Principle · MindLiemandt argues kids will do hard things and meet high adult-set standards when those standards are connected to something the child loves and feel like a fair trade. He illustrates this with his schools' approach: sports participation is gated by academic requirements, and upper high schoolers can earn their cell phone back because the real world won't hide it from them — they must learn self-regulation. He also warns that pounding kids into a cell they don't want is a major cause of mental health issues in elite schools, so the standards must be paired with something the kid genuinely wants.
presented as Liemandt's endorsed heuristic
Andrew Huberman · How to Accelerate Learning & Improve Education | Joe Liemandt
Creator perspective · MindLiemandt argues that kids want to do hard things and have to do hard things, that the key to a child's happiness is high standards, that he could not run a school kids love more than vacation if he had low standards, and that kids think they're awesome and they are awesome, but need an environment to do hard things with their friends.
Andrew Huberman · How to Accelerate Learning & Improve Education | Joe Liemandt
Creator perspective · MindJoe Liemandt argues that measuring schools on achievement is a big problem because good schools are often just selection effects; schools should instead be measured on the rate of learning — the slope of the line — while looking at both whether kids are at grade level and how fast they're learning, since a sufficient slope means they'll eventually get there.
Andrew Huberman · How to Accelerate Learning & Improve Education | Joe Liemandt
Principle · MindJoe Liemandt states as a principle that life skills are trainable just like other skills — you can teach grit, hard work, and financial literacy.
Andrew Huberman · How to Accelerate Learning & Improve Education | Joe Liemandt
Creator perspective · MindLiemandt argues that with AI tutors, every kid can now be given perfectly tailored lessons at their level, at a mastery level, cost-effectively, removing the historical economic barrier to one-on-one education.
Andrew Huberman · How to Accelerate Learning & Improve Education | Joe Liemandt
Principle · MindWorking memory holds a limited number of unique chunks; using slots for one thing means they can't be used for another. Fluency in basics like multiplication tables frees slots for higher-order thinking, which is why the speaker claims he can predict who will struggle with a math problem even among students who know the material.
described as the speaker's account; not externally verified
Andrew Huberman · How to Accelerate Learning & Improve Education | Joe Liemandt
Principle · MindLiemandt describes two failure modes: throwing a kid in with high expectations but no support, and shielding a kid from all stress with low standards and high support — both are bad. The fix is Dr. Yeager's mentor mindset program (high standards, high support), which all adults in Alpha buildings follow, with a coach or guide positioned to support the kid while they struggle.
Andrew Huberman · How to Accelerate Learning & Improve Education | Joe Liemandt
Creator perspective · MindLiemandt describes that the key to developing a child is that they struggle and fail on their road to success supported by a caring adult, and that you just want to run the struggle-fail-support loop repeatedly. He notes parents believe kids learn valuable life skills in after-school sports through struggle, failure, and a supportive coach, and that parents pick a coach because they know he cares about their kid achieving and will support them. He argues schools should run the same struggle-fail-support loops that sports do for the other seven hours.
Andrew Huberman · How to Accelerate Learning & Improve Education | Joe Liemandt
Creator perspective · MindLiemandt contrasts sports with classrooms: a football team where every player got identical instruction would fail because the quarterback needs something different than the wide receiver, yet a teacher with 25-30 students must give everyone the exact same (median) instruction. This is why society has universal access to education but not universal learning: kids need individualized instruction, right down to the true individual.
Andrew Huberman · How to Accelerate Learning & Improve Education | Joe Liemandt
Creator perspective · MindLiemandt argues that learning science (cognitive load theory, cognitive science, how learning works) is not a main part of the curriculum in schools of education, so average teachers don't know the science of learning. At his staff days, when he asked his hundreds of teachers how many learned direct instruction, only about 5% raised their hand, and none had read Project Follow Through during their master's programs. He says this knowledge gap needs to be fixed.
Andrew Huberman · How to Accelerate Learning & Improve Education | Joe Liemandt
Creator perspective · MindLiemandt argues that high standards lead to loving school and happiness, citing Dr. Yeager's book '10 to 25, The Science of Motivating Kids' which is about high standards and high support, and that Dr. Yeager created Alpha's guide training where all adults follow his high standards, high support mentor mindset program. He describes that most parents are either high standards/low support or low standards/high support, and both are bad, and that the real answer is to do both high standards and high support — as schools should, and as great coaches do, as shown in thousands of sports movies where a coach turns a last-place team into state champions.
Andrew Huberman · How to Accelerate Learning & Improve Education | Joe Liemandt
Creator perspective · MindLiemandt argues that what you know is the basis of what you can learn, so having a foundation and building on it is critical. If material is too hard for a learner, they will not learn it no matter how long they are given it. He uses himself and the MCAT as an example: he doesn't know the words on the page, and grit or growth mindset would not get him to pass; he would need to be rolled back to freshman bio and built up from there.
Andrew Huberman · How to Accelerate Learning & Improve Education | Joe Liemandt
Creator perspective · MindLiemandt argues that individualized one-on-one tutoring is the best educational model (as in the Plato, Socrates, Aristotle, Alexander the Great lineage), but it is not cost-effective, which is why societies settled on 30-to-1 ratios 100 years ago. Smaller ratios are better (20-to-1 beats 30-to-1, 10-to-1 beats 20-to-1, 5-to-1 beats 10-to-1, one-to-one is best), and the goal in education is to get to one-on-one. He notes class size is the number one thing parents measure schools on.
Andrew Huberman · How to Accelerate Learning & Improve Education | Joe Liemandt
Creator perspective · MoneyMichael Lee says Sequence Holdings has two problems: an engineering problem and a human engineering problem, and the human engineering problem is much more difficult. Post-investment at the bank, Sequence learned they need to be much more hyper-attuned and sensitive to how employees feel, because technological change draws a lot of anxiety about what it means for the organization. The key is to make people feel like bringing AI into the company elevates what they do at their job, exercises the best parts of being human, makes the job more fun, and takes rote repetitive work out of their workflow. Sequence also aims to make employees feel like their organization is winning, and this playbook will continue to be refined at Baldwin after learning a lot at the bank. He expects the human
creator's learned heuristic
No Priors: Artificial Intelligence | Technology | Startups · Re-Founding Incumbents for the AI Era with Sequence Holdings Co-Founder and CEO Michael Lee
Creator perspective · MoneySequence Holdings views the current AI moment as the equivalent of the next industrial revolution. Michael Lee argues that when machines can work 24/7 and scale with electricity — accomplishing what no individual human or group of humans can do — the right answer is to reorganize what an organization should be to meet the opportunity of available technology. He sees the current push by software companies to give small machines to every human in the human assembly line as better than nothing but insufficient.
creator's interpretation and endorsement
No Priors: Artificial Intelligence | Technology | Startups · Re-Founding Incumbents for the AI Era with Sequence Holdings Co-Founder and CEO Michael Lee
Report · MoneyThe acceptance rate is the number of tokens accepted by the target model over the total tokens generated by the draft model, and a high acceptance rate is the key thing to note in a successful speculative decoding run.
AI Engineer · Is Speculative Decoding Worth It? Profiling vLLM on NVIDIA Blackwell — Akamai
Report · MoneySpeculative decoding uses a smaller draft model to speculate on a few tokens, which is faster and cheaper because the model is smaller. The target model then verifies the draft tokens in one forward pass, approving or rejecting them; for all rejected tokens the target model recomputes the token itself. Because the target model still verifies everything, output quality stays the same. The draft model typically generates between three and five tokens per cycle.
AI Engineer · Is Speculative Decoding Worth It? Profiling vLLM on NVIDIA Blackwell — Akamai
Creator perspective · MoneyIf you provide large enough context, models start to over-explore and come up with hypotheses that don't even exist or might not get you to the right answer. If you supply very limited context, models under-explore and won't have visibility into other pathways or potential hypotheses. For production incidents, telemetry is literally infinite — you can create as many log lines and metrics as you want — which makes the precise amount of context the real problem.
AI Engineer · The 6 Pillars of an Agentic Harness for Production — Varun Krovvidi, Resolve AI
Creator perspective · MoneyAI might determine the cleanest possible fix is deletion of a particular code snippet or database — you can't fault it for it, you just have to build guardrails around the system. Build guardrails so the AI works in the right fashion: determine the least privileged access AI can operate with in any given scenario.
AI Engineer · The 6 Pillars of an Agentic Harness for Production — Varun Krovvidi, Resolve AI
Creator perspective · MoneyFor production incidents, the only thing you're looking for is a causal chain of evidence, like a detective — the steps that took place that led to the incident. When solving a hard problem like production reasoning, you need causal reasoning built into the system, not just coherent answers.
AI Engineer · The 6 Pillars of an Agentic Harness for Production — Varun Krovvidi, Resolve AI