Bay Area Frontier Research Club #23 | Founders Edition: The Knowledge Supply Chain (dinner + paper discussion @ AGI House SF)
A deep-dive into the knowledge supply chain for frontier AI.
Model improvement has a supply chain: expert human judgment on one end, the scientific record on the other. AT FRC #23, the people running both ends take the stage.
Curtis Northcutt — inventor of confident learning, co-founder & CEO of Cleanlab through its 2026 acquisition by Handshake, now leading AI Research & Strategy at the company supplying expert data to the frontier labs — opens with a question nobody has measured at scale: can AI teach as well as expert human tutors? He debuts unreleased research from his team, presented in this room a week before its public release.
Kalpit Dixit — founder of Paper Lantern, previously the AWS Bedrock scientist whose teams delivered trillions of pretraining tokens — closes with a controlled experiment: give Karpathy's autoresearch agent access to two million research papers, and it rebuilds GPT-2 with 26% less training time. Same agent, same starting conditions — the literature was the difference.
Founders Edition is for researchers who became founders, presenting the technical work inside their companies: methods, results, and open questions. Ten minutes on stage, then extended Q&A that goes as deep as the room wants.
Co-hosted with AGI House SF.
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 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 open
5:30pm – 6:30pm: Networking + light dinner
6:30pm – 8:00pm: Research presentations + discussion
8:00pm – 8:30pm: Networking
Presenters & topics
Talk 1: StudentBench — Can AI Teach as Well as Expert Human Tutors?
Curtis Northcutt — Head of AI Research & Strategy, Handshake AI · Co-founder & CEO of Cleanlab (acquired by Handshake, 2026) · Inventor of confident learning · MIT PhD
Few people have shaped how the field thinks about data quality more than Curtis. He invented confident learning during his MIT PhD under Ike Chuang, worked at FAIR under Yann LeCun and at Oculus Research, then co-founded Cleanlab and led it as CEO for five years — technology now used by over half the Fortune 500 — through its acquisition by Handshake, where he leads AI research & strategy.
Can AI teach as effectively as an expert human tutor — and what does it cost to raise a human's score by one point?
StudentBench tests a necessary condition for recursive human self-improvement: can AI teach as effectively as human tutors? In a study of thousands of real students, Curtis's team measured GRE learning gains after one hour of LLM tutoring versus expert human tutoring — and found them statistically equivalent. He'll present those results alongside expert evaluations of teaching quality, tutoring-cost comparisons, and questions StudentBench makes answerable for the first time — including the cost to augment a human's intelligence by one percentage point on a standardized exam. Presented in this room a week before the paper's public release.
Pre-read: Northcutt, Jiang & Chuang, Confident Learning: Estimating Uncertainty in Dataset Labels, Journal of Artificial Intelligence Research.
Talk 2: Scientific Literature in the Autonomous Research Loop
Kalpit Dixit — Founder of Paper Lantern · Former Senior Applied Scientist, AWS Bedrock · Stanford MS · IIT Bombay
Kalpit has spent his career at the data layer of frontier AI. At AWS Bedrock, his teams delivered trillions of pretraining tokens and shipped multiple AI products to GA; he's published NLP research at Amazon Science; and he now builds Paper Lantern, which puts the world's research literature — more than two million papers — directly inside the reasoning loop of AI agents. He brings the receipts: a controlled experiment on what the literature is actually worth.
When an autonomous agent improves a training run, is it discovering something new — or finally reading what the field already knew?
Kalpit will present a case study connecting Paper Lantern to Andrej Karpathy's autoresearch framework: two agents, same starting conditions, one difference — access to the literature. After training the best configuration from each run for two hours, the literature-grounded agent reached 3.2% lower validation loss on a roughly seven-million-parameter language model — and hit target loss with 26% less training time. The discussion will examine the experimental setup, which research-informed changes actually helped, and what the results establish about literature access in autonomous research.
Pre-read: Paper Lantern Improves Autoresearch.
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
Researchers working on human data, post-training, and evaluation
Founders and engineers building AI agents and supporting infrastructure
Teams developing autonomous research and model-training workflows
Investors focused on AI infrastructure and research-driven companies
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.
Previous Session Recap — The 20-Watt Problem
At Mission Robotics, FRC #22 examined energy efficiency in biological and artificial intelligence. Marta Gajowa, UC Berkeley neuroscientist and founder of Neuraffica, discussed how living neural circuits represent and process information. Ezra Wolf, founding silicon engineer at Zettascale Computing, examined energy use in AI computation, memory bottlenecks, and accelerator design.
The presentations were followed by discussion with both speakers and robot demonstrations. Recordings will be shared on our YouTube channel.
🌐 Connect with Frontier Research Club
Luma Calendar: luma.com/frontiersyndicate
YouTube: FrontierResearchClub
LinkedIn: frontier-research-club
Instagram: @frontierresearchclub
Email: kristopher@frontiersyndicate.vc
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.