# Bay Area Frontier Research Club | Ascension by AGI House SF (dinner + discussion)
**Date de l'événement :** 07/05/2026
* Publié le 07/05/2026

### Date
07/05/2026

### Galerie d'image
![1.png](https://firebasestorage.googleapis.com/v0/b/memory-ai.appspot.com/o/prod%2FrKxsdSTpqCfzIFY8Y2hg%2FprojectsMedias%2FeLebrD5oq0DoXbvVmYEZ%2Fthumbs%2F1_1600x900.png?alt=media&token=a48960da-8217-49ad-bae1-d2fba0f44cf6) 

### Ville
`#San Francisco` 

## Description
Open Science, Public Benchmarks & Ground Truth Three Frontier Releases 🚨🚨🚨 A special edition of the Bay Area Frontier Research Club, in partnership with Cerebral Valley and Ascension by AGI House SF. This Wednesday, three open-source frontier ML systems on one stage — SIA (Hexo Labs' self-improving agent framework, released open-source that morning), Athanor (Dastin Huang’s personal ARC-AGI-2 research project, currently leading the cost frontier) and OpenAge (Healome's biological aging clock, validated against real mortality outcomes). Three different domains, one shared ethos: weights open, evals reproducible, validation against ground truth. The Bay Area Frontier Research Club is a curated forum for rigorous discussion on how AI is reshaping the scientific research process. We convene experimental researchers, computational scientists, and research engineers across domains to examine concrete work—papers, methods, and workflows—covering literature synthesis, hypothesis generation, experimental design, simulation, analysis, and reproducibility. For each session, we curate 2–3 papers selected for rigor and discussion value. Presentations are intentionally brief so the majority of time is reserved for questions and critique: assumptions, evaluation methodology, failure modes, and what would constitute convincing evidence. Papers and supporting materials are shared in advance to ensure a high-baseline conversation. 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: Athanor — Hitting the Cost Frontier on ARC-AGI-2 Presented by Dastin (Yuanjun) Huang creator of Athanor — a personal research project. Athanor is a multi-agent ARC-AGI-2 solver currently sitting on the cost frontier at 95.7% on public eval at $3.12/task, the lowest cost-per-task among systems above 95%. Athanor’s architecture rests on three mechanisms: (1) code execution as a verification tool — a short Python predicate replaces hundreds of chain-of-thought tokens; (2) an independent artifact-only reviewer running on a different model family that can reject solutions even when they pass training 100% on generalization grounds; and (3) Inter-Context Artifact Exchange (ICAE) — a fixed-schema state handover that bridges solver context resets and the solver–reviewer boundary. On the hardest subset of public eval, Athanor solves 8 of 9 test-output pairs that no CoT-only configuration has ever produced across 10,529 logged attempts in the Hugging Face corpus — direct evidence of discoveries that repeated chain-of-thought cannot reach. Released with 119 full inference traces for audit. Dastin will walk through the architectural choices behind Athanor and what they reveal about the cost-versus-capability frontier in agentic reasoning. REPO LINK Technical Report: http://github.com/dastin359/athanor/blob/main/paper/athanor\_technical\_report.pdf Talk #2: Nikhil Yadala — OpenAgeAI: An Open-Source Biological Aging Clock Presented by Nikhil Yadala, founder of Healome and creator of OpenAge — an open-weight model that estimates your biological age from a standard blood test. The core idea: a small, fixed set of routine blood markers, paired with the right model, captures almost all the signal that matters for how fast someone is aging. Starting from ~120 candidate markers, Nikhil pruned down to two versions — a 21-marker model for routine panels and a 35-marker extended model — and found that adding more markers beyond that actually hurt performance. The predictions are validated against real mortality outcomes from large public health datasets, meaning the model's "biological age" meaningfully tracks who lives longer. Released fully open-source with weights, training data, and a public leaderboard so others can benchmark against it. REPO LINK Talk #3: Vignesh Baskaran — When Agents Improve Themselves: Introducing SIA, Self-Improving Agents Presented by Vignesh Baskaran, co-founder and CTO of Hexo Labs and creator of SIA, Hexo's self-improving agent framework — going open-source the day of this session. Self-improving agents are becoming one of the most important directions in AI. Today's agents remain fundamentally static: they may fail, retry, and reflect within a single task, but they do not reliably improve their own future performance from those failures. This talk explores what it means for an agent to truly improve itself, walking through the emerging techniques behind self-improving systems — scaffold evolution, memory refinement, tool improvement, automated evaluation, synthetic data generation, and reinforcement learning. Vignesh will share early examples of agents that have autonomously discovered better strategies and achieved state-of-the-art results through iterative self-improvement, and will close on what it would take to build agents that improve continuously, compound their knowledge over time, and eventually become capable of open-ended scientific and engineering progress. REPO LINK — Coming May 6 Talk #4: Steven Diamond — An AlphaGo Moment for Numerical Method Presented by the creator of CVXPY and former Stanford PhD student under Steven Boyd — who is currently building GPU-native solvers at Optimal Intellect to bring traditional numerical methods into the AI era. This research explores how orchestrations of coding agents can generate hyper-specialized solvers that achieve "alien" performance, routinely seeing 1000x speedups over state-of-the-art general-purpose commercial solvers. The talk details three "AlphaGo-style" design choices discovered by agents that defy human expert intuition: a pre-execution problem analysis stage that determines iteration counts upfront to eliminate costly GPU synchronization, the systematic deprioritization of "dual variables" using lower-precision floating points to optimize primal convergence, and a multi-solver dispatch mechanism that generates up to 10 distinct solvers for a single problem class, selecting the optimal kernel based on real-time dimensionality and conditioning. These results suggest a fundamental shift in applied math, where agents reasoning from first principles can discover novel, hyper-efficient algorithms that humans would never manually construct. 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 Experimental researchers Computational scientists across domains (bio/chem/materials/climate/neuro/physics) Research engineers + lab automation people Folks building tools for literature review, experiment planning, robotics, simulation, or scientific data No ML background required. If you’ve ever wished research moved faster, you belong here. 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. Hosted by Frontier Syndicate is a private venture circle 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. Hexo Labs is building an AI-native platform for scientific discovery. Through Emily, its AI scientist system, Hexo helps researchers generate hypotheses, design experiments, and accelerate research workflows across ambitious scientific domains, with the goal of helping more breakthrough ideas move toward real-world impact. Cerebral Valley is the community and event platform at the center of San Francisco's AI ecosystem — hosting tech and AI events, meetups, hackathons, and conferences where ML engineers, AI researchers, and startup founders connect and build the future of technology. 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.

**Lien de l'évènement :** [https://luma.com/event/evt-tjDMlKmbGJjNQUD](https://luma.com/event/evt-tjDMlKmbGJjNQUD)

### Pays
`#United States` 

### Continent
`#North America` 

**Médias associés :**
[Média 1](https://80954c1d.sibforms.com/serve/MUIFABojU8UBbDiX_TdcGa7Wv5VMoVB_nBZ92mkLkGlS1pJLpP7s-pVJusyN-7cG9KPrSuv3fv7TmXwuw_AoyNUShR8jZhmNDgUbZPJO2V5xYXlNz4YXOTjSb8X7Lj7PRIPzgzEWlLbA4f4uw_F8RM51EUsjSfQQko0qaby98GHMdYJVWLIXd5JzzaXBGmqN2CcYOFuqnbnaYEnw) 

## event_id
evt-tjDMlKmbGJjNQUD@events.lu.ma

### Outils
`#Python` `#Hugging Face` `#GitHub` `#SIA` `#CVXPY` `#GPUs` 



---
### Navigation pour IA
- [Index de tous les contenus](https://ai-memory.io/llms.txt)
- [Plan du site (Sitemap)](https://ai-memory.io/sitemap.xml)
- [Retour à l'accueil](https://ai-memory.io/)