Raj Sanshi’s Post

🚩 If you want to learn more about digitalization but feel confused by all the different terms and how various fields interact, a good way to understand and visualize it, might be via creating an Euler diagram for the taxonomy of AI which can explain it in a very simplified way. 📎 Here, you can see that deep learning is a subfield of machine learning, and machine learning is a subfield of AI, which itself is a subfield of computer science. Data science is an umbrella term that covers machine learning, statistics, and a bit of computer sciences too; referring that it includes algorithms, data storage, and web application development. Ref: Elements of AI - University of Helsinki #Digitalization #EulerDiagram #AITaxonomy #DeepLearning #MachineLearning #ArtificialIntelligence #ComputerScience #DataScience #TechTerms #Interdisciplinary #TechExplained #SimplifiedTech #LearningAI #TechEducation #Algorithms #DataStorage #WebDevelopment #Statistics #TechCommunity #Innovation #TechTrends #FutureOfTech

🔍 Are LLMs really deceptive? 🤖🧮 Recently, the media spotlight has once again highlighted concerns about how Large Language Models (LLMs) can be deceptive, sometimes outright lying to their users. Yet many rely on chatbots for their convincingly comprehensive answers. This dichotomy is something we need to explore in detail. 🧐 As technology shapes our understanding of the world, it's important to understand both its potential and its shortcomings. We can't miss the forest for the trees. 🌲🌳 The widespread use of LLMs has often been compared to the introduction of calculators in schools. Remember when we weren't allowed to use calculators until we were about 14? It was important to understand mathematical concepts first, to get a 'feel' for calculations and results before relying on calculators. 📚🧮 Once you've mastered the basics, using a calculator or computer becomes incredibly convenient and saves a lot of time. Interestingly, this perspective changed a bit when I was doing my PhD in protein crystallography. There was this one professor who did all the necessary calculations in her head - no computer needed! It was awe-inspiring for us postgraduate students. 🌟 But while earlier PhDs might have involved calculating crystal lattices by hand, that wouldn't have been possible when I defended my thesis. 🧑🎓✨ Back to LLMs and calculators: The main difference is how they work. Calculators and computer programs operate deterministically, giving the same output every time for a given input. In contrast, LLMs operate probabilistically, giving different outputs for the same input based on probabilities. You can even test this for yourself by adjusting the temperature in OpenAI's playground to see how it affects the creativity of the responses. 🌡️💡 Understanding this distinction is crucial to using LLMs effectively in everyday life. They're not deceptive; they work with probabilities, which means there's no one "right" answer - at least not for an LLM. 🤔 Let's keep that in mind as we continue to navigate the fascinating world of AI! 🚀🔍 #AI #LLMs #Technology #Innovation #ProbabilisticThinking #AIResearch Image DALL-E: and it "lied" to me too - its attempts at creating calulators were abysmal 🤣 .

🌟 **Exploring the Boundaries of Machine Learning: From Vision to Finance** 🌟 Machine Learning is transforming how we interact with technology, and its applications are as diverse as human activities themselves. To truly harness its potential, we need to understand and replicate what comes naturally to humans. Let's dive into some key areas where machine learning is making a significant impact: 👁️ **Computer Vision**: Imagine teaching a computer to see! This involves replicating the complex process of human vision—where our eyes capture information, our brains process it, and then make decisions. Developing algorithms for object recognition, for example, aims to mimic how humans can instantly identify a tree in a picture. The challenges are immense, but the possibilities are endless. 🗣️ **Speech Recognition**: Human speech is unstructured data. Different people say the same thing in various ways, yet we effortlessly understand each other. Machine learning algorithms strive to give computers the same ability, understanding and processing natural language as we do. This field is key to advancements in voice-activated assistants and real-time translation. 📝 **Text Analysis**: From deciphering handwriting to extracting meaning from billions of documents on the internet, text analysis is another crucial area. Machine learning can help us understand the semantics of written text, making sense of unstructured data and unlocking valuable insights. 🧠 **Brain-Inspired Learning**: The human brain, with its millions of interconnected neurons, remains the holy grail of machine learning. Understanding how our brain learns and makes decisions can revolutionize AI, though we are still taking baby steps towards this goal. 📈 **Financial Predictions**: The stock market is a prime example of machine learning in action. By analyzing historical data, algorithms can predict stock movements, aiding in smarter investment decisions. 🔬 **Scientific Applications**: From biological experiments to materials engineering, machine learning is a powerful tool in science and engineering. Any experiment generating large datasets can benefit from machine learning, leading to new discoveries and innovations. Machine learning is not just about mimicking human capabilities but also enhancing them. As we make strides in this exciting field, the future holds tremendous potential for smarter, more intuitive technologies. Happy Learning ! 😊 #MachineLearning #ArtificialIntelligence #ComputerVision #SpeechRecognition #TextAnalysis #AI #DataScience #Innovation

Come check Vector Institute papers!!

I knew that humans have lats to be strong, but recently I learned that LLMs have LATS too, for the same purpose of making them stronger and better. 💪 I came across an article https://lnkd.in/dm_TRrw8. which introduced me to LATS — an innovative approach that significantly enhances the reasoning and decision-making capabilities of Large Language Models. ✨ It allows LLMs to plan, reason and act more effectively by leveraging external feedback by the integration of Monte Carlo Tree Search (MCTS) with LLM powered value function and self-reflection. 🚀 The main steps involved: 1️⃣ Selection: At first, it begins by determining which tree segment is best suited for growth. 2️⃣ Expansion: Executes n actions to expand the tree. This results in n new child nodes stored in an external long-term memory structure. 3️⃣ Evaluation: The third step is to evaluate or assigns a score to each child node for the next selection. The value function is calculated by reasoning about a given state. 4️⃣ Simulation: Expand the next node until a terminal state is reached. Feedback is calculated based on the success of the final state otherwise we have to follow two more steps. 5️⃣ Backpropagation: Similar to neural networks, it refines the tree structure. 6️⃣ Reflection: Stores both successful and failed trajectories for providing context to the agent and value function. This optimizes learning by integrating semantically meaningful experiences. LATS overcomes the limitations of previous tree-based methods like ToT prompting, which relied solely on internal reasoning. This approach allows LLMs to consider multiple reasoning paths and adapt to environmental conditions without additional training. 🦾 LATS has demonstrated SOTA performance across various domains, including programming, QA, web navigation, and mathematics, achieving a 92.7% accuracy on humaneval with gpt-4. 💡 Looking forward to seeing how this evolves! 🔎 #LLM #LATS #AI #Genai #Research

LLMs are alien beasts. It is deeply troubling that our frontier models can both achieve silver medal in Math Olympiad but also fail to answer "which number is bigger, 9.11 or 9.9"? The latter question broke the internet recently because none of GPT-4o, Claude-3.5 or Llama-3 could get it right 100% of the time. - Jim Fan This is the current state of large language models (LLMs) and their inconsistent performance across different tasks. It’s striking how LLMs can excel in complex mathematical reasoning, like Math Olympiad problems, yet struggle with basic numerical comparisons such as 9.11 vs. 9.9. This discrepancy highlights several key aspects of LLMs: - Inconsistency: LLMs can perform exceptionally well on complex tasks but fail on simpler ones. - Lack of True Understanding: They may process information intelligently but often lack a fundamental grasp of concepts like numerical magnitude. - Generalization Challenges: LLMs may struggle to apply their capabilities across different contexts and question formats. - "Alien" Nature of AI: Their cognitive processes differ fundamentally from human cognition, leading to unpredictable performance patterns. It also calls for caution when deploying these models in real-world applications where reliable performance on basic tasks is crucial. While LLMs have made impressive strides, they are still far from achieving human-like general intelligence and remain tools with specific strengths and weaknesses. #AI #GENAI #LLM

Based on the insights from the paper "Attention is All you Need" by Google Researchers in 2017, the transformative 'Transformers' architecture has revolutionized the realm of General Artificial Intelligence (Gen AI) and Large Language Model (LLM). Here is a simplified overview of the different components of this architecture for everyone's comprehension: 1. **Tokenization:** The model processes prompts by breaking down sentences into words or subwords, assigning a unique token to each. Think of this as creating a dictionary for the model, where every word and subword has a distinct position for easy reference. The size of this dictionary can vary across different models. 2. **Word Embedding:** Following tokenization, each token is embedded into a multi-dimensional coordinate based on its features, reflecting the word's meaning from its various appearances in sentences. In essence, each word/subword is transformed into a unique vector, with its magnitude as the token and direction as the word embedding. 3. **Positional Encoding:** An array is generated based on the positional order of each word/subword in the prompt, aiding in maintaining the positional information. 4. **Multi-Head Self Attention:** This component enables the model to grasp the context of the current word by considering other words within the sentence. 5. **MLP (Multi-Layer Perceptron):** Drawing on pre-training data from sources like Wikipedia and Quora, the model incorporates knowledge gained during training, enhancing its understanding and predictive capabilities. 6. **Prediction:** Utilizing context, knowledge, and temperature settings, the model predicts the subsequent word in the prompt. The temperature setting, akin to a TV tuner, influences the randomness of the machine's response, with lower values yielding more deterministic predictions and higher values introducing greater variability. This foundational architecture forms the cornerstone of various Gen AI concepts, paving the way for advanced language processing capabilities. Special thanks to Sumit Mittal Sir for the enlightening course "AI - The Ultimate Masters Program for Data Engineers #Sumitmittal #AIforDataEngineers #AI #LLM #AIBASICS #GENAI #DataEngineering #AI #ContinuousLearning

𝐄𝐱𝐩𝐥𝐨𝐫𝐢𝐧𝐠 𝐋𝐢𝐯𝐞𝐁𝐞𝐧𝐜𝐡: 𝐀 𝐍𝐞𝐰 𝐅𝐫𝐨𝐧𝐭𝐢𝐞𝐫 𝐢𝐧 𝐋𝐋𝐌 𝐁𝐞𝐧𝐜𝐡𝐦𝐚𝐫𝐤𝐢𝐧𝐠 🌟 Great news for the AI and machine learning community! Let's dive into 𝐋𝐢𝐯𝐞𝐁𝐞𝐧𝐜𝐡, a pioneering benchmark developed by researchers aiming to robustly evaluate Large Language Models (LLMs) without the common pitfalls of dataset contamination or biases from human or LLM judging. 🔍 𝐊𝐞𝐲 𝐅𝐞𝐚𝐭𝐮𝐫𝐞𝐬 𝐨𝐟 𝐋𝐢𝐯𝐞𝐁𝐞𝐧𝐜𝐡: 𝐑𝐞𝐠𝐮𝐥𝐚𝐫𝐥𝐲 𝐔𝐩𝐝𝐚𝐭𝐞𝐝: LiveBench is dynamic, with questions sourced from the latest math competitions and academic papers, updated monthly. 𝐎𝐛𝐣𝐞𝐜𝐭𝐢𝐯𝐞 𝐄𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐨𝐧: Utilizes automatic scoring based on objective truth, avoiding biases that can occur with human or automated judges. 𝐂𝐨𝐦𝐩𝐫𝐞𝐡𝐞𝐧𝐬𝐢𝐯𝐞 𝐓𝐞𝐬𝐭𝐢𝐧𝐠: Includes a variety of tasks such as math, coding, reasoning, language, instruction following, and data analysis, designed to thoroughly test LLM capabilities. 📈 The rigorous nature of LiveBench is evident as even the top performing models achieve below 60% accuracy, illustrating the challenge it poses and its role in driving model improvements. 🌍 𝐂𝐨𝐦𝐦𝐮𝐧𝐢𝐭𝐲 𝐈𝐧𝐯𝐨𝐥𝐯𝐞𝐦𝐞𝐧𝐭: The benchmark invites contributions and collaboration from the global AI community. To get involved or to learn more about how you can use LiveBench for your projects, check out their resources on GitHub or visit the LiveBench.ai leaderboard. 🛠️ 𝐖𝐡𝐲 𝐢𝐬 𝐋𝐢𝐯𝐞𝐁𝐞𝐧𝐜𝐡 𝐢𝐦𝐩𝐨𝐫𝐭𝐚𝐧𝐭? Traditional benchmarks quickly become outdated as they are absorbed into the training datasets of new models. LiveBench addresses this by offering a continually updated and contamination-free testing environment. 👀 Stay tuned for updates and developments from this exciting initiative, which is setting new standards in the evaluation of AI models! #AI #GenAI #LLM #MachineLearning #DataScience #Benchmarking #TechnologyUpdates #ArtificialIntelligence

LoRA is making fine-tuning large language models much easier and cheaper! Wondering how? Let me explain. Low-Rank Adaptation (LoRA) is a technique that fine-tunes large models without touching most of the original parameters. This means less computing power, less memory, and way faster training. Here’s why LoRA matters: 1. Efficiency: It freezes the original model weights and adds small, low-rank matrices to make task-specific tweaks. So, you don’t need to retrain the whole model from scratch. 2. Cost-Effective: You can fine-tune a huge model without needing a supercomputer. 3. Memory-Friendly: Reduces GPU memory usage by up to 3x. Perfect for resource-constrained environments. But it gets even better with these advanced variants: 1. LoRA+: Speeds up fine-tuning by up to 2x by introducing different learning rates for adapters. Perfect for complex tasks where standard LoRA falls short. 2. QLoRA: Uses quantization to shrink model size drastically. You can fine-tune 65 billion parameter models on just a 48GB GPU. No more struggling with massive hardware requirements. 3. DyLoRA: Dynamically adjusts low-rank adapters during training. This means no more trial and error on finding the right rank – it does it for you, saving tons of time. 4. LoRA-FA: Freezes some weight matrices to reduce memory use even further while keeping performance intact. 5. DoRA: Breaks down weights into direction and magnitude, improving performance while keeping changes to the original model minimal. What’s the point? LoRA techniques are making it easier to fine-tune massive models for specific tasks—without breaking the bank or your machine. Whether you're working on NLP or other data-heavy tasks, these methods ensure you get the most out of your model. Key takeaways: - LoRA allows efficient task adaptation without full retraining. - Memory and resource savings make it perfect for limited hardware setups. - Variants like QLoRA and LoRA+ bring even more benefits, from faster training to handling massive models. Ready to fine-tune smarter, not harder? — I share my learning journey here. Join me and let's grow together. Enjoy this? Repost it to your network and follow Karn Singh for more. #AI #MachineLearning #LLM #LoRA #AIResearch #Finetuning #TechInnovation

🤔 Let's Clear Up a Common Misconception: AI ≠ Superset of ML I often notice a widespread oversimplification: The idea that Machine Learning is simply a subset of Artificial Intelligence. 📚 The Academic Reality: 1️⃣ Historical Context: ML evolved from pattern recognition and computational learning theory AI emerged from the quest to create "thinking machines" These fields developed largely in parallel, not hierarchically 2️⃣ Fundamental Differences: ML focuses on statistical learning and optimization AI encompasses reasoning, knowledge representation, and planning Some ML applications have nothing to do with AI (e.g., statistical regression in economics) 3️⃣ The Intersection: Modern AI systems often leverage ML techniques Modern ML systems sometimes incorporate AI principles BUT: They remain distinct academic disciplines with unique foundations 🎯 Key Point: While there's significant overlap in practice, treating ML as a mere subset of AI oversimplifies rich academic traditions and distinct theoretical foundations. 🧪 Consider This: Statistical learning theory (core to ML) existed before AI Many ML applications in finance and statistics have no AI component Some AI systems (like expert systems) use no ML at all 💭 Why This Matters: Understanding these distinctions helps us: Design better systems Choose appropriate solutions Advance both fields independently What's your take on this? Have you noticed this oversimplification in industry discussions? #MachineLearning #ArtificialIntelligence #ComputerScience #DataScience #AcademicPerspective #TechEducation 🔍 Curious to hear thoughts from other academics and practitioners in the field. What's your experience with this distinction?