[Paper Reading]: Base Models Know How to Reason, Thinking Models Learn When
- Date
- 2025-10-16
- Location
- Zoom Link: https://us02web.zoom.us/j/81282737577, Fremont, CA, USA
- Host
- SupportVectors AI Events & Meetings
About this event
If you care about how modern AI systems actually reason, this reading session is built for you. We’ll dig into "Base Models Know How to Reason, Thinking Models Learn When" together, using the paper as a starting point for a sharper conversation about what reasoning models are doing well, where they differ from base models, and what that means for people building with or studying them. This is not a passive lecture and it is not a vague networking hour. It’s a focused, community-driven paper reading designed for people who want to show up, read critically, compare interpretations, and leave with a clearer understanding of an important idea in the current AI landscape. About the Event This event is a paper reading meetup centered on the paper "Base Models Know How to Reason, Thinking Models Learn When." The goal is simple: create a space where technically curious people can examine the paper’s claims, unpack the central argument, and discuss what the results suggest about model behavior in practice. Rather than treating the paper as a piece of hype or taking the title at face value, we’ll use the session to ask better questions. What does it mean to say a base model “knows how” to reason? What does it mean for a thinking model to “learn when”? How should we interpret these distinctions if we care about evaluation, prompting, product decisions, or research direction? The format is community-oriented and discussion-friendly. Expect a structured conversation rather than a formal presentation-heavy event. Attendees should come ready to think carefully, react to the paper’s framing, and build on each other’s ideas. Because this is a meetup setting, there is room for both close reading and broader interpretation. Whether you are approaching the paper from a research angle, an engineering perspective, or pure curiosity, the event is designed to make the conversation accessible without flattening the nuance. What to Expect The session will likely move through the paper in a practical, readable way, focusing on the core claims and the most discussion-worthy takeaways rather than trying to perform a line-by-line recitation. The emphasis is on understanding the argument, evaluating the evidence, and testing how the ideas hold up when translated into real-world reasoning about models. You can expect the evening to include: A quick framing of the paper so everyone is aligned on the main question and why it matters Guided discussion of key ideas and claims from the paper Space for interpretation and critique, including where attendees agree, disagree, or want more evidence Conversation with other attendees interested in AI, model behavior, and current research directions Time for questions and informal exchange, especially around implications for practice This is a good fit for active participation. You do not need to arrive with a polished take, but you should expect to engage. Often the most useful part of a paper reading is not just the paper itself, but hearing how different people parse the same result through different lenses. Because the event sits at the intersection of research discussion, community meetup, and networking, there’s also value in the side conversations. If you’ve been wanting a setting where you can talk seriously about AI ideas with people who are paying attention, this is that kind of room. Why Attend A lot of AI discussion happens at the level of headlines, summaries, and secondhand takes. This event offers something more useful: a chance to slow down and work directly with a specific paper that touches a live and important question in the field. You’ll leave better equipped to talk about the topic with precision instead of relying on surface-level interpretations. If you work with language models, study them, or simply follow the space closely, the distinction implied by the paper’s title is worth examining carefully. Claims about reasoning, model behavior, and the value of “thinking” approaches increasingly shape how people evaluate systems and make decisions about deployment, prompting, and product design. A strong group discussion can help separate what the paper actually supports from what people might be tempted to infer. Attending can help you: Sharpen your understanding of an active research conversation around reasoning models Practice reading papers critically instead of accepting conclusions too quickly Hear multiple perspectives from a community of interested peers Build better intuitions for how research claims map to practical use cases Meet others who enjoy substantive technical conversation in a relaxed meetup format It is also a good opportunity if you want more community around serious AI discussion. Not every event creates room for careful disagreement, curiosity, and collaborative interpretation. This one is designed to. Practical Details This event takes place Wednesday, October 15 at 7:00 PM PDT. The listed format is in person, with the location noted as Fremont, USA, and the event information also includes a Zoom link: https://us02web.zoom.us/j/81282737577. Because both an in-person location reference and a Zoom link are provided, attendees should pay close attention to the event listing and access details before the start time. If you plan to attend, it’s a good idea to confirm how you’ll join and be ready a few minutes early so the discussion can begin smoothly. A few ways to get the most out of the session: Read the paper in advance if you can, even if only once Bring a few questions or notes on the claims that stood out to you Be ready to discuss, not just listen Join on time, especially if you want the framing at the beginning This is a meetup for people who want thoughtful conversation, not performative hot takes. If that sounds like your kind of evening, you’ll likely find the discussion rewarding.
Who should attend
This is for people who enjoy serious ideas, clear discussion, and learning through conversation rather than passive listening. - You’re **curious about AI reasoning** and want to understand what current papers are actually claiming, not just how they’re summarized online. - You **work with language models** as a builder, researcher, student, or practitioner and want better intuition for how reasoning-related results may affect evaluation, prompting, or product decisions. - You like **reading papers in community** because other people’s questions and interpretations help you notice what you would have missed on your own. - You want a space where you can **ask thoughtful questions, disagree constructively, and test ideas** without needing to be the loudest person in the room. - You’ve been looking for a **meetup with substance**: something social and conversational, but anchored in a concrete topic worth unpacking. - You do not need to be a specialist, but you’ll get the most from this event if you’re **comfortable engaging with technical concepts** and interested in discussing them carefully.