90/30 Club (ML reading) #50: LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics
- Date
- 2026-04-28
- Location
- San Francisco, CA, USA
- Host
- Luma
About this event
If you care about where machine learning is headed beyond benchmark chasing and brittle recipes, this session is worth your Monday evening. For the 50th edition of 90/30 Club, the group is digging into LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics—a paper that speaks directly to one of the biggest questions in modern ML: can we build strong self-supervised systems without leaning on a pile of ad hoc tricks? This is an in-person San Francisco reading group for people who like technical ideas in the company of other curious builders and researchers. Expect a focused conversation around the paper itself, but also a broader discussion about what its claims might mean for representation learning, autonomy, and how we think about scalable ML systems. What Is This? 90/30 Club is an ML reading gathering built around serious engagement with one piece of work at a time. Rather than skimming headlines or trading surface-level takes, the format is designed to help attendees read closely, compare interpretations, and pressure-test ideas with others who actually want to get into the substance. This edition centers on LeJEPA, a paper whose title alone signals an ambitious goal: self-supervised learning that is both provable and scalable, without relying on the usual heuristics. That makes it a strong pick for a group conversation, because it sits at the intersection of theory, systems thinking, and practical ML design. The event also carries the energy of a milestone: #50. If you have been meaning to find an ML community in San Francisco that values thoughtful technical discussion over posturing, this is a good room to step into. Because the tags include ml, autonomy, community, and networking, you should expect a mix of motivations in the room. Some people will come for the paper itself, some for adjacent ideas in autonomous systems and representation learning, and some because they want to meet others who care about these topics enough to spend an evening discussing them in depth. What to Expect The core of the evening is a shared reading discussion, not a passive talk. You should expect the paper to anchor the conversation: what problem it is addressing, what assumptions it makes, how the method is positioned against prior self-supervised approaches, and where its strongest or weakest claims may be. A typical flow for a reading event like this usually works best when it moves from orientation to analysis to open discussion. In practice, that means the evening will likely feel something like this: Brief arrival and informal conversation before the discussion gets going A framing of the paper and why this specific work was chosen Group discussion of the main ideas, methods, and claims Questions, disagreements, and comparative perspectives from attendees Time to continue conversations with people in the room afterward You do not need to show up with a polished opinion. What matters more is a willingness to engage seriously: asking where the paper is genuinely novel, whether “without the heuristics” holds up under scrutiny, and what “provable and scalable” should mean in a field where those words are often used loosely. Because this is in person, expect the conversation to be more dynamic than a comment thread or webinar. People can interrupt constructively, sketch intuitions, challenge assumptions, and connect the paper to their own work in a way that usually leads to sharper understanding for everyone involved. Why Attend If you read ML papers alone, you already know the tradeoff: you can move fast, but it is easy to miss key assumptions, over-credit a result, or gloss over what is actually new. A good reading group helps fix that. It gives you access to multiple technical lenses at once—research, engineering, product intuition, and theoretical skepticism—all applied to the same text. This paper is especially well suited to that format because it invites both close reading and debate. Claims about self-supervised learning without heuristics are not just technical details; they touch on bigger questions about robustness, simplicity, and whether current ML practice can be made more principled without losing performance. You should come if you want one or more of the following: A sharper understanding of an important self-supervised learning paper Better intuition for current debates in representation learning A chance to test your own reading against other strong technical perspectives Real conversation with people working on or around ML and autonomy A community setting that rewards substance over performance There is also straightforward professional value here. The people who make time for paper discussions tend to be the ones thinking carefully about foundations, not just shipping buzzwords. If you want to meet peers in San Francisco who care about ML at that level, this event gives you a natural context to start those conversations. Practical Details This event takes place in person in San Francisco, USA on Monday, April 27 at 7:00 PM PDT. It is an evening gathering, which makes it a practical option for people coming after work, research meetings, or a day spent heads-down on technical projects. Because the event is in person, plan for a format that rewards showing up ready to participate. If you can, read or at least skim the paper beforehand so you can follow the discussion more easily and contribute where you have questions or insight. Even if you have only reviewed the abstract, intro, or key claims, that will still make the session more useful. A few practical ways to get the most out of it: Bring notes or a few questions you want answered Be ready to explain what you think the paper is claiming in plain language Listen for where others disagree; those moments are often the most informative Leave time after the discussion if you want to meet people and continue talking If the title alone made you pause and think, “I want to know whether this really works and why it matters,” you are exactly the kind of attendee this event is built for.
Who should attend
This is for people who want a thoughtful, technically grounded ML conversation and prefer discussing real papers with real peers over passively consuming summaries. - You work in **machine learning, AI research, applied ML, or autonomy** and want to stay close to ideas that may shape how self-supervised systems are built. - You are the kind of person who reads paper titles like **“Provable and Scalable Self-Supervised Learning Without the Heuristics”** and immediately wants to know what is actually being claimed. - You enjoy **reading groups, research discussions, and technical debate**, especially when people are willing to challenge assumptions rather than just nod along. - You are building, studying, or exploring **representation learning, self-supervised learning, or adjacent areas**, and you want stronger intuition about current approaches and tradeoffs. - You want to meet a **San Francisco ML community** that values substance, curiosity, and good conversation, whether you are new to the scene or already well connected. - You do not need to be a specialist in this exact paper to belong here; if you are willing to prepare a bit, ask good questions, and engage seriously, you will get a lot out of the room.