Library
12 sources captured.
I’ve always struggled to keep facts straight after a busy week, and this tiny paragraph finally clicked for me. It explains why cramming never sticks and why pulling an answer from memory feels like a mental workout that actually builds strength. Knowing that each review should land just before I would forget gives me a concrete schedule instead of vague guesswork. The idea that the forgetting curve flattens after every successful recall turns review into a progressive ladder rather than a repetitive chore. Now I can trust spaced repetition to turn fleeting knowledge into lasting competence for my career.
spaced repetitionforgetting curveretrieval practicereview interval schedulingmemory strengthening6 cardsWorld Models
processedhttps://worldmodels.github.io/
I love the idea that an AI can essentially learn by dreaming. Instead of constantly interacting with a heavy, expensive real-world simulator, the agent builds its own internal, compressed mental model of reality. By separating the heavy lifting—understanding space and time—into a massive world model and keeping the decision-maker tiny and simple, it solves the massive headache of credit assignment. The insight about 'temperature' is brilliant too; if the dream is too perfect, the agent just learns to exploit glitches. You have to make the dream a bit messy to make the agent truly robust.
World ModelsReinforcement LearningVariational AutoencoderLatent SpaceMixture Density Network+1 more7 cardsI found this chapter on descriptive statistics really insightful because it laid out the foundational concepts for summarizing and understanding data, which is crucial in business and quantitative analysis. The chapter's explanations of measures of central tendency and variability, as well as data visualization techniques, gave me a solid grasp of how to effectively describe and communicate data insights. This knowledge will help me to better analyze and interpret data in my own work, making more informed decisions. Descriptive statistics is a fundamental skill that I realize I need to master.
Descriptive StatisticsMeasures of Central TendencyMeasures of VariabilityData VisualizationMean+2 more6 cardssample IEP document
processedI cared about this IEP document because it shows how a customized education plan can help a student like Samuel Johnson, who struggles with reading, to catch up with his peers. The document outlines specific goals, accommodations, and services that will support Samuel in overcoming his challenges. It's impressive to see how the IEP team, including teachers, a parent, and a school psychologist, come together to create a comprehensive plan that addresses Samuel's unique needs. The document also highlights the importance of progress monitoring and regular review to ensure the plan's effectiveness.
Individualized Education Program (IEP)Specific Learning DisabilityAccommodations and ModificationsProgress MonitoringIEP Team+1 more12 cardsI found it interesting that spaced repetition works not during the reviews themselves, but during the gaps between reviews, when memories consolidate. This makes sense because it explains why reviewing at the right moments can significantly improve long-term retention while reducing study time. It's a powerful strategy for efficient learning.
Spaced repetitionSpacing effectForgetting curveMemory consolidation4 cardsAgents are Loops
processedI was surprised to learn that agents, like those used in coding tools, are simply LLMs (large language models) wrapped in a loop that allows them to execute tools and iterate on tasks. This architecture, which includes an LLM, tools, and a loop, enables agents to perform complex tasks by making decisions based on previous results. The simplicity of this core architecture belies the complexity of production agents, which require careful engineering, safety guardrails, and context management to function reliably.
LLMAgentToolLoopContext Management+1 more3 cardsI find the Pareto principle really interesting because it highlights how unevenly distributed effects and causes can be. In product development, this means that a small number of features can have a disproportionately large impact. This idea resonates with me because I've seen it play out in various projects where a few key features drive the majority of user engagement. Understanding this principle can help teams focus on what really matters and avoid wasting resources on less impactful work.
Pareto principlePrioritizationProduct development80/20 ruleEngineering teams3 cardsI keep running into situations where I or someone on my team wants to rip out some old process, delete legacy code, or scrap a rule that looks pointless. Every time we do it without asking why it existed, something breaks in a subtle and embarrassing way. Chesterton's Fence gives me a memorable name for the discipline I already knew I should practice but kept skipping when I was in a hurry. Having a crisp label makes it easier to invoke in a meeting without sounding obstructive — I can just say the name and people get it.
Chesterton's FenceSecond-order thinkingUnintended consequencesReform and change management3 cardsI keep hearing that I should invest early, but I never really felt the weight of it until I saw the doubling logic spelled out. A 7% annual return sounds modest — almost boring — yet it quietly doubles my money every decade without me doing anything clever. That realization flips my whole frame: the enemy isn't low returns, it's waiting. Every year I delay is a doubling cycle I'm handing back. I want to remember this not as a financial tip but as a principle about how exponential curves work, because it applies anywhere patience and consistency can compound over time.
compound interestexponential growthtime in marketconsistency over optimization3 cardsBeating the Averages
failedSelf-attention feels like one of those rare conceptual unlocks that changed everything downstream. I keep encountering Transformers as the backbone of basically every serious model I use or read about, and I realized I'd been hand-waving past the actual mechanism. Understanding why recurrence was the bottleneck — and how replacing it with scaled dot-product attention suddenly made parallel training tractable — gives me a cleaner mental model for why scaling worked at all. This is foundational context I want to carry forward whenever I'm reasoning about model architecture, training costs, or why the current generation of frontier models looks the way it does.
self-attentionTransformer architecturescaled dot-product attentionparallelizable trainingrecurrence elimination3 cardsI keep getting tempted to build clever, structured solutions — systems where I encode what I think I know about a problem domain. The Bitter Lesson is a direct warning against that instinct. Every time researchers did exactly what I'd be inclined to do — carefully crafting features, rules, and representations — they eventually got beaten by something dumber that just had more compute. I want to remember this so I stop over-investing in architectures that flatter my own understanding and start asking whether the approach I'm choosing actually scales with data and compute, or whether I'm just making myself feel smart.
Bitter LessonCompute scalingHand-engineered featuresLearning-based methods3 cards