Jason Mentors The Next Generation + Why AI Is Bad at Teaching (Aristotle Interview) | E2346

Jason’s advice to young finance students was to use AI to extend, not replace, formal coursework: identify gaps in skills such as reconciliation and modeling, study advanced topics alongside classes, and build personal AI agents for one-on-one instruction. He said independent projects could demonstrate AI fluency to employers. For venture-capital applicants, he recommended researching a fund’s investments, decisions, exits, and long-term returns, and reading at least 10 biographies a year, including outside one’s target industry, to build perspective on how businesses develop.

Jason told Agastia to anticipate that interviewers might stereotype him as an Indian immigrant and to distinguish himself through AI fluency, independent work, and broad interests. He asked David to read five biographies, take notes, and return with two lessons from each; his recommendations included Walter Isaacson’s Einstein biography and Stephen King’s On Writing. Jason said biographies can provide both practical lessons and inspiration. On a hypothetical $500,000 windfall, he said he might invest in skills, a company, markets, or himself; in 1996, he would have given his magazine more runway or angel-invested in technology companies.

Jason urged aspiring founders to learn company-building end to end by doing projects that involve taxes, contracts, LLC formation, payroll, hiring, and negotiation, arguing that narrowly scoped jobs can limit that experience. He said a startup could operate on $10,000–$20,000 a month with one employee, while spending $25,000–$50,000 monthly for 10 months could become a distraction. David said he would move from Long Island to the city to save about two commuting hours a day, wait for an idea he cared about rather than build around his first idea, and consider putting $250,000 into angel deals, including companies founded by NYU and Columbia students.

For a caller developing protein-folding technology for drug discovery, Jason stressed that AI in medicine was outside his expertise and characterized adoption as slow and methodical, with experimentation at the fringe. He said fringe projects can become startups that large pharmaceutical companies later acquire or copy, and recommended customer discovery for tools sold to pharmaceutical companies, which are easier to pursue than products requiring regulatory approval. He advised separating contact-finding and scheduling from discovery interviews, using a person or bot for outreach, and considering incentives such as coffee or gift cards. He acknowledged that interviews can expose an idea but was more concerned about other entrepreneurs copying it than employees at large companies; he urged legal research and warned against espionage or inducing anyone to breach an NDA.

Sean Ready described Aristotle as an AI lab focused on human learning, with a voice tutor aimed initially at high school students, the age group behind most U.S. tutoring demand. He argued that advances in general models’ reasoning, coding, and enterprise problem-solving have not automatically made them better teachers: a long answer may finish a task without helping a student understand it. Effective tutoring, he said, must preserve productive struggle rather than let students delegate learning. Aristotle aims to approximate working beside a human tutor by setting an agenda, asking questions, diagnosing misunderstandings, and adapting to a student’s level. Its system predicts the next pedagogical action, which could be an explanation, silence, an example problem, a diagram, or a visual highlight.

Ready cited Bloom’s 2 Sigma finding that one-on-one tutoring can produce a two-sigma improvement in learning outcomes, while saying fewer than one in 52 U.S. students has a personal tutor. Aristotle’s session-based pricing includes about $50 a month for eight sessions and about $200 for unlimited access; the host estimated annual costs of $600 and $2,400. Ready said tutoring depends on contextual cues, including voice, facial expression, attention, and how to interpret responses such as “hmm,” that text-only models often miss. Aristotle is building its own tutoring harness, benchmarks, and evaluations, hand-grading thousands of hours of sessions, and using multiple agents and specialized models for tasks such as generating math problems and reading whiteboards. Ready said sparse multimodal data makes one generalized model a poor fit for every tutoring task.

Ready identified dyslexic students and their parents as a particularly strong user market. He said session recordings could become valuable proprietary data but include information from students under 18, which Aristotle intends to keep locked down and not sell. He framed the product as a teaching assistant that could help teachers support students individually, while Jason predicted successful student-learning products would enter classrooms as teachers seek progress dashboards and ways to intervene. Ready envisioned AI-paced curricula alongside greater emphasis in schools on emotional, social, and physical education and shared experiences. A parent reportedly said her daughter learned more in three hours with Aristotle than in the previous month of public school; this was an individual account, not independently established evidence.