Hard vs Fair: Training Better Cyber Agents (and COLM Pre-Party)

​Debate on the Frontier returns—this time as we celebrate going to COLM!

The Debate: Hard vs Fair

What makes a better training environment for cyber agents: harder tasks, or fairer ones? As agents move from benchmarks toward real systems, we’ll debate which kinds of challenge build capability we can trust—and which create noise, brittle incentives, or misleading scores.

Join us during the run-up to COLM for a lively discussion with guest speakers working at the frontier of cyber AI, audience Q&A, and the chance to meet researchers and builders.

​The COLM Pre-Party

🥘 Dosa + 🥛 Lassi

We're going to COLM! Come to our private pre-party hungry. We’ll have dosa and lassi throughout the evening.

📍 The Details

Thu, October 1, · 6:00–8:30 PM
Collinear Sunnyvale office
530 Lakeside Dr, Suite 101, Sunnyvale, CA 94085

Nazneen Rajani will be moderating a panel of expert guest speakers from the industry.

​About Us

Collinear builds agent evals, RL environments, and verifier-graded training data for frontier models. Our researchers will be at COLM and look forward to seeing you at the debate—or at your poster!

​Previous Debate

Our last Debate on the Frontier brought together researchers to argue, challenge assumptions, and connect over the future of AI. This next chapter brings the conversation to cyber agents—and to our new home in San Francisco.