<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Olumide Ogundepo on Medium]]></title>
        <description><![CDATA[Stories by Olumide Ogundepo on Medium]]></description>
        <link>https://medium.com/@olumidebenjamin16?source=rss-364c71a1c1eb------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*eGI5U9jlwiDi1H5n</url>
            <title>Stories by Olumide Ogundepo on Medium</title>
            <link>https://medium.com/@olumidebenjamin16?source=rss-364c71a1c1eb------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 21:23:44 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@olumidebenjamin16/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[From Cramming to Synthesis — Intelligence in the Era of AI in Nigeria]]></title>
            <link>https://medium.com/@olumidebenjamin16/from-cramming-to-synthesis-intelligence-in-the-era-of-ai-in-nigeria-4bfbf51d126e?source=rss-364c71a1c1eb------2</link>
            <guid isPermaLink="false">https://medium.com/p/4bfbf51d126e</guid>
            <category><![CDATA[intelligence]]></category>
            <category><![CDATA[creativity]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[nigeria]]></category>
            <category><![CDATA[education-reform]]></category>
            <dc:creator><![CDATA[Olumide Ogundepo]]></dc:creator>
            <pubDate>Wed, 18 Mar 2026 11:14:49 GMT</pubDate>
            <atom:updated>2026-03-18T11:41:20.460Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Introduction</strong></p><p>Some months ago, <a href="https://www.linkedin.com/posts/olumide-ogundepo-76500a15b_zero-is-not-nothing-rethinking-education-activity-7382008677668392960-dcvf?utm_source=share&amp;utm_medium=member_desktop&amp;rcm=ACoAACZEN70BGz6r5_-ZasVWJMpgH1JHW4NNkXY">I wrote about rethinking zero</a>, how what’s potentially interpreted to be nothing is actually the structural foundation of progress such that when combined with one, gave birth to all amazing digital technology in the world today. Zero isn’t failure or absence; it is the enabling structure of our potential or computational world.</p><p>Today, I want to explore a revelation: our foundational misunderstanding of intelligence itself.</p><p>For several generations, we’ve measured intelligence by a student’s ability to memorize patterns, cram past questions, and replicate answers. However, the rise of artificial intelligence has exposed an uncomfortable truth such that what we call intelligence was a narrow optimization for stable environments. True intelligence that scales and adapts operate on a whole different principle. Just as zero seemed like nothing until we realized it enabled everything, cramming seemed like intelligence until AI showed us what learning actually means.</p><p><strong>The Cramming Paradigm</strong></p><p>Walk into any examination hall across institutions in Nigeria, or anywhere in the world and you will see the culmination of what we’ve over time considered intelligent behavior range from memorized hundreds of past questions, reproduced answers exactly as taught, optimized for one context (that is, passing specific course exam), and learned basically from their own several attempts and corrections.</p><p>We rewarded this. We called it being smart. Students who could detect patterns in past papers, predict possible questions, and retrieve memorized answers were considered top performers. The system worked with provision of high score, university admissions, and professional certificates.</p><p>But here is what we missed: this isn’t intelligence, it is narrow optimization.</p><p>Technically, it is what reinforcement learning researchers call on-policy learning, an agent (a student) that learns exclusively from their own experience, optimizes for a specific environment (specific course exam), and discards knowledge when contexts change. This works when the environment is structured or stable such that the curriculum doesn’t change much, evaluation is narrow (answer accordingly or fail), sample efficiency doesn’t matter (that is, there is lots of time to prepare for exams). Hence, the cramming paradigm was intelligent for a static world.</p><p>The AI Revelation: When Machine Exposed Our Blind Spot</p><p>The arrival of GPT, Claude, Grok, Gemini, and other large language models exposed what we defined as intelligence by passing exams they never take time to study for. Let’s consider these scenarios:</p><p>1. The crammed student: 10 hours memorizing 500 past questions, achieves 80% on the test, but struggles to apply knowledge in a novel approach.</p><p>2. The AI: It was trained on billions of different examples across several human knowledge domains, achieved 95% on tests it really never prepared for, achieved this by transferring understanding across infinite domains.</p><p>The AI never crammed. It never saw past questions or papers for tests, yet it outperformed humans with several hours of practice that we assume is intelligent.</p><p><strong>The Difference</strong></p><p>While students were engaging in on-policy learning (learning only from their own narrow experience). AI machines were doing something fundamentally different: off-policy learning at scale, absorbing patterns from every source imaginable, synthesizing across contexts, and building transferable understanding.</p><p>The revelation is not that AI is intelligent and students are not. The revelation is that humans have been optimized for the wrong kind of intelligence.</p><p><strong>True Intelligence: The Off-Policy Paradigm</strong></p><p>Permit me to share a technical insight from reinforcement learning that reflect everything:</p><p>On-policy approach: It learns from an agent’s own experience. They are stable, predictable, reliable, and work well when there are infinite samples and stable environments. But they are sample-inefficient and don’t transfer learned experience well.</p><p>Off-policy approach: It learns from any experience such as its own, others’, simulated and diverse sources. They are more complex, harder to stabilize, but dramatically more sample-efficient and better at generalization.</p><p>The greatest breakthrough of AI machines today is the combination of on-policy and off-policy with off-policy for learning everything while on-policy is fine-tuned for specific optimization. Literally, this is how GPT-4 works, pretrained on the breadth of human knowledge (that is, off-policy), then fine-tuned for specific behaviors (on-policy). This is how DeepMind’s AlphaGo worked, it learns from several human games patterns, then optimized and tailored through self-play (on-policy) for each opponent faced.</p><p>The implication for human intelligence is profound as true intelligence in the era of AI is not about memorizing more patterns but about learning from diverse sources thinking through off-policy that consists of experiences and mistakes of other people, cross-domain principles and patterns, historical knowledge and potential possibilities, as well as success and failure analysis from multiple contexts. Intelligence is not retrieval of stored answers but combination of patterns in a novel approach, transferring domain understanding, generation of solutions to unseen problems, and adaptation of principles to new contexts. Knowledge does not expire after the exam hence the need for building cumulative and connected understanding, updating belief with new evidence, learning from observation before taking action, development of mental models not answer banks. Additionally, the most valuable skill is not what you know but how quickly you can acquire new knowledge, how effectively you can synthesize across domains, how adaptable you are to changing contexts and how well you can transfer learning to proffer novel solutions.</p><p>Just as zero is not nothing but the structure that enables everything, intelligence is not the possession of knowledge but the capacity to acquire, synthesize, and apply it. To buttress my point, mathematics without zero is functional but fundamentally limited as you can count but you cannot compute. Intelligence without synthesis is the same, you can pass tests, but you cannot adapt.</p><p><strong>Why This Matter: The Real-World Crisis</strong></p><p>The cramming paradigm is failing in the modern world, and we’re seeing the consequences.</p><p><strong>The Nigeria Context</strong></p><p>The Nigeria students excel at WAEC and JAMB but struggle in university coursework that sometimes require application, some graduates can recite theory but cannot solve novel problems, some first-class students sometimes cannot function in a workplace that requires adaptation. Also, in a professional setting, careers require consistent reinvention, problems do not come with past questions, the synthesizing of cross-domain becomes the most valuable skill. However, there is a need to realize that machines now handles memorization and pattern recognition better than humans because over a long time we produced students optimized for a world that no longer exists, hence makes our definition of traditional intelligence obsolete in the presence of AI, hence human value now lies in the ability to synthesize, become creative through curiosity, therefore those that only cram are being discarded with automation while those that can synthesize and adapt become irreplaceable.</p><p><strong>Recommendation</strong></p><p><strong>Academics and Educators</strong></p><p>The proposed new study approach should be informed from a review of national curriculum such that the answers that will be inferred from students’ answers to exam questions can be any of the responses from broad multiple sources of analysis and exploration across domains not necessarily from what was taught in classes, this value novel problem-solving over answer reproduction. Hence, assessment will be redesigned to stop rewarding pattern recognition and start rewarding synthesis and knowledge transfer as such understanding is being evaluated and not memorization or narrow mastery, therefore learning capacity is being assessed over knowledge possession. Furthermore, off-policy learning should be integrated into the teaching culture such as creating case studies from multiple sources, exposure to diverse perspectives, learning from historical failures, developing real-world and complex problems.</p><p><strong>Institutions and Policy-Makers</strong></p><p>The hard truth is that our education system is optimized for an on-policy approach that births narrow intelligence. Therefore, we need structural reform that redefines success metrics through adaptation, synthesis ability, and learning ability with speed, not just exam scores. The need to invest in novel ways of how to learn, not just what to learn. Development of cross-disciplinary programs with project-based learning, and real-world problems. Rather than competing with AI in memorization, let us craft our path through synthesis, creativity born through curiosity and leave memorization for AI. Let us continuously optimize education for lifelong adaptation such that it does not end.</p><p><strong>Final Note</strong></p><p>Traditional academic intelligence (cramming) is to true intelligence what counting without zero is to mathematics, it is functional but fundamentally limited. In the Era of AI, intelligence is not what you have memorized, it is your capacity to synthesize from all available knowledge, that is your own and others to generate novel understanding and adapt to changing contexts.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4bfbf51d126e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Reinforcement Learning with Tic-Tac-Toe Illustration]]></title>
            <link>https://medium.com/@olumidebenjamin16/reinforcement-learning-with-tic-tac-toe-illustration-8f91518189d6?source=rss-364c71a1c1eb------2</link>
            <guid isPermaLink="false">https://medium.com/p/8f91518189d6</guid>
            <category><![CDATA[fundamentals]]></category>
            <category><![CDATA[robotics]]></category>
            <category><![CDATA[reinforcement-learning]]></category>
            <category><![CDATA[lrs]]></category>
            <dc:creator><![CDATA[Olumide Ogundepo]]></dc:creator>
            <pubDate>Thu, 09 Oct 2025 01:51:09 GMT</pubDate>
            <atom:updated>2025-10-09T01:51:09.172Z</atom:updated>
            <content:encoded><![CDATA[<p>As a kid you might be familiar with the Tic-Tac-Toe, a pencil and paper sketched game being played between two players (with X and O assigned to each players) that sequentially take turns and the winner is declared when the Xs or Os marks are uniformly represented in a diagonal or straight line, in the absence of such representation the game continues. Hence, this game will be used to illustrate the concepts of Reinforcement Learning contained in Introduction to Reinforcement Learning book by Richard Sutton and Andrew Barto, such that it could be relatable and solidify the needed understanding to develop more complex reinforcement learning driven systems.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/276/1*2QDoywOJc6ulvRFCXRDBKw.png" /><figcaption>Tic-Tac-Toe board illustration</figcaption></figure><p>In the process of playing the game each of the player (say player X) continues to position in such a manner that it put the opponent in an uncomfortable position or attains any of the winning position in the figure above while simultaneously wish that (player O) did not block such winning positioning. These players keep learning on how to win while playing the game, monitor and capitalize on each other’s oversight or weakness to win. In Reinforcement Learning terms, these players are referred to as agent (A), initial state (S) as the board game state without any players move, reward (r) as steps taken at each point to win (+1) or loss (-1), return (G) as the cumulative reward at the end of each game either a win, draw or lose, policy (π) as the strategy adopted by each player to win the game, while the agent, states and rules of play is referred to as the environment.</p><p>The players (agents) learn from series of interaction on the board with each other in order to win the game, the reward and feedback for each game is delayed until after the game is won or lost, also it is difficult to evaluate a rewarding steps that will lead to a win in a game because of several iterations and learning that will be involved. Hence, a machine agent could use (value estimate, more on it later) to evaluate the steps that would lead to optimal result (win) through breaking down of all potential steps into subproblems and select the steps that will lead to a win. The agent also engage in exploration and exploitation methods, strategies that helps to try out new steps to achieve higher rewards (exploration) or use of existing playing knowledge to maximize future rewards (exploitation). However, there is a need for the agent to trade-off between exploration and exploitation to enhance agents learning and improve decision making.</p><p><strong>Decision Making Without Knowing The Outcome</strong></p><p>In the process of making an optimal decision that maximizes win or a situation where the agent need to decide between three possible moves (left, right, or down), first, the agent randomly explore the present state and possible actions then develop a technique called <em>value estimates </em>for each action in a given state. The Q-table is then used to store the value estimate. It is represented as:</p><p>Q(<em>s, a</em>): It is the expected reward from taking a specific action<em> a (either right, left, or down) </em>in a given state <em>s</em>. The agent keeps updating the Q-values incrementally, using the equation:</p><pre>Q(s,a)=Q(s,a)+α[r-Q(s,a)]</pre><p>Such that:</p><p>a is the action</p><p>s is the state</p><p>α is the learning rate</p><p>r is the reward after a given action has been taken.</p><p>In order to decide the move to take, the agent mostly adopts ɛ-greedy policy, such that;</p><p>With probability ɛ, the agent select a random move (exploration)</p><p>With probability 1-ɛ, the agent select the best-known move (exploitation).</p><p>Conclusively, this agent starts from being a novice — making mistakes through random lose, wins or draws. However, over a period of time, it learns from trial and error, develop a policy that performs better and better. This is the fundamental concept of reinforcement learning.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8f91518189d6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Introduction to Reinforcement Learning in R]]></title>
            <link>https://medium.com/@olumidebenjamin16/introduction-to-reinforcement-learning-in-r-41f14041b68?source=rss-364c71a1c1eb------2</link>
            <guid isPermaLink="false">https://medium.com/p/41f14041b68</guid>
            <category><![CDATA[r]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[robotics]]></category>
            <dc:creator><![CDATA[Olumide Ogundepo]]></dc:creator>
            <pubDate>Fri, 13 Dec 2019 05:03:03 GMT</pubDate>
            <atom:updated>2019-12-13T05:03:03.073Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Motivation for the study</strong></p><p>Tremendous increase in the application of the robotics intelligence, and consolidating of the daily activities performed by humans with automation and transactions motivated the development of this article. Reinforcement learning is an aspect of machine learning that is made up of an Agent<em>, </em>interacting with an environment, and performing some specific actions with reward in respect to prior actions. R programming language will be used to develop systems consider in this learning.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/582/1*c2LGkU4VqdJR40VsdBfH5Q.png" /><figcaption>Figure 1: The agent-environment interaction in a Markov decision process</figcaption></figure><p>Over time, there has been several applications of this technique in fascinating field of study such as:</p><p>i. Bidding and Advertising</p><p>ii. Games</p><p>iii. Traffic light control</p><p>iv. Robotics</p><p>v. Chemistry</p><p>vi. Resources management in computer clusters</p><p>vii. Healthcare</p><p>viii. Finance. etc.</p><p>In this article, our focus will be on how reinforcement learning makes decisions in solving some specific problems from both stationary and non-stationary perspectives. Sit back and enjoy!!!.</p><p>One distinguishing factor between reinforcement learning and other machine learning technique is the use of evaluative feedback instead of classical instructive feedback used by other learning. Evaluative feedback in RL is an indication of how good the action was, but not considering if the taken action was the worst or best action possible. Unlike instructive feedback that signals the correct action irrespective of previous action taken. And this is the cause for pattern classification, artificial neural network, etc. Distinctively, evaluative feedback is dependent on the previous actions taken while instructive feedback is independent of the previous action taken.</p><p>In this article we focus on evaluative part of reinforcement learning, studying this aspect enables us to understand how important it is to RL. And how it can be integrated with instructive feedback to design a fast and intelligent system.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/533/1*ZW_GtvRx6Zr9Ed4GSFMguA.png" /><figcaption>Figure 2: Rabbit-Tiger Environment</figcaption></figure><p>For instance, imagine the rabbit-tiger environment in figure 1 above, it consists of a rabbit in an environment where pellets and carrot are scattered on the grid, and a tiger waiting in a corner for the rabbit to come closer so as to prey on it. In addition, the (rabbit) aim is to maximize expected reward over some period of time; for instance, 50 (steps) actions. So, when the rabbit eats pellets, its reward is given as +5, and when it eats carrot as +10, when eaten by the tiger as death but when it eventually eats the pellets and fruits (still maintain the step) without being eaten by the tiger it moves to the next phase (regarded as success). Afterwards, for each selected action, the agent receives a numerical reward. The RL in this context can be modeled as:</p><p>1. The agent (rabbit) is present in a state in the environment, and based on the entry state say <em>S0,</em> the agent takes an action <em>A0.</em></p><p>2. There is an environment transition to a new state <em>S1 </em>based on the selected action <em>A0,</em> and it takes another action <em>A1.</em></p><p>3. As the agent move closer to eat pellets (+5), and moves closer to eating carrot (+10), then the environment continue to transit and position the rabbit in a difficult position to finish the phase or being eaten by the tiger (-100) where the episode ends (rabbit lifespan cut short); and it reduces its chances of survival and ends the phase, as displayed in figure 2.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/624/1*8tY_WhrerQGJMzATVokxRw.png" /><figcaption>Figure 3: Rabbit-Tiger Environment</figcaption></figure><p>In the rabbit-tiger context, each action has an expected (mean) reward given that the action is selected. In accordance with this study, we refer to it as <strong><em>value</em></strong><em> </em>of that action. Let’s represent the selected action at time step<em> t</em> as <em>At, </em>and the respective reward as <em>Rt.</em> Then, the value of an arbitrary action <em>a, </em>denoted as <em>q*(a),</em> is the mean reward given that <em>a</em> is chosen:</p><p>Assuming we know the value of each action, thereafter, we would always select the action with the highest value. This is referred to as <em>greedy action</em>. When any of these actions are selected, we say that the current knowledge of the values is being exploited. Otherwise, non-greedy actions are selected, and referred to as explore because it helps to improve the estimate of non-greedy actions value. Exploitation is better when considering the maximization of the mean reward on the next step, but exploration may produce the best total reward in the future. For example, in the rabbit-tiger context, assuming the nearest pellet to the left of the grid possesses a greedy action’s value, while several other actions (pellet to the right positions the rabbit in a difficult position to survive, carrot almost near tiger in front, etc.) are estimated to be almost good but surrounded with uncertainty.</p><p>One of the uncertainties (carrot almost near tiger in front) actions might be actually better than the greedy action but the agent is unaware. If there is enough time steps to still select actions, then it will be better to explore the non-greedy actions, and detect which is better than the greedy action. It should be noted that reward is comparably lower in the next step (short run) during exploration, but much higher in the next steps (long run) because after the discovery of the better actions, they can be exploited many times. Either to explore or exploit depends on these three factors; uncertainties, estimated values, and number of steps left. Because it is quite impossible to explore and exploit a selected single action. In that case, there will be a conflict in exploration and exploitation. There are many techniques to strike a balance between exploration and exploitation, which will be extensively discussed in our next article.</p><p>Hope this article has been of help to you in anyway; look out for the continuation in our next series.</p><p>It is good to note that, understanding the framework is more important and necessary before diving into code writing.</p><p>My name is Ogundepo Olumide Benjamin, thanks for reading. Hope, to see you in the next episode.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=41f14041b68" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>