Podcast Summary

Podcast: AI & I
Host: Dan Shipper
Episode: Why Your AI Learning Projects Keep Fizzling Out
Guest: Near Zickerman (Co-founder & CEO of Oboe)
Date: January 14, 2026

Episode Overview

This episode unpacks why so many personal AI learning projects start strong but quickly lose steam, diving deep into the mechanics of learning with AI tools. Host Dan Shipper interviews Near Zickerman, CEO of the AI learning platform Oboe, discussing what current LLMs (like ChatGPT) get wrong about learning—and what a true AI-powered learning platform should aim to do differently. They demo Oboe live on the show, explore how people really learn, and riff on both the philosophy and science of AI, knowledge, and the mind.

Key Discussion Points & Insights

1. Why Do AI Learning Projects Fizzle Out?

[00:00–04:30]

• Dan asks: Why shouldn't learners just use ChatGPT to learn everything? Why build an entirely separate platform for learning with AI?
• Near explains:
  • Modern LLMs are designed as general tools, not specifically for learning. They lack features specific to the learning process.
  • Most real-world learning is multi-modal and often passive, not purely through chat interactions.
  • People naturally use a combination of tools and formats: "You may use [LLMs] as one tool in the arsenal... but they are just one modality. People learn through multimodality." — Near Zickerman [03:41]

2. Active vs. Passive Learning

[04:30–07:25]

• Dan is surprised: Isn’t good learning always active and intentional?
• Near clarifies:
  • There are two dimensions: intentionality (objective-driven motivation) and the modality (active vs. passive).
  • Most people learn passively (consuming lectures, reading, etc.), with active engagement as just one reinforcement tool.
  • "Most of the time that you were learning, you were learning from [teachers] teaching to you... not actively participating." — Near Zickerman [06:28]

3. What’s Missing in LLM-Based Learning?

[07:25–10:01]

• LLMs compress information and are excellent for quick answers.
• True learning requires stepwise progress toward a goal, with "scaffolding" to bring users back to their objective—something LLMs lose easily in open-ended chats.
• "Anybody who has spent more than a few minutes in an LLM conversation... has probably learned LLMs are not great at that. They lose context very quickly." — Near Zickerman [09:43]

4. Live Demo: Building a Wittgenstein Course in Oboe

[10:01–24:25]

• Dan screenshares creating a custom course on Wittgenstein’s Philosophical Investigations.
  • Oboe makes the course feel "lightweight and achievable" by chunking material and giving instant progress, rather than overwhelming users.
  • Notable Quote: "A big part of our value proposition... is you can learn anything, anything that you ever thought was too hard to learn. We could at least get you started." — Near Zickerman [14:51]
• Speed & user experience: Oboe starts generating content instantly, so learners aren't left waiting.
• Multimodality: Oboe incorporates multiple formats (text, quizzes, AI-generated podcasts) to match learning style and topic.
  • "Multimodality is a big piece of what we believe in... you can't just be getting a bunch of text in the way that an LLM would give it to you." — Near Zickerman [16:48]
• Quiz/flashcards are “embedded formats” that Oboe injects when helpful for retention.

Demo Timestamps:

• [11:11] – Dan proposes a course on Wittgenstein
• [14:00] – Oboe generates the course live
• [16:48] – Explanation of multimodal content and in-course podcasts

5. User Perspective & Customization

[23:32–24:25]

• Dan asks: Can users tell Oboe to “go deeper” or tweak existing courses?
• Near: Currently, you refine by making more prompts; upcoming features will allow dynamic course editing and more granular control (e.g., adjusting depth, tone, or structure).

6. Oboe’s Growth & User Behavior

[24:25–27:55]

• Oboe launched September 2025; growth fueled by users’ strong objective-oriented learning goals.
• Over two-thirds of user prompts fall under “objective-based learning.”
• "Most people today struggle with this gap... I know what I want to achieve. I just have no idea how to get there." — Near Zickerman [25:44]

7. Why Motivation and Context Matter

[27:55–33:38]

• Dan recounts: Even with persistent reminders from LLMs, his learning stalls when material gets tough and context is lost.
  • "I had to do now even more work to build back up to this, and that's even less appealing." — Dan Shipper [29:09]
• Near: Agents need better guardrails and flexibility to help users recover context, reignite motivation, and dynamically adjust—just like a great human teacher.

8. Balancing Ephemerality and Substance

[33:38–35:52]

• Dan’s concern: Oboe courses feel ephemeral; he forgets to return to them.
• Near: The challenge is creating content that feels both lightweight and long-lasting—engaging enough to revisit but not intimidating.

9. What the CEO Is Learning & Physics Tangent

[35:52–44:08]

• Near uses Oboe to study advanced math and physics, like the Stern-Gerlach experiment.
• The conversation veers into quantum physics, history of experiments, and why Newtonian and quantum ideas permeate society differently.

10. AI, Embedding Spaces, and the Limits of Knowledge

[44:08–54:18]

• Dan draws parallels: LLMs’ high-dimensional, implicit knowledge resembles quantum mechanics’ probabilism and weirdness.
  • "There are dimensions of human output... that we're completely blind to." — Near Zickerman [46:38]
• Notable Moment:
  • Dan: "Language models know things we don't know... That’s because we think of knowledge as something explicit, but language models... embody a corpus of knowledge that exists in us in a different way." [50:28–51:30]
• Wittgenstein’s point: The limits of language, and by extension, of knowledge and what LLMs can “know.”
• Near reflects on whether the human mind is evolutionarily limited in self-understanding and the implications for AI and consciousness.

Notable Quotes

• Near Zickerman:
  • "LLMs are incredible... but they are just one modality. People learn through multimodality." [03:41]
  • "A real learning platform has to be built as a learning platform." [13:04]
  • "A big part of our value proposition... is you can learn anything, anything that you ever thought was too hard to learn." [14:51]
  • "There are dimensions of human output... that we're completely blind to." [46:38]
• Dan Shipper:
  • "At some point, I got to a piece where it was a little hard. And then... I had to do now even more work to build back up... and that's even less appealing." [29:09]
  • "LLMs know things that we don’t know. But that's because we think of knowledge as something... explicit. Language models... embody a corpus of knowledge that exists in us in a different way." [50:28]

Key Timestamps

• [00:00] – Introductions & big-picture vision for Oboe
• [03:41] – Multimodal learning vs. LLM-only learning
• [06:28] – Passive vs. active modalities in education
• [09:43] – LLMs’ weaknesses for long, goal-driven learning
• [14:00] – Live demonstration: building a course on Wittgenstein
• [16:48] – How Oboe’s multimodal/embedded formats work
• [29:09] – Why learners lose momentum and context
• [46:38] – Human limits vs. LLM dimensionality and implicit patterns
• [50:28] – Reconceptualizing knowledge: explicit vs. intuitive/embedded

Final Thoughts

This episode offers a nuanced take on why so many self-initiated AI learning projects lose steam: mainstream LLMs aren’t built as teachers, real learning is more complex and multi-dimensional, and the onus of structuring knowledge shouldn't fall solely on the learner. The conversation mixes practical product insights, big questions in philosophy and science, and a hands-on demo of Oboe, making it valuable for anyone interested in the future of AI-driven education—and the limitations and possibilities of human learning itself.

To try Oboe: oboe.com

For more from Dan Shipper: every.to/chain-of-thought?sort=newest