The researchers-turned-founders take the stage at AGI House.

Founders Edition is FRC's recurring session for founders who were researchers first — PhDs and lab alumni now building companies on top of their own results. This is not a pitch night: each founder presents the open research problem inside their company — real data, working systems, falsifiable claims — with the same rigor the work had in the lab. Same bar as every FRC session; being a founder earns no discount on it. Co-hosted with AGI House SF.

This session is built around one question: what keeps an agent honest? As agents take on real work, the answer lives in two grounding layers. The first is expert human judgment — the delta between plausible and correct that only expert data closes, and that frontier labs pay real money for. The second is the scientific record — what an autonomous research agent gains when it can actually read the literature instead of navigating by feel. This session takes the question from both directions: one talk from the person building the judgment layer for frontier labs, one from the person who gave an autonomous researcher the map.

The Frontier Research Club is a curated forum for rigorous, technical discussion at the frontier of AI. We convene researchers from the frontier labs, Stanford, Berkeley, and the teams building in production to examine concrete work — papers, methods, and results — with a bias toward assumptions, evaluation methodology, failure modes, and what would count as convincing evidence.

Presentations are intentionally brief so the majority of time is reserved for questions and critique. Materials are shared in advance so the conversation starts at depth.

Agenda

5:30pm: Doors open5:30pm – 6:30pm: Networking + light dinner6:30pm – 8:00pm: Research presentations + discussion8:00pm – 8:30pm: Networking

Presenters & topics

Talk 1 — Ground Truth: The Expert Human-Data Layer

What is the delta that only expert human judgment closes — and why do frontier labs pay for it?

Curtis opens the night from the buy side of ground truth: what expert human data actually is, where models still fail without it, what frontier labs purchase and why, and where the judgment layer for agents goes next. He'll be debuting unreleased research from his team — shared with this room first, ahead of a public release the following week. The paper stays in the room until it drops; attendees get it, and the announcement to reshare, the day it does.

Curtis Northcutt, PhD, is Director of AI Research at Handshake. He co-founded Cleanlab and led it as CEO — $30M raised, 100+ Fortune 500 companies served — through its 2026 acquisition by Handshake. He holds a computer science PhD from MIT, where he invented confident learning, the foundation of the widely used open-source cleanlab library for finding and fixing label errors in real-world datasets. He writes at curtisnorthcutt.com.

Pre-read: Northcutt, Jiang & Chuang, "Confident Learning: Estimating Uncertainty in Dataset Labels" (JAIR) — https://arxiv.org/abs/1911.00068

Talk 2 — What the Literature Is Worth: An AI Agent Rebuilds GPT-2, 26% Faster

When an autonomous research agent improves a training run, is it discovering something new — or finally remembering what the field already knew?

Kalpit connected Paper Lantern — which gives agents the ability to search, read, and reason across 2M+ research papers inside the reasoning loop — to Karpathy's autoresearch framework and set it loose on rebuilding OpenAI's GPT-2. The literature-grounded agent surfaced 32 research-backed ideas that combined to cut training time by 26% — and in head-to-head runs reached 3.2% lower validation loss at 10% lower training cost than the same agent without the papers. Same ideas available to both agents; one had the map. The talk walks through what changed, and what it says about grounding autonomous research.

Kalpit Dixit is the founder of Paper Lantern, building research-grounded autonomy for model pre- and post-training. He's a Frontier Research Club regular, returning for his third talk after sessions at Stanford and the Pebblebed self-improving-agents night.

Pre-read: Paper Lantern autoresearch case study — https://www.paperlantern.ai/blog/auto-research-case-study

Want to present your work?

If you have a research paper you’d like to discuss at one of our next sessions, please submit it for consideration. ​Submit your paper here!

Who should attend

  • Founders building agents, evals, and the infrastructure underneath them

  • Researchers in post-training, human data, and evaluation

  • Frontier-lab data and eval teams — the people who buy and build ground truth

  • Engineers working on autonomous research and training pipelines

  • Investors backing agent infrastructure

Capacity is limited.

We will take photos and short video clips for event recap and promotion. By attending, you consent to being photographed and recorded, and to the use of those images and clips by the organizers on social media and other event marketing channels.

Last Session Recap — FRC #22: The 20-Watt Problem

Last week the club took over Mission Robotics — 8,500 square feet of brick-and-timber robotics floor, a dozen robots in the room — for a session built around one number: the twenty watts on which the human brain runs general embodied intelligence. Marta Gajowa (UC Berkeley neuroscientist; Founder & CEO, Neuraffica) opened from biology's side, drawing on fifteen years reverse-engineering how living neural circuits represent and process information — including technology that writes activity directly into living circuits and watches them learn. Ezra Wolf (founding silicon engineer, Zettascale Computing — and an FRC regular who crossed from the audience to the stage) answered from silicon's side: where the watts actually go when a model runs, the memory wall, and how far accelerator design can close the gap toward the brain's budget. A room at capacity, guided discussion with both speakers, then robots. Talk videos are coming to our YouTube.

🌐 Connect with Frontier Research Club

Hosted by

Frontier Syndicate is a venture community connecting frontier tech researchers, builders, and investors through curated convenings and early-stage capital. Across the Bay Area, we host a recurring series of research forums, builder nights, and intimate investor dinners — and back exceptional companies emerging from the labs, communities, and technical networks we convene.

Ascension by AGI House SF is a community of AI founders and researchers accelerating humanity's transition to AGI, hosting merit-based gatherings, hackathons, and technical events that draw leading AI minds from around the world.