Mox Taco Tuesday: How Neural Networks Think in Shapes (Goodfire's Neural Geometry)
- Date
- 2026-06-17
- Location
- Door code will be posted day of event!, San Francisco, CA, USA
- Host
- Mox
About this event
Most people use neural networks every day without having a clear mental model of what they are actually doing. This Taco Tuesday session digs into a more intuitive way to understand them: not as black boxes, but as systems that organize information into shapes, boundaries, and structure. If you are curious about AI and want a sharper way to think about how models represent the world, this is a strong place to start. About the Event Mox Taco Tuesday: How Neural Networks Think in Shapes centers on Goodfire's neural geometry, a way of looking at neural networks through the lens of spatial structure and representation. Rather than staying at the level of buzzwords, the event is built around the core idea that modern AI systems can be better understood by examining how they arrange concepts, relationships, and decisions in high-dimensional space. This is an in-person community gathering in San Francisco, designed for people who want a thoughtful, social, and grounded conversation about AI. The Taco Tuesday framing suggests a relaxed atmosphere, but the topic itself is substantial: how neural networks encode meaning, how internal representations form, and why geometry offers a useful language for understanding model behavior. Expect something that sits between a technical meetup, a learning session, and a community conversation. You do not need to arrive with all the answers. The value here is in showing up ready to listen, ask questions, and compare notes with other people who care about machine learning, neural networks, and where AI research is heading. What to Expect The evening begins at 6:30 PM PDT on Tuesday, June 16 and takes place in person in San Francisco. This is not a passive webinar or a loosely defined networking mixer. The event has a clear topic and a clear point of view: neural networks can be understood more deeply when you think in terms of shapes, directions, clusters, and geometric structure. You can expect a format that supports both learning and conversation. Depending on the flow of the night, that may include a focused talk, an explanation of key ideas behind neural geometry, and time for discussion with other attendees. Because the event is tied to a specific concept rather than general AI chatter, the conversation is likely to stay concrete and idea-driven. Likely themes include: How neural networks represent information internally Why geometry is a useful metaphor and analytical tool for AI systems What it means for a model to separate, cluster, or transform concepts How better interpretability can lead to better intuition about model behavior Open questions and practical implications for people building or studying AI There is also a community element built in. With tags spanning neuralnetwork, ai, community, and networking, this is a good setting for meeting people who are actively thinking about similar problems from research, engineering, product, or pure curiosity angles. If you enjoy the kind of event where the best conversations continue after the main presentation, this format should feel natural. Why Attend A lot of AI discussion gets stuck in two modes: either overly abstract hype or highly specialized jargon. This event offers a better middle ground. It gives you a conceptual handle on neural networks that is intellectually serious without requiring you to sift through a full research paper alone. If you work with machine learning systems, understanding representation and geometry can improve the way you reason about models. Even when you are not doing interpretability research directly, better intuitions about how networks organize information can influence how you debug, evaluate, and communicate about AI systems. If you are earlier in your AI journey, this is useful for a different reason. It can help replace vague mental models with sharper ones. Instead of thinking of a neural network as magic hidden layers doing mysterious work, you start to see a framework for asking better questions: What is being separated? What is being grouped? What changes from layer to layer? What kind of structure is the model learning? You should also attend for the people in the room. San Francisco has no shortage of AI events, but topic quality matters. A session focused on neural geometry is likely to attract attendees who want more than surface-level takes, which makes the networking more relevant and the conversations more rewarding. Practical Details This event is in person in San Francisco, USA. The listed location notes that the door code will be posted the day of the event, so plan to check event updates before you head over. Since access details are being shared same-day, it is worth giving yourself a little buffer before arrival rather than cutting timing too close. Key details at a glance: Event: Mox Taco Tuesday: How Neural Networks Think in Shapes (Goodfire's Neural Geometry) Date: Tuesday, June 16 Time: 6:30 PM PDT Format: In-person community event Location: San Francisco, USA Entry note: Door code will be posted on the day of the event Because this is an in-person evening gathering, expect a more conversational pace than a formal conference talk. Bring your questions, your current AI confusions, and your best attempt at explaining how you think models work. This is the kind of event where a single good idea can reshape how you read papers, evaluate systems, or talk about neural networks with other people in the field.
Who should attend
This is for people who want a more precise, intuitive way to think about AI than the usual surface-level conversation. - **You work with machine learning or neural networks** and want better mental models for how representations form inside models, not just how to call an API or train from a template. - **You are AI-curious but thoughtful about fundamentals** and want a grounded introduction to concepts like representation, geometry, and interpretability without needing a full academic background. - **You enjoy technical community events** where the topic is specific enough to attract serious interest, but the atmosphere still leaves room for conversation and meeting people. - **You are a founder, product builder, researcher, or engineer** trying to keep up with deeper shifts in AI understanding, especially around how models organize information internally. - **You like asking “what is the model actually doing?”** and want language that helps you reason about clusters, boundaries, structure, and internal concepts. - **You are based in or around San Francisco** and want an in-person evening event that combines learning with relevant networking around neural networks and AI.