)
• Experience with pre-training or post-training at scale (260B+ parameter models)
• Familiarity with RLHF, reward modeling, or alignment research

• Location: Toronto or New York preferred; flexible for exceptional remote candidates
• Partnerships: Direct collaboration with research groups at six universities
• Traction: Live contracts, accepted papers, millions of players in games using our agents
• Runway: $3.6M raised, two years funded

Good Start Labs is committed to building a diverse team. We encourage applications from people underrepresented in AI research. We provide visa sponsorship for exceptional candidates.

Life's better with play.