Suggestions
Mia Mabanta
Head of Programs and Talent at YC Continuity
Mia Mabanta is the Head of Programs and Talent at YC Continuity, the growth-stage fund of Y Combinator, where she has been in this role since April 2018. In her position, she focuses on working with the top 1% of Y Combinator startups, which include well-known companies like Stripe, Coinbase, and DoorDash.13
Professional Background
Mia has a diverse professional history. Before joining Y Combinator, she was the Head of Brand and Product Marketing at Quartz, where she played a crucial role in scaling the company significantly in terms of revenue and workforce. Her efforts included leading market validation and product storytelling, which contributed to Quartz's growth during her tenure.12
Additionally, Mia co-founded HelloWallet, a fintech startup that was acquired by Morningstar, which stemmed from her research at the Brookings Institution. Her early career involved working as a research analyst at Brookings, where she contributed to initiatives focused on underserved consumer markets.12
Education
Mia holds an MBA from the Stanford University Graduate School of Business and a Bachelor's degree from Colgate University.12
Personal Insights
Mia has shared insights about her career journey, emphasizing the importance of resilience and adaptability. She immigrated to the U.S. at 17 and has since developed a strong ability to navigate challenges, which she attributes to her early experiences.2
Highlights
The way we align and shape model behavior is evolving quickly. Last week, we had the pleasure of gathering SF's post-training experts from companies like @appliedcompute, @OpenAI, @cursor_ai, @thinkymachines, @mercor_ai, @cognition, @xai, @flappyairplanes, and others.
Post-training is a nascent field, and I learned a ton:
1/ Reward modeling is becoming the bottleneck. The gap between what we can train models to do and what we can reliably evaluate is widening fast.
2/ The line between post-training and product is blurring. Teams that treat RLHF and fine-tuning as an afterthought are falling behind those that treat it as a core part of the product loop.
3/ Evaluation is still surprisingly artisanal. The tooling isn't where it needs to be, and there's a real opening for better infrastructure.
This event was part of @firstharmonic's Builders Table series: intimated, curated dinners where you almost certainly won't be the smartest person in the room. DM me if you're interested in coming to the next one.


