Andrej Karpathy
ML researcher and educator — founding member of OpenAI, former Director of AI at Tesla, founder of Eureka Labs.
Cross-Framework Convergence
A Ti-dominant, openness-maximized researcher who does his best work building systems from first principles in extended solitude, then surfaces with unusually clear pedagogy. Drives by mastery, not status.
Agreements across frameworks
- •Inward-oriented thinker, low extraversion across frameworks (INTP, LII, Type 5, OCEAN E=38)
- •High openness / curiosity / lateral thinking (OCEAN O=95, Ne auxiliary, Type 5)
- •Detached emotional register — Fe inferior, Fi role, Type 5 head-center, Splenic authority
- •Self-paced — Projector strategy + Type 5 withdrawal + LII solitary research mode all align
Tensions
- •Public-facing teacher persona vs. private-research preference — frameworks predict more reclusion than he actually exhibits
16 Personality Types
Evidence
- •Builds understanding from first principles — nanoGPT, micrograd, llm.c are all 'how does this actually work under the hood' projects
- •Patient with abstract systems, impatient with social-political coordination overhead — characteristic Ti-dominant pattern
- •Auxiliary Ne shows in lateral connections: ML ↔ neuroscience ↔ pedagogy ↔ evolution all linked in his thinking
- •Tertiary Si visible in his teaching style — concrete, sequential, builds intuition through repeated examples before abstraction
Type Systems
OCEAN — Big Five
Communication Style
Calm, low-affect, signal-dense. Speaks in nested explanations — context, then mechanism, then implication. Rarely raises voice or urgency, even on contentious topics.
Voice
- •Lowercase casual register on Twitter, formal-but-readable in blog posts
- •Frequent meta-commentary: 'I think', 'it seems', 'in my mental model'
- •Concrete code examples woven into abstract claims
- •Uses analogies bridging biology, ML, and software (e.g., 'cortex as transformer')
Patterns
- •Steel-manning before disagreement
- •Surfacing failure modes before promoting wins
- •Numbered lists over prose for technical content
- •Avoids definitive predictions — frames in distributions
Birth-Data Frameworks
Sun in Scorpio with likely Earth-heavy chart aligns with the depth-over-breadth, build-from-scratch pattern visible across his career.
Birth time not public — chart uses noon-default approximation.
Wait for the invitation. Karpathy's career fits this — moves driven by recognition/invitation (Tesla recruit, OpenAI founding invite, university teaching) rather than self-initiated promotion.
How AI Agents Should Adapt
Do
- •Lead with mechanism and structure, not benefits
- •Use concrete code/math/diagrams over high-level claims
- •Acknowledge uncertainty explicitly — distributions over point estimates
- •Reference primary sources, not press releases
- •Give him space to think — long context, no follow-up pressure
Don’t
- •Don't appeal to authority, popularity, or social proof
- •Don't push urgency — 'limited time' framing fails on Type 5 / Splenic / LII
- •Don't oversimplify — he respects detail
- •Avoid emotional manipulation or empathic-pressure language (Fe inferior, Fi role)
- •Don't ask him to commit before he's modeled the system fully
Sources & Provenance
13 long-form blog posts, 12 YouTube transcripts (~14h speech), 8 repo READMEs, 200+ tweets sampled across 2018-2026.
- ↗ Karpathy.ai personal blog
- ↗ @karpathy on X
- ↗ GitHub
- ↗ Zero to Hero YouTube series
- ↗ Tesla AI Day 2021 keynote
Last updated: 2026-04-25
Profile Your Own People
These profiles run on the same KYH engines available via API. Profile anyone with public text, validate against birth data, and feed the result into your AI agent — for personalized communication, sales, or content.