Cover Image for The Debrief · Lunch & Open Q&A

Presented by

Scale with Precision. No pitches, just progress.

San Francisco is buzzing with AI innovation. But a great idea is only 10% of the battle. The other 90% is finding the talent and support to build your models fast — with reason, judgment, and real skill.

Whether you're in town for CoreWeave Fully Connected, The AI Conference, or you build here year-round, stop by. No decks, no sales pitches, no panels. Drinks, a short walk from the show floor, and a room of engineering leaders, GenAI researchers and AI leads.

The Harlequin, 68 4th St. Tuesday Sept 29, 3 PM to 5 PM.

Hosted By

The Debrief · Lunch & Open Q&A

Name: The Debrief · Lunch & Open Q&A
Start: 2026-10-05T12:00:00.000-07:00
End: 2026-10-05T13:00:00.000-07:00
Location: Virtual — joining link sent on registration

Hugo × Frontier Labs · Bay Area

Virtual — joining link sent on registration

Approval Required

Your registration is subject to host approval.

Welcome! To join the event, please register below.

About Event

🍜 Start Monday with lunch on us. Bring a question.

Three days of CoreWeave Fully Connected and The AI Conference, then a weekend to let it settle. Most of the good questions never got asked out loud. This is where they get asked.

Here is the deal. Tell us the question you did not get to ask. Then build your lunch order on Uber Eats and paste the share-cart link on the form. We pay for it, it lands before we start, and we answer your question live with the people who actually do this work.

🎙 What we can actually answer

• Process Supervision vs. Outcome Bias: Why grading step-by-step reasoning trajectories (PRMs) beats simple final-answer reward functions.

• The Sourcing Paradigm Shift: How native academic English and process-heavy education architectures change the math on global talent selection.

• Agentic evaluation that survives contact with production, and what breaks first.

• Auditing chain-of-thought when the answer is right and the reasoning is not.

• Process reward models, RLHF preference data, and where they get gamed.

• What expert capacity costs, how fast it can be stood up, and what quality actually holds at volume.

• The unglamorous operational stuff: inter-annotator agreement, first-pass quality, throughput, and how you know any of it is real.

• Deployment Velocity: How to deploy dedicated evaluation pods in days and get calibrated outputs on your internal tooling within 48 hours.

📮 How this works

Everything is on the registration form. The question is required and we read every one before approving a spot, so make it a real one. The lunch bits are optional. Get your order in by the Friday before so it arrives on time, and you are very welcome to skip lunch entirely and just come for the conversation.

No slides. Sixty minutes. If we do not know the answer we will say so.

Enjoy lunch, connect with fellow AI engineering directors, and get direct answers to your hardest post-training questions.

🧠 About Hugo Inc.

Hugo is your verification and judgment layer for agentic and reasoning systems. 730+ vetted specialists, 100 percent university-educated, 64 percent holding four-year STEM or CS degrees, working inside your environment rather than around it.

Full rundown: hugoinc.com/industry/frontier-reasoning-labs

Hosted by the Hugo team.

Location