<?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 Moudather Chelbi on Medium]]></title>
        <description><![CDATA[Stories by Moudather Chelbi on Medium]]></description>
        <link>https://medium.com/@moudather.chelbi?source=rss-ddbda4ec986------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*zjaX0EIsh_1J7H1SaOVAFg.jpeg</url>
            <title>Stories by Moudather Chelbi on Medium</title>
            <link>https://medium.com/@moudather.chelbi?source=rss-ddbda4ec986------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 00:42:43 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@moudather.chelbi/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[SCSI: Reimagining Global Data Analysis for Environmental Science]]></title>
            <link>https://medium.com/@moudather.chelbi/scsi-reimagining-global-data-analysis-for-environmental-science-ef1dfb3ad63f?source=rss-ddbda4ec986------2</link>
            <guid isPermaLink="false">https://medium.com/p/ef1dfb3ad63f</guid>
            <dc:creator><![CDATA[Moudather Chelbi]]></dc:creator>
            <pubDate>Fri, 27 Dec 2024 17:01:43 GMT</pubDate>
            <atom:updated>2024-12-27T17:01:43.394Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*91vRjWps2GBJ8sxWNxJLTQ.jpeg" /></figure><p>In an era where understanding our planet’s complex systems is more crucial than ever, we need powerful tools that can handle global-scale data analysis with precision and efficiency. The Satellite Constellation Spatial Index (SCSI) project emerged from this need, offering a fresh perspective on how we process and analyze Earth’s environmental data.</p><p><strong>The Vision</strong></p><p>Imagine being able to track the intricate dance of global weather patterns, monitor the health of our oceans, or predict the spread of atmospheric phenomena with unprecedented accuracy. This was the driving vision behind SCSI — to create a framework that could handle these complex analyses while maintaining scientific accuracy and computational efficiency.</p><p><strong>Inspiration from Nature</strong></p><p>The project draws inspiration from an unexpected source: the geometric patterns found in nature. Just as a honeycomb provides the most efficient way to pack cells in a plane, SCSI uses a special geometric pattern based on the icosahedron (a twenty-sided polyhedron) to divide Earth’s surface into nearly uniform regions. This approach mirrors natural systems’ tendency to find optimal solutions for complex problems.</p><p><strong>Beyond Traditional Methods</strong></p><p>Traditional approaches to global data analysis often struggle with a fundamental problem: the Earth is round, but we typically analyze it using flat maps. This leads to distortions, particularly near the poles — think of how Greenland appears massive on many world maps. SCSI tackles this challenge head-on by working directly on the spherical surface, maintaining accuracy regardless of latitude.</p><p><strong>Real-World Applications</strong></p><p>The potential applications of SCSI span numerous fields:</p><p><strong>Climate Science</strong></p><p>- Modeling global temperature distributions</p><p>- Tracking atmospheric circulation patterns</p><p>- Analyzing ocean current systems</p><p>- Studying climate change impacts</p><p><strong>Environmental Monitoring</strong></p><p>- Tracking pollution dispersion</p><p>- Monitoring forest coverage changes</p><p>- Analyzing biodiversity distributions</p><p>- Studying ocean acidification patterns</p><p><strong>Disaster Response</strong></p><p>- Predicting weather pattern impacts</p><p>- Tracking wildfire spread</p><p>- Monitoring volcanic ash dispersion</p><p>- Analyzing tsunami wave propagation</p><p><strong>Satellite Operations</strong></p><p>- Optimizing constellation coverage</p><p>- Planning Earth observation missions</p><p>- Managing communication networks</p><p>- Analyzing global sensor networks</p><p><strong>The Power of Integration</strong></p><p>What makes SCSI particularly powerful is its ability to integrate multiple types of data and analyses. It can simultaneously track temperature, pressure, humidity, and other variables, understanding how they interact and influence each other. This holistic approach mirrors the interconnected nature of Earth’s systems.</p><p><strong>Looking to the Future</strong></p><p>The potential for SCSI extends far beyond its current capabilities. Future developments could include:</p><p><strong>Enhanced Analysis</strong></p><p>- Machine learning integration for pattern recognition</p><p>- Real-time data processing capabilities</p><p>- Advanced predictive modeling</p><p>- Integration with satellite data streams</p><p><strong>Expanded Applications</strong></p><p>- Urban planning and development</p><p>- Renewable energy optimization</p><p>- Agricultural planning and monitoring</p><p>- Marine resource management</p><p><strong>Technical Advancements</strong></p><p>- Cloud-based processing capabilities</p><p>- Enhanced visualization tools</p><p>- Mobile platform support</p><p>- API integrations</p><p><strong>Impact on Scientific Research</strong></p><p>SCSI represents more than just a technical solution — it’s a new way of thinking about global data analysis. By providing researchers with tools that match the complexity of the systems they study, we enable deeper insights and more accurate predictions.</p><p><strong>Community and Collaboration</strong></p><p>The project is designed with collaboration in mind. Scientists, researchers, and developers can build upon the framework, adding new capabilities and adapting it to their specific needs. This open approach ensures that SCSI can evolve alongside our understanding of Earth’s systems.</p><p><strong>Conclusion</strong></p><p>In a world where understanding global patterns and interactions is increasingly crucial, SCSI provides a powerful framework for analysis and insight. Whether tracking climate change, optimizing satellite operations, or studying environmental phenomena, it offers the tools needed to work with global data effectively.</p><p>The future of environmental science lies in our ability to process and understand complex, interrelated systems on a global scale. SCSI takes us one step closer to this goal, providing a foundation for the next generation of Earth science research and applications.</p><p>As we face growing environmental challenges, tools like SCSI become increasingly important. They help us understand our planet better, make more informed decisions, and work towards a sustainable future. The journey has just begun, and the possibilities are as vast as the globe itself.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ef1dfb3ad63f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Jailbreaking Large Language Models: A Growing Threat]]></title>
            <link>https://medium.com/@moudather.chelbi/jailbreaking-large-language-models-a-growing-threat-eb7db55bf2f8?source=rss-ddbda4ec986------2</link>
            <guid isPermaLink="false">https://medium.com/p/eb7db55bf2f8</guid>
            <category><![CDATA[jailbreak]]></category>
            <category><![CDATA[llm]]></category>
            <dc:creator><![CDATA[Moudather Chelbi]]></dc:creator>
            <pubDate>Wed, 23 Oct 2024 13:55:37 GMT</pubDate>
            <atom:updated>2024-12-17T17:01:30.150Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*8vndKfoOzKsFBann" /><figcaption>Photo by <a href="https://unsplash.com/@growtika?utm_source=medium&amp;utm_medium=referral">Growtika</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Large language models (LLMs) are a powerful technology capable of amazing feats like writing code and generating creative text. However, the same capabilities that make them so impressive also leave them open to exploitation through “jailbreaking.” Jailbreaking is a type of adversarial attack designed to bypass an LLM’s safety mechanisms and force it to produce harmful or inappropriate content.</p><h3>How Jailbreaking Works</h3><p>LLMs learn to generate text by identifying patterns in massive datasets. Jailbreaking takes advantage of this by feeding the model carefully crafted prompts designed to trigger the generation of harmful or unethical content.</p><p>These attacks can take many forms, including:</p><p>● <strong>Adversarial prompting:</strong> Creating prompts that exploit the LLM’s training data or its ability to understand instructions.</p><p>● <strong>In-context learning attacks:</strong> Providing the model with harmful examples during a learning task, causing it to incorporate the malicious information into its responses.</p><p>● <strong>Multi-turn prompting:</strong> Engaging the model in a conversation that gradually escalates until it produces the desired harmful output.</p><h4>A Multifaceted Threat</h4><p>Jailbreaking can also target the model’s internal architecture or exploit vulnerabilities in its training process. This includes attacks such as:</p><p>● <strong>Backdoor attacks:</strong> Introducing malicious data or code into the model during training, creating a “backdoor” that can be triggered by specific inputs.</p><p>● <strong>Model interrogation:</strong> Analyzing the model’s internal representations to extract sensitive information or manipulate its decoding process.</p><p>● <strong>Activation steering:</strong> Directly manipulating the model’s internal activations to alter its behavior without changing its code or training data.</p><p>The rise of multimodal LLMs, which can process both text and images, has led to new types of attacks that exploit interactions between these modalities.</p><blockquote>Additionally, attackers are exploiting the fact that most LLMs are primarily trained on English data, leaving them more vulnerable to jailbreaking in other languages.</blockquote><h3>The Need for Robust Defenses</h3><p>A range of defense mechanisms have been developed to counter jailbreaking, including:</p><p>● <strong>Prompt filtering and transformation:</strong> Analyzing and manipulating input prompts to detect and neutralize malicious content.</p><p>● <strong>Adversarial training and safety fine-tuning:</strong> Training the model on datasets containing both benign and adversarial examples to improve its resistance to attacks.</p><p>● <strong>Multi-agent defenses:</strong> Using multiple LLMs with different roles and perspectives to analyze and filter out harmful responses.</p><p>Despite these efforts, jailbreaking remains a significant threat. New vulnerabilities and attack strategies are constantly emerging, requiring ongoing research and development of robust defenses.</p><h3>The Future of LLM Security</h3><p>Ensuring the safe and responsible use of LLMs requires a multifaceted approach that goes beyond simply scaling models or implementing superficial safety measures.</p><h4>Researchers must focus on developing:</h4><p>● <strong>Robust alignment techniques:</strong> Teaching LLMs to understand context and intent rather than relying on memorizing patterns.</p><p>● <strong>Comprehensive defense strategies:</strong> Combining prompt-level, model-level, and multi-agent defenses to protect against a wide range of attacks.</p><p>● <strong>Automatic detection and mitigation mechanisms:</strong> Leveraging machine learning to identify and neutralize jailbreak attempts.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=eb7db55bf2f8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Navigating the Complexities of Evaluating Retrieval-Augmented Generation Systems]]></title>
            <link>https://medium.com/@moudather.chelbi/navigating-the-complexities-of-evaluating-retrieval-augmented-generation-systems-e7dc6b9274a8?source=rss-ddbda4ec986------2</link>
            <guid isPermaLink="false">https://medium.com/p/e7dc6b9274a8</guid>
            <category><![CDATA[retrieval-augmented-gen]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[llm-evaluation]]></category>
            <dc:creator><![CDATA[Moudather Chelbi]]></dc:creator>
            <pubDate>Tue, 22 Oct 2024 15:06:12 GMT</pubDate>
            <atom:updated>2024-10-22T15:18:21.794Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*eHZN3Rm8fMrECfp9" /><figcaption>Photo by <a href="https://unsplash.com/@priscilladupreez?utm_source=medium&amp;utm_medium=referral">Priscilla Du Preez 🇨🇦</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Retrieval-Augmented Generation (RAG) systems, which cleverly combine information retrieval with generative language models, are rapidly gaining traction in the field of natural language processing. But evaluating these sophisticated systems is a challenging endeavor, requiring a nuanced understanding of their components and the dynamic interplay between them. This blog post will explore the intricacies of evaluating RAG systems, examining the challenges, targets, and metrics involved, drawing insights from a recent research survey on the subject.</p><h3>The Challenges of Evaluating RAG Systems</h3><p>Evaluating RAG systems isn’t as simple as assessing individual components in isolation. Several inherent challenges complicate the evaluation process.</p><p>● <strong>Dynamic and Diverse Knowledge Sources:</strong> One of the primary challenges stems from the vast and ever-changing nature of the knowledge sources used by RAG systems. These sources can range from structured databases to the entirety of the web, making it difficult to measure the precision, recall, and relevance of the retrieved information. The temporal aspect of information adds another layer of complexity, as the accuracy and relevance of data can shift over time.</p><p>● <strong>Subjectivity in Generation:</strong> Evaluating the quality of the text generated by RAG systems can be inherently subjective, especially for tasks that involve creativity or open-ended responses. What constitutes a “good” response can vary based on human preferences and the specific context of the query.</p><p>● <strong>Interplay Between Retrieval and Generation:</strong> The performance of a RAG system hinges on the synergy between its retrieval and generation components. Evaluating each component separately doesn’t provide a complete picture of the system’s overall effectiveness. The system’s ability to leverage retrieved information to generate high-quality responses is a crucial aspect of evaluation.</p><h3>A Unified Evaluation Process: AUEPORA</h3><p>To address these complexities, <a href="https://arxiv.org/pdf/2405.07437">https://arxiv.org/pdf/2405.07437</a> introduced a structured analytical framework called <strong>A Unified Evaluation Process of RAG (AUEPORA)</strong>. AUEPORA guides the understanding and analysis of RAG benchmarks by focusing on three key questions:</p><p>1.<strong>What to Evaluate?</strong> (Target)</p><p>2.<strong>How to Evaluate?</strong> (Dataset)</p><p>3.<strong>How to Measure?</strong> (Metric)</p><p>This framework emphasizes a systematic approach to evaluating RAG systems, considering various factors and potential output/ground truth pairings.</p><h3>Evaluation Targets: What Aspects to Focus On?</h3><p>AUEPORA helps break down the evaluation process into specific targets, focusing on both the <strong>retrieval</strong> and <strong>generation</strong> components, as well as additional requirements. Let’s explore these targets in more detail:</p><h4>Retrieval Targets</h4><p>● <strong>Relevance:</strong> This target evaluates how well the retrieved documents align with the user’s query. Essentially, it measures the precision and specificity of the retrieval process, ensuring that the retrieved information actually addresses the user’s information needs.</p><p>● <strong>Accuracy:</strong> This target assesses how accurately the retrieval component identifies relevant documents from a pool of candidates. It focuses on the system’s ability to distinguish between relevant and irrelevant documents and rank them accordingly.</p><h4>Generation Targets</h4><p>● <strong>Relevance:</strong> Similar to retrieval relevance, this target measures how well the generated response aligns with the user’s initial query. It ensures that the response stays on topic and addresses the specific requirements of the query.</p><p>● <strong>Faithfulness:</strong> This target evaluates whether the generated response accurately reflects the information present in the retrieved documents. It measures the consistency and grounding of the response, making sure the generated content is supported by the retrieved evidence.</p><p>● <strong>Correctness:</strong> This target assesses the factual accuracy and appropriateness of the generated response in relation to the query. It ensures that the response provides correct information and aligns with the expected context.</p><h4>Additional Requirements</h4><p>Beyond the core retrieval and generation targets, AUEPORA highlights several additional requirements crucial for real-world RAG applications:</p><p>● <strong>Latency:</strong> This requirement focuses on the system’s responsiveness — how quickly it can retrieve information and generate a response. Low latency is essential for a positive user experience, particularly in interactive applications.</p><p>● <strong>Diversity:</strong> This requirement assesses the variety of information presented in both the retrieved documents and the generated response. A diverse system avoids redundancy and presents a broader range of perspectives on the query topic.</p><p>● <strong>Noise Robustness:</strong> Real-world data is often messy and contains irrelevant or misleading information. This requirement evaluates how well a RAG system can handle such noisy data without compromising the quality of its output.</p><p>● <strong>Negative Rejection:</strong> This requirement assesses the system’s ability to recognize when the available information is insufficient to provide a confident response. Instead of generating potentially inaccurate or misleading information, the system should ideally refrain from answering in such cases.</p><p>● <strong>Counterfactual Robustness:</strong> This requirement measures how well a RAG system can identify and disregard incorrect information, even when presented with compelling but false claims.</p><h3>Datasets for Evaluating RAG Systems</h3><p>The selection and creation of datasets play a vital role in evaluating RAG systems. Datasets tailored for specific metrics or tasks can provide more accurate insights into the system’s capabilities and guide its development.</p><p>● <strong>Leveraging Existing Datasets:</strong> Some benchmarks utilize well-established datasets like those from KILT (Knowledge Intensive Language Tasks) or SuperGLUE. These datasets provide a baseline for comparison but might not always capture the complexities of real-world scenarios.</p><p>● <strong>Generating New Datasets:</strong> Thanks to the power of LLMs, researchers can now create datasets specifically designed to evaluate certain aspects of RAG systems. Benchmarks like RGB, MultiHop-RAG, and CRUD-RAG have generated their own datasets using online news articles to test how well systems handle dynamic, real-world information.</p><p>● <strong>Dynamically Updated Datasets:</strong> Some newer benchmarks, like DomainRAG, go a step further by using datasets that are updated regularly. This prevents LLMs from “cheating” by relying on memorized information and forces them to utilize the provided, up-to-date data.</p><h4>Evaluation Metrics: Quantifying Performance</h4><p>While identifying the targets of evaluation is important, having the right metrics to quantify performance is equally crucial. A wide range of metrics can be used to measure different aspects of RAG systems, catering to both retrieval and generation aspects.</p><h4>Retrieval Metrics</h4><p>Retrieval metrics focus on assessing the relevance, accuracy, diversity, and robustness of the information retrieved by the system. These metrics can be broadly categorized as:</p><p>● <strong>Non-Rank Based Metrics:</strong> These metrics evaluate whether a retrieved item is relevant without considering its position in a ranked list. Some common examples include:</p><p>○ <strong>Accuracy:</strong> The proportion of correct results (both true positives and true negatives) out of all cases examined.</p><p>○ <strong>Precision:</strong> The fraction of relevant instances among the retrieved instances.</p><p>○ <strong>Recall@k:</strong> The fraction of relevant instances retrieved out of all relevant cases, considering only the top <em>k</em> results.</p><p>● <strong>Rank-Based Metrics:</strong> These metrics take into account the order in which items are retrieved, placing higher importance on the top-ranked results. Examples include:</p><p>○ <strong>Mean Reciprocal Rank (MRR):</strong> The average of the reciprocal ranks of the first correct answer for a set of queries.</p><p>○ <strong>Mean Average Precision (MAP):</strong> The mean of the average precision scores for each query.</p><h4>Generation Metrics</h4><p>Generation metrics go beyond simply measuring accuracy and assess the quality of the generated text, taking into account aspects like coherence, relevance, fluency, and alignment with human judgment.</p><p>● <strong>Traditional Metrics:</strong> These metrics measure the overlap or similarity between generated text and a reference text. Some commonly used examples include:</p><p>○ <strong>ROUGE (Recall-Oriented Understudy for Gisting Evaluation):</strong> A set of metrics that evaluate summary quality by comparing them to human-generated summaries.</p><p>○ <strong>BLEU (Bilingual Evaluation Understudy):</strong> A metric originally designed for evaluating machine translation quality, which calculates the precision of n-grams in the generated text compared to a reference translation.</p><p>○ <strong>BertScore:</strong> This metric leverages contextual embeddings from pre-trained transformers like BERT to evaluate the semantic similarity between generated and reference text.</p><p>● <strong>LLM as a Judge:</strong> This emerging approach utilizes LLMs themselves to evaluate the quality of generated text. LLMs can be trained on human judgments to score text based on criteria like coherence, relevance, and fluency. This method has the potential to provide more nuanced assessments, but aligning LLM judgments with human preferences remains a challenge.</p><h3>Metrics for Additional Requirements</h3><p>Specific metrics are used to evaluate the additional requirements mentioned earlier. For instance, <strong>latency</strong> can be measured by the average time taken to process a single query. <strong>Diversity</strong> can be assessed using metrics like cosine similarity, where lower similarity scores indicate higher diversity.</p><h4>Looking Ahead: Future Directions in RAG Evaluation</h4><p>Several areas for future development in RAG evaluation. These include:</p><p>● <strong>Moving Beyond Traditional QA Datasets and Metrics:</strong> While QA-based benchmarks are useful, they may not adequately capture the unique challenges and capabilities of RAG systems. More diverse and RAG-specific benchmarks are needed to thoroughly assess the system’s overall performance.</p><p>● <strong>Addressing Challenges with LLM-Based Evaluation:</strong> While using LLMs as judges holds promise, aligning their assessments with human judgment is crucial. Further research is needed to develop standardized grading scales and prompting techniques to ensure consistency and fairness in LLM-based evaluation.</p><p>● <strong>Developing Resource-Efficient Evaluation Methods:</strong> The use of LLMs for data generation and evaluation can be computationally expensive. Exploring more resource-efficient methods that can effectively assess RAG systems using smaller datasets while maintaining validity and reliability is an important area for future research.</p><h3>Conclusion</h3><p>Evaluating RAG systems is a complex and multifaceted task, requiring a thorough understanding of the interplay between retrieval and generation. While significant progress has been made in developing benchmarks and metrics, the field continues to evolve. Addressing known challenges and exploring new avenues for evaluation will be crucial for driving further advancements in this area of natural language processing.</p><p>Source: <a href="https://arxiv.org/pdf/2405.07437">https://arxiv.org/pdf/2405.07437</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e7dc6b9274a8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Introduction to Abstraction Hacking]]></title>
            <link>https://medium.com/@moudather.chelbi/introduction-to-abstraction-hacking-72bae4fa44c4?source=rss-ddbda4ec986------2</link>
            <guid isPermaLink="false">https://medium.com/p/72bae4fa44c4</guid>
            <dc:creator><![CDATA[Moudather Chelbi]]></dc:creator>
            <pubDate>Sat, 05 Oct 2024 18:06:41 GMT</pubDate>
            <atom:updated>2024-10-05T18:06:41.085Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*8_AmKTaK8lf-vKb_" /><figcaption>Photo by <a href="https://unsplash.com/@dynamicwang?utm_source=medium&amp;utm_medium=referral">Dynamic Wang</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>One-Sentence Summary</h3><p>Abstraction hacking involves understanding complex systems deeply enough to creatively manipulate or exploit their rules in ways that the creators never intended.</p><h3>One-Paragraph Summary</h3><p>Abstraction hacking is the art of breaking down complex systems — be it law, finance, art, or technology — and understanding their underlying components to creatively manipulate them for unintended but often beneficial purposes. Lawyers exploit legal loopholes, accountants navigate tax codes to find deductions, speedrunners exploit game mechanics, artists challenge the boundaries of what constitutes art, and entrepreneurs disrupt industries by reimagining traditional business models. From social activists bending norms to magicians exploiting human perception, abstraction hacking is at the core of creative problem-solving and innovation across many domains.</p><h3>Foundation</h3><p>In a world defined by rules, codes, and systems, there are individuals who navigate these abstract frameworks not by following the intended path but by finding creative, sometimes unexpected, ways to use them. These people can be called abstraction hackers. Abstraction hacking is the art of understanding a system at such a deep level that you can bend or exploit its rules in ways that the creators never imagined. It’s a concept that spans fields as diverse as law, technology, finance, and even art.</p><h4>Lawyers as Abstraction Hackers</h4><p>Take lawyers, for example. On the surface, a lawyer’s job is to ensure their clients operate within the boundaries of the law. However, the real skill lies in understanding the law so well that they can identify and exploit loopholes. A loophole is, essentially, an unintended gap in the framework — an opportunity that arises when the complexity of the legal system meets human ingenuity. Lawyers deconstruct statutes, contracts, and legal precedents, finding places where the language or intentions of lawmakers leave room for creative interpretation. They are hacking an abstraction — the law — just as a software engineer might reverse-engineer a piece of code.</p><h4>Tax Accountants as Abstraction Hackers</h4><p>Consider tax accountants. The tax code is a dense web of regulations, filled with nuanced rules. Expert accountants are, in many ways, abstraction hackers of the financial system. They explore every deduction, exemption, and credit, working to minimize their clients’ tax liability. The tax code, like any system, is an abstraction — a representation of the government’s intent. But within that abstraction, accountants find cracks and opportunities, just like a hacker finding a way to bypass security measures. By creatively interpreting the rules, they help clients save money while staying within the boundaries of legality.</p><h4>Gamers as Abstraction Hackers</h4><p>This concept can also be seen in the world of gaming. Speedrunners are gamers who play video games with the goal of completing them as fast as possible. To do this, they exploit glitches, bugs, and unintended behaviors in the game’s design. A wall that doesn’t quite have collision detection, a particular sequence of actions that lets you skip entire levels — these are all examples of players manipulating the game’s rules to achieve their goals. Speedrunners learn the inner workings of the game’s mechanics, transforming them into opportunities for creative problem-solving. Here, the game is the abstraction, and the speedrunners are the hackers.</p><h4>Artists as Abstraction Hackers</h4><p>Street artists, too, can be considered abstraction hackers. Rather than confining their work to galleries, they take their art to the streets, challenging conventional boundaries of where art is “allowed” to exist and who has access to it. Banksy, for instance, uses the public space as his canvas, subverting the traditional structures of the art world and democratizing art. By choosing walls, bridges, or abandoned buildings as his medium, Banksy effectively hacks the abstraction of art ownership and access, creating work that anyone can see but no one can possess. His art makes us rethink our understanding of public versus private spaces, ownership, and value.</p><h4>Social Activists as Abstraction Hackers</h4><p>Social norms are also abstractions that can be hacked. Consider activists who find ways to creatively challenge societal rules to promote change. For instance, civil rights activists have historically used nonviolent protests to highlight injustices. By understanding the social expectations and leveraging them in unexpected ways, these activists hack the abstraction of social behavior, forcing society to confront its inconsistencies and reconsider its norms. They utilize the gaps in social constructs to bring about change in ways that the existing system often did not anticipate.</p><h4>Entrepreneurs as Abstraction Hackers</h4><p>Entrepreneurs often hack the abstractions of existing markets and industries. They look at the established norms of a business sector and find innovative ways to disrupt them. Consider how companies like Airbnb and Uber reimagined hospitality and transportation. Instead of building hotels or buying fleets of cars, they found a way to leverage existing resources — people’s homes and personal vehicles. They hacked the abstraction of ownership in these industries and created entirely new business models that transformed their respective sectors.</p><h4>Magicians as Abstraction Hackers</h4><p>Magicians are also abstraction hackers of human perception. Magic tricks rely on manipulating the assumptions our brains make about how the world works. Magicians exploit these assumptions — whether it’s how we expect an object to move, or how we think our attention can be divided. By understanding the abstract principles of human perception and attention, magicians create illusions that defy our understanding, effectively hacking the abstraction of how we process reality.</p><h3>The Essence of Abstraction Hacking</h3><p>At its core, abstraction hacking is about breaking down a system into its fundamental components, understanding how those components interact, and finding ways to creatively manipulate them. Whether it’s the law, tax codes, games, social norms, art, markets, or human perception, abstraction hackers exploit the gap between how a system is supposed to work and how it actually does. This mindset doesn’t belong solely to malicious actors — it’s also the foundation of innovation. It’s what allows entrepreneurs to disrupt industries, artists to break boundaries, and magicians to create wonder.</p><p>Abstraction hacking is like the immune system of a complex society: it identifies weak spots, adapts, and ultimately makes the system more robust by challenging its inconsistencies. Those who learn to hack abstractions often find opportunities where others see only barriers, making systems stronger, more resilient, and more adaptable.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=72bae4fa44c4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Turning Habits into Financial Instruments: A Thought-Experiment]]></title>
            <link>https://medium.com/@moudather.chelbi/turning-habits-into-financial-instruments-a-thought-experiment-c4449266ecaf?source=rss-ddbda4ec986------2</link>
            <guid isPermaLink="false">https://medium.com/p/c4449266ecaf</guid>
            <category><![CDATA[finance]]></category>
            <category><![CDATA[thought-experiment]]></category>
            <category><![CDATA[habits]]></category>
            <dc:creator><![CDATA[Moudather Chelbi]]></dc:creator>
            <pubDate>Fri, 27 Sep 2024 10:38:42 GMT</pubDate>
            <atom:updated>2024-09-27T10:41:50.755Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Or9Q23FPbwVUtKAJ" /><figcaption>Photo by <a href="https://unsplash.com/@austindistel?utm_source=medium&amp;utm_medium=referral">Austin Distel</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>In the journey of personal development, building and maintaining good habits is often essential. From exercising regularly to reading daily, consistent habits can lead to significant improvements in our lives. But what if we could take this a step further by turning our habits into financial instruments? This idea blends personal commitment with financial mechanisms, offering a unique approach to habit formation.</p><h3>What Are Habit Contracts?</h3><p>At the heart of this concept are <strong>habit contracts</strong> — agreements where an individual commits to maintaining a specific habit over a defined period. Traditionally, these contracts might involve personal pledges or agreements with friends and family to stay accountable. However, by integrating financial instruments, habit contracts can take on a new dimension, allowing them to be tokenized and traded like traditional financial assets.</p><h3>How It Works: Examples and Scenarios</h3><p>Let’s explore how habits can be transformed into financial instruments through some practical scenarios:</p><h4>Scenario 1: Exercising Regularly</h4><p>Imagine you commit to exercising three times a week for three months. This commitment is formalized into a habit contract and represented as a <strong>Base Habit Token (BHT)</strong>. This token acts as the foundational asset.</p><ul><li><strong>Futures Contracts:</strong> You could create a futures contract based on your BHT. If you successfully maintain your exercise routine, the value of the futures contract increases. If you miss sessions, the value decreases. Investors can buy and sell these futures contracts, betting on your ability to stick to your exercise plan.</li><li><strong>Options Contracts:</strong> Suppose you have an option contract that gives you the right, but not the obligation, to sell your BHT at a predetermined price. If you consistently exercise, the option becomes more valuable, allowing you to potentially profit from your commitment.</li></ul><h4>Scenario 2: Reading Daily</h4><p>Consider a commitment to read for 30 minutes every day for six months. This habit is tokenized into a BHT.</p><ul><li><strong>Swaps:</strong> You could enter into a swap agreement where you exchange your BHT with someone else’s financial instrument, based on the performance of your reading habit. If you meet your reading goals, both parties benefit; if not, there may be financial penalties or adjustments.</li><li><strong>Derivatives Trading:</strong> Traders can speculate on the performance of your reading habit by buying and selling derivatives based on your BHT. Successful habit maintenance could drive the demand and value of these derivatives.</li></ul><h3>Potential Benefits</h3><ol><li><strong>Enhanced Motivation:</strong> By attaching financial incentives to habit formation, individuals might feel a stronger motivation to maintain their commitments. The potential for financial gain can serve as an additional driver beyond personal satisfaction.</li><li><strong>Increased Accountability:</strong> Financial stakes can enhance personal accountability. Knowing that there is a tangible investment tied to your habits may encourage more consistent effort.</li><li><strong>Market Engagement:</strong> This concept introduces an innovative way to engage with personal development, making the process more interactive and potentially more engaging for those interested in finance.</li></ol><h3>Challenges</h3><p>While the idea is intriguing, several challenges need to be addressed:</p><ul><li><strong>Privacy Concerns:</strong> Tracking and verifying personal habits could raise significant privacy issues. Ensuring that individuals’ personal data is protected is crucial.</li><li><strong>Ethical Implications:</strong> Financializing personal goals might lead to unintended consequences, such as increased stress or the commodification of personal well-being.</li><li><strong>Impact on Intrinsic Motivation:</strong> Relying on financial incentives could undermine intrinsic motivation, making individuals dependent on external rewards rather than internal drives.</li></ul><h3>Simulation</h3><p><a href="https://github.com/vinerya/habitfi">https://github.com/vinerya/habitfi</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c4449266ecaf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Cycle of Extremism: How Far-Right Actions Spark Liberal Waves and Potential Resurgence]]></title>
            <link>https://medium.com/@moudather.chelbi/the-cycle-of-extremism-how-far-right-actions-spark-liberal-waves-and-potential-resurgence-44d35d17d79c?source=rss-ddbda4ec986------2</link>
            <guid isPermaLink="false">https://medium.com/p/44d35d17d79c</guid>
            <dc:creator><![CDATA[Moudather Chelbi]]></dc:creator>
            <pubDate>Thu, 19 Sep 2024 09:56:02 GMT</pubDate>
            <atom:updated>2024-09-27T11:20:27.015Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Abstract</em>: This article explores the historical pattern of political extremism swinging between far-right conservatism and liberalism. It examines how the extreme actions of far-right movements, particularly fascism during the early 20th century, led to a significant shift toward liberal ideologies. The discussion also delves into the current trajectory of liberalism and considers whether its extreme manifestations could precipitate a resurgence of far-right extremism.</p><h3>Introduction</h3><p>History often unfolds in cycles, especially in the realm of political ideologies. The pendulum swings between conservatism and liberalism, sometimes veering into extreme territories on either side. Observers have noted that the extreme actions of far-right movements, notably during the World Wars, catalyzed a shift toward liberalism. However, there is growing concern that the current trajectory of liberalism, if it continues toward extreme ends, might inadvertently pave the way for a resurgence of far-right ideologies.</p><h3>The Rise of Far-Right Extremism and Fascism</h3><p>The early 20th century witnessed the ascendancy of far-right extremism in several parts of the world. Fascist regimes in Italy under Benito Mussolini and Nazi Germany under Adolf Hitler epitomized this shift. These movements were characterized by authoritarianism, nationalism, and the suppression of opposition. The consequences were devastating: widespread human rights abuses, World War II, and the Holocaust.</p><p>The global community’s reaction to these atrocities was one of horror and a collective resolve to prevent such extremism from reoccurring. This led to the establishment of international bodies like the United Nations and a push toward more liberal, democratic governance structures in many countries.</p><h3>The Shift Toward Liberalism</h3><p>In the aftermath of World War II, there was a significant ideological shift. Liberalism, with its emphasis on individual rights, equality, and democratic governance, gained prominence. The civil rights movement in the United States, the decolonization of Africa and Asia, and the expansion of social welfare programs in Europe are testaments to this liberal wave.</p><p>This period also saw the rise of progressive values, including gender equality, environmentalism, and multiculturalism. Governments began to adopt policies that aimed to rectify historical injustices and promote inclusivity.</p><h3>The Trajectory Toward Extreme Liberalism</h3><p>While liberalism has contributed to significant social advancements, some argue that it is reaching extreme levels in certain contexts. Critics point to phenomena such as “cancel culture,” perceived restrictions on free speech, and policies they view as overreaching in areas like immigration and identity politics.</p><p>These critics contend that when liberalism suppresses dissenting viewpoints or enforces conformity, it strays from its foundational principles of open dialogue and individual freedoms. Such extreme manifestations may alienate segments of the population who feel their values and opinions are marginalized.</p><h3>Potential Resurgence of Far-Right Extremism</h3><p>History suggests that when one ideological extreme becomes dominant, it can provoke a counter-reaction. The fear is that extreme liberalism may inadvertently fuel the very forces it opposes. Evidence of a far-right resurgence can be seen in the rise of nationalist parties in Europe, increased polarization in U.S. politics, and the popularity of leaders with authoritarian tendencies in various countries.</p><p>These movements often capitalize on economic anxieties, cultural shifts, and a sense of disenfranchisement among certain groups. They may present themselves as defenders of traditional values and national identity against the perceived excesses of liberal policies.</p><h3>The Cyclical Nature of Political Extremism</h3><p>The oscillation between ideological extremes raises questions about the cyclical nature of politics. Philosophers and historians, such as Hegel with his dialectical method, have long debated this phenomenon. The pattern suggests that society swings from one extreme to another, seeking balance but often overcorrecting in the process.</p><p>Understanding this cycle is crucial for developing strategies to mitigate extreme shifts. Promoting open dialogue, addressing legitimate concerns across the political spectrum, and fostering inclusive policies that consider diverse perspectives may help prevent the pendulum from swinging too far in any one direction.</p><h3>Conclusion</h3><p>The historical pattern of extremism begetting extremism serves as a cautionary tale. While liberalism has played a vital role in advancing human rights and equality, it must be vigilant against adopting extreme positions that could undermine its core values. By striving for balance and fostering an environment where multiple viewpoints can coexist, society may avoid repeating the cycles of the past.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=44d35d17d79c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Teaching AI to Decode Magic]]></title>
            <link>https://medium.com/@moudather.chelbi/teaching-ai-to-decode-magic-81051c64d39c?source=rss-ddbda4ec986------2</link>
            <guid isPermaLink="false">https://medium.com/p/81051c64d39c</guid>
            <category><![CDATA[magic]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Moudather Chelbi]]></dc:creator>
            <pubDate>Tue, 17 Sep 2024 17:28:27 GMT</pubDate>
            <atom:updated>2024-09-17T17:28:27.156Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction</h3><p>Artificial Intelligence has made remarkable strides in recent years, especially in the realms of computer vision and natural language processing. Models like CLIP and other vision-language frameworks have enabled machines to recognize images and generate descriptive text with impressive accuracy.</p><p>But as we push the boundaries of what AI can achieve, a fascinating thought experiment emerges:</p><p><strong>What if we could test AI models by having them understand and explain a magician’s card trick?</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WEGJAi-1g7ajEJguY2OCoA.jpeg" /></figure><h3>The Magic of Testing AI with Card Tricks</h3><p>Magic tricks — especially card tricks — are a unique blend of storytelling, sleight of hand, misdirection, and psychological manipulation. Designed to deceive the observer, these tricks make people believe in the impossible.</p><p>For an AI, unraveling a magic trick isn’t just about recognizing objects or actions; it’s about understanding subtle cues, predicting hidden movements, and interpreting intent.</p><h3>Why Card Tricks Make an Ideal AI Test</h3><ol><li><strong>Complex Visual Sequences</strong>:<br>Card tricks often involve rapid hand movements and intricate sequences that challenge an AI’s ability to track and interpret visual data over time.</li><li><strong>Misdirection and Deception</strong>:<br>Magicians use misdirection to draw attention away from the secret move. Testing AI models on these scenarios evaluates their focus and ability to detect anomalies.</li><li><strong>Narrative Context</strong>:<br>Many tricks are presented within a story or narrative framework. Understanding the trick requires the AI to comprehend the language and context, not just the visuals.</li><li><strong>Inference of Hidden Actions</strong>:<br>The essence of magic lies in what the audience doesn’t see. AI must infer actions that are intentionally concealed, pushing the model’s reasoning capabilities.</li></ol><h3>Proposed Testing Framework</h3><h4>1. Video Analysis of Magic Tricks</h4><ul><li><strong>Input</strong>: Provide the AI model with a video of a magician performing a card trick.</li><li><strong>Objective</strong>: Have the AI observe the performance and generate a step-by-step explanation of how the trick was executed.</li></ul><h4>2. Evaluation Criteria</h4><ul><li><strong>Accuracy of Explanation</strong>: Assess how accurately the AI can explain the mechanics of the trick.</li><li><strong>Attention to Detail</strong>: Evaluate the model’s ability to notice subtle cues and slight movements.</li><li><strong>Temporal Reasoning</strong>: Test the understanding of the sequence and timing of events.</li><li><strong>Language Comprehension</strong>: Determine how well the AI interprets the magician’s narrative or patter accompanying the trick.</li></ul><h3>Benefits of This Testing Approach</h3><h4>Advancing Visual Understanding</h4><p>By analyzing magic tricks, AI models are challenged to process complex visual information that goes beyond static images or straightforward videos. This could lead to more sophisticated visual processing algorithms.</p><h4>Enhancing Reasoning Abilities</h4><p>Understanding a magic trick requires reasoning about unseen events and intentions. This can lead to improvements in an AI’s ability to make inferences and understand causality.</p><h4>Improving Multimodal Integration</h4><p>Magic performances often combine visual elements with verbal storytelling. Testing AI models in this context promotes better integration between visual and linguistic data processing.</p><h3>Potential Applications</h3><ul><li><strong>Surveillance and Security</strong>: Improved ability to detect deceptive or anomalous behavior in surveillance footage.</li><li><strong>Robotics</strong>: Enhanced understanding of human actions can lead to more intuitive human-robot interactions.</li><li><strong>Healthcare</strong>: Better interpretation of complex visual data, such as reading medical imagery or monitoring patient movements.</li></ul><h3>Conclusion</h3><p>Testing AI models with magician card tricks presents an exciting opportunity to push the boundaries of artificial intelligence. It challenges models to move beyond basic recognition tasks and toward true understanding and reasoning.</p><p>By embracing this innovative approach, we can accelerate the development of AI systems that are more perceptive, intelligent, and capable of navigating the complexities of the real world.</p><h3>Join the Conversation</h3><p>Are you an AI researcher, magician, or enthusiast intrigued by this idea? Share your thoughts and let’s explore how we can collaboratively push AI to new limits.</p><p><strong>Tags</strong>: AI, Artificial Intelligence, Machine Learning, Vision-Language Models, CLIP, Innovation, Technology, Future of AI, Magic Tricks</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=81051c64d39c" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>