Poppy Crum on Huberman Lab: How AI Reshapes Learning

Crum distinguishes between using AI as a cognitive amplifier versus a cognitive replacement. When you use LLMs to *generate* work (e.g., write the paper for you), you drastically reduce “germane cognitive load”—the mental effort that builds schemas and long-term understanding. When you use AI to *test* yourself, surface weak spots, quiz you, or structure a learning plan (as Huberman does with papers), you preserve and even enhance that deep learning while gaining efficiency.

Powerful neuroplasticity comes from closed loops: clear goals, measurable performance, and real-time feedback. Crum’s Stanford course has students design tools that sense performance (e.g., contrast sensitivity in gamers, acceleration in soccer players), compute relevant metrics, and feed back auditory or other cues. This increases ‘neural resolution’—the brain’s ability to detect fine differences—and accelerates skill acquisition in sports, music, and cognitive tasks.

You don’t need to code to build useful AI tools. Crum describes using AI and no-code platforms (e.g., Perplexity Labs, Replit, mobile computer-vision kits) to create apps that analyze swimming strokes or running gait from phone video, then output actionable metrics (entry angle, velocity, cadence consistency). The same structure can be applied to work focus, posture, or any repeatable behavior: capture data → define metrics → have AI create the analytics and basic interface.

Crum argues we won’t need to cover ourselves in sensors. Microphones, cameras, CO₂ sensors, and thermal imagers in homes, vehicles, and offices—and AI to integrate them—can infer heart rate, breathing, stress, focus, and group states non-invasively. HVAC systems, lighting, sound, and haptics can then be adjusted to support current goals (deep work vs. rest vs. social connection) much like an advanced sleep mattress already adjusts temperature to optimize sleep stages.

Crum proposes a simple lens for tools and AI agents: are you using them to (1) give you *more* cognitive skill and insight, or (2) to replace skills just to go faster? GPS is a classic replacement—extremely useful but it degrades spatial mapping. Using AI to compress research, surface patterns, or detect health risks can make you smarter. Using it to do all the hard mental work (writing, reasoning) can produce short-term gains but long-term losses in capability and independence.