Adel Nehme

👉🏼 Detection of ChatGPT fake science with the xFakeSci learning algorithm 🤓 Ahmed Abdeen Hamed 👇🏻 https://lnkd.in/eSZQ59Hr 🔍 Focus on data insights: - The study demonstrates how AI-generated content can be distinguished from scientific articles. - Introduction of xFakeSci, a learning algorithm capable of identifying ChatGPT-generated articles. - Calibration step incorporated to address overfitting issues using data-driven heuristics. 💡 Main outcomes and implications: - xFakeSci algorithm achieved F1 scores ranging from 80 to 94%, outperforming traditional data mining algorithms. - Promising performance in predicting fake science generated by ChatGPT. - Significant step towards combating fake science in the field. 📚 Field significance: - Generative AI tools like ChatGPT are reshaping the landscape of scientific content creation. - Importance of developing algorithms like xFakeSci to maintain the integrity of scientific publications. 🗄️: [#detection #ChatGPT #fakeScience #xFakeSci #learningAlgorithm #dataInsights]

1

Note #40 on Generative AI and foundation models: How to create an effective RAG system tailored to an organization's specific domain and needs? Creating a Retrieval-Augmented Generation (#rag) system using a #llm for a domain-specific application within an organization involves several essential steps. Here are the steps to do so: 1- Define objectives/use cases: e.g. answering customer queries, generating reports, or assisting with research. 2- Data collection/preparation: collect data and clean relevant documents, databases, and other resources within the organization's domain. 3- Indexing: index the data to make it searchable. e.g. elastic-search or other search engines can be used. 4- Select and fine-tune the LLM: e.g. GPT-4, BERT (w.r.t. the use case) and fine-tune the LLM with domain-specific data to improve its understanding and generation capabilities in your specific context. 5- Implement retrieval component: via developing or integrating a search engine to retrieve relevant documents from the indexed data. Also implement a system to process user queries and convert them into a format suitable for the search engine. 6- Develop the generation component: via using the retrieved documents to provide context to the LLM for generating accurate and relevant responses. Then integrate the retrieval component with the LLM to form a coherent RAG pipeline. 7- Design the system architecture: set up backend services to handle the retrieval, processing, and generation tasks and create APIs to facilitate communication between different components of the RAG system. 8- Evaluation and testing: we of course need to assign metrics to evaluate the performance of the RAG system: e.g. accuracy, relevance, and response time. You have to continuously refine the system based on feedback and performance metrics. 9- Deployment/monitoring: within the organization’s infrastructure or on a cloud platform and implement monitoring tools to track system performance, detect issues, and ensure reliability and scalability 10- User training and support: providing training for users to effectively utilize the prepared RAG system via set up support channels to assist users with any issues or questions. 11- Maintenance and updates: the RAG system needs to be updated and maintained regularly for new data, improvements, its smooth operation and relevance. pic ref: developer.nvidia.com #machinelearning #deeplearning #neuralnetwork #nlp #generatieveai #llm #llms #datascience #dataanalytics #foundationmodels #genai #ml #cloudcomputing #cloudai #nlp #ai #transformers #foundatinmodels #finetune #finetunellm #rag #ragsystem

5

This is probably the best DSPy course on the internet. And I'm frankly not biased at all. Learn to program, not prompt, language models. Are you fed up with the prompt shenanigans that lead to your overall system becoming brittle over time? DSPy is here to help you build robust LLM-based applications.  I find this tutorial amazing because it is practical. No nonsense, just practical ways to learn how to harness the power of DSPy.

5

We had this coming: MoA LLM -> Mixture of Experts (MoE) Agents -> #Mixture of Agents (MoA) Let's understand them in simple terms. #MoE: Within one large model, sub-models are created on specific parts of data to develop an expert on that part of data. E.g. One sub-model specializes in predicting health conditions for Native Hawaiians while another sub-model is specialized for Asians. MoE reduces training time, computational costs while maintaining (or improving) accuracies of #LLMs. Twitter / X's #Grok-1, Mistral AI's #Mixtral 8x7B (8 experts of 7 billion parameters each), Databricks's #DBRX are based on this architecture. Ensemble models (EM): Why am I suddenly jumping to EM? Because, in your discussions, this may come up as a concept similar to MoE at #LLM-level, just like EM is at traditional #ML-level (e.g. Gradient Boosting Random Forest). But there is one distinction - In EM, each model is trained on all the data while in MoE, each sub-model (expert) is trained on a subset of data. #MoA: While MoE combines multiple neutral networks (#experts), MoA combines outputs from multiple Agents (potentially accessing multiple LLMs internally). Advantage: More accurate response with improved quality. Disadvantage: Longer inference time; to the time order of 5x. Solution: Groq. It reduces inference time to as low as 1x! #ArtificialIntelligence #AI

20

What TikTok Teaches Us on Algorithm Transparency 📱🔍 There’s a reasonably obvious joke in data circles that ‘whatever developers don’t want to explain is called The Algorithm’. 🤔 And to a certain extent that’s correct - and entirely unacceptable. I’ve heard the name of The Algorithm taken in vain by lots of people - whether they’re talking about ‘not wanting to raise their kids on The Algorithm’, or ‘we must make content which does well in The Algorithm’. Hell, there’s entire industries out there predicated on having a better guess as to what happens inside the recommendation engines of big tech firms. 🏢 But what are we actually talking about? Can you imagine if you went into a business meeting having made a key decision, and the only thing you had to say as to your reasoning was ‘The Algorithm said so’. Obviously that’s madness and AI cult behaviour of the highest order - you have to understand the mechanisms behind your decision making process. 🧠 So what can we learn from TikTok? On the face of it, not a lot. However, I’ve been regularly posting on the platform for over a year now - and have some guesses as to what’s going on under the hood. (Also, my handle is @MLGonzo1). 📱 Firstly, one of the main predictors of success within the platform is categorization of your videos. Whenever you interact with a video, TikTok will try to recommend more videos for you to watch, and one of the main drivers of this is through a pre-filled search bar to discover other relevant content. High performing videos have two things in common: they fit into one of the search categories which TikTok has, and they are in a high performing category (I believe ‘cats of TikTok’ is one of the highest performing, obviously). If you fit into one of the categories that TikTok’s model has determined to exist, then that algorithm ‘knows’ (quotes intentional) what to do with your video. Conversely, if you’re like me and make obscure videos about obsolete audio formats - then the algorithm has no clue what you’re doing and so your videos don’t gain the same traction. 🐱 Because of this insight, we effectively have a feedback loop whereby users can create the sorts of videos that keep people on the platform. This transparency is actually A Good Thing. Imagine the converse situation where videos were promoted entirely at random. Creators would have absolutely no idea what did and didn’t work - and would either be driven mad or give up. 🎥 In your business, you don’t have all the proprietary problems which TikTok has regarding secrets and model theft (or at least internally I hope you don’t!), and so the more information your users have about the levers to pull to get either better insights or predictions from your AI the better. More education is always good. 📚 ❗️ Want more insights like this? Check out our free, Ultimate Guide to AI. Hit the link in the comments! 🔔 Want more of this content in your feed? Don’t forget to hit the bell on my profile.

18

3 Comments

Large Language models are are transforming the way we interact with machines and each other. ✨ As someone who's passionate about the latest developments in artificial intelligence, I've created a video series that delves into the fascinating world of Conversational LLM models. 🌟 The Making of SRKGPT: Crafting an AI with Shahrukh Khan's Style | LLM | LLama v2 Finetuning youtu.be/gYPwx0DR7zc  🌟 Supercharging LLama-2: Enhancing Performance on Any Task with ChatGPT Dataset youtu.be/paGr-t1wSOQ 🌟 Elevating Base Falcon Model with ChatGPT Dataset: A Game-Changing Approach youtu.be/lo11Iczb0Vc 🌟 LLAMA-2 Open-Source LLM: Custom Fine-tuning Made Easy on a Single-GPU Colab Instance | PEFT | LORA youtu.be/8cc4bJtycOA 🌟 Falcon Open-Source LLM: Custom Fine-tuning Made Easy on a Single-GPU Colab Instance | PEFT | LORA youtu.be/CxqZ5j3xlt0 🌟 Conversational AI : Understanding the Technology behind Chat-GPT | GPT| RLHF | Few Shot Inferences youtu.be/JKoJ5YIr2O4 🌟 The Good, the Bad, & the Ugly : Exploring the Dual Nature of Conversational LLMs | Open-Source LLMs youtu.be/MHfzoHC4kek 🌟 Pushing the Boundaries of LLMs: Sparse & Flash Attention, Quantisation, Pruning, Distillation, LORA youtu.be/mF7OM_XU2S4 #datatrek #datascience #machinelearning #statistics #deeplearning #ai

17

2 Comments

Large Language models are are transforming the way we interact with machines and each other. ✨ As someone who's passionate about the latest developments in artificial intelligence, I've created a video series that delves into the fascinating world of Conversational LLM models. 🌟 The Making of SRKGPT: Crafting an AI with Shahrukh Khan's Style | LLM | LLama v2 Finetuning youtu.be/gYPwx0DR7zc  🌟 Supercharging LLama-2: Enhancing Performance on Any Task with ChatGPT Dataset youtu.be/paGr-t1wSOQ 🌟 Elevating Base Falcon Model with ChatGPT Dataset: A Game-Changing Approach youtu.be/lo11Iczb0Vc 🌟 LLAMA-2 Open-Source LLM: Custom Fine-tuning Made Easy on a Single-GPU Colab Instance | PEFT | LORA youtu.be/8cc4bJtycOA 🌟 Falcon Open-Source LLM: Custom Fine-tuning Made Easy on a Single-GPU Colab Instance | PEFT | LORA youtu.be/CxqZ5j3xlt0 🌟 Conversational AI : Understanding the Technology behind Chat-GPT | GPT| RLHF | Few Shot Inferences youtu.be/JKoJ5YIr2O4 🌟 The Good, the Bad, & the Ugly : Exploring the Dual Nature of Conversational LLMs | Open-Source LLMs youtu.be/MHfzoHC4kek 🌟 Pushing the Boundaries of LLMs: Sparse & Flash Attention, Quantisation, Pruning, Distillation, LORA youtu.be/mF7OM_XU2S4 #datatrek #datascience #machinelearning #statistics #deeplearning #ai

6

1 Comment

Many students who transition to machine learning don’t pay enough attention to the optimiser used. They blindly use the ADAM optimiser. There is a very rich theory behind why the ADAM optimiser works so well. To truly understand the ADAM optimiser, you to need to understand: (a) Vanilla Gradient Descent (b) Momentum in Gradient Descent (c) RMSProp ADAM is just a combination of Momentum and RMSProp. It all started with simple gradient descent. This is also called Vanilla Gradient Descent. It was the first algorithm which could optimise the neural network weights and minimise the loss. Slowly, we discovered that vanilla gradient descent has 4 major problems: (1) It is slow. (2) It gets stuck in local minima. (3) It is prone to oscillations for distorted loss function landscapes. (4) It used a fixed step size. Introducing momentum fixed issues (1), (2) and (3). The simplest way to think of momentum is: the past matters. Instead of only using the current value of gradient, we give a lot of weightage to past values. This speeds up gradient descent, avoids local minima and also reduces oscillations. What about issue (4): fixed step size? Slowly updating weights (small gradients) of the neural network should have a larger step size. Quickly updating weights (large gradients) of the neural network should have a smaller step size. RMSProp implemented such an adaptive step size. It fixed issue (4). ADAM combined Momentum and RMSProp. We give weightage to the past and we also use an adaptive step size. It works beautifully. Most modern ML frameworks use the ADAM optimiser. I recorded 4 videos which will help you masters Neural Network optimisers: (1) Vanilla Gradient Descent in Neural Network Optimization: https://lnkd.in/gCbciAu9 (2) Momentum in training neural networks: https://lnkd.in/gZwFz46b (3) Coding the RMSProp Optimizer with Neural Network training: https://lnkd.in/gryU7Rsw (4) Coding the ADAM optimizer for neural networks: https://lnkd.in/grn2V7Yg Learn neural networks from basics. Don’t skip sections and develop superficial knowledge. I constructed one master image to explain it all (for those who are lazy to go through the videos). I have attached this image in the post

40

Looking how to select better representatives from your clusters? Take a look at Lexin Chen's PRIME repo: https://lnkd.in/enM_Btbx Now with updated functionality and tutorials! Great collaboration with Arup Mondal and Alberto Perez! #moleculardynamics #clustering #machinelearning

65

👉🏼 Multimodal ChatGPT-4V for ECG Interpretation: Promise and Limitations 🤓 Lingxuan Zhu 👇🏻 https://lnkd.in/egpNsnJq 🔍 Focus on data insights: - ChatGPT-4V achieved an overall accuracy of 83.87% in ECG interpretation. - Performance varied across question types, with lower accuracy in counting-based questions. - Indicates potential for assisting clinicians in ECG interpretation and decision-making. 💡 Main outcomes and implications: - ChatGPT-4V shows promise in ECG interpretation but needs improvement in diagnostic reliability and quantitative analysis. - Further large-scale studies are needed to evaluate its capabilities fully. - Multimodal AI like ChatGPT may have a significant role in cardiovascular care with technological advancements. 📚 Field significance: - Advancements in AI technology like ChatGPT-4V could enhance ECG interpretation in clinical settings. - The study highlights the potential for AI to support clinicians in cardiovascular medicine. - Ongoing training and improvements are essential for the real-world application of ChatGPT-4V. 🗄️: [#ECGinterpretation #ChatGPT-4V #AItechnology #CardiovascularMedicine]

1

Data science unveils the power of converting raw data into valuable insights. It goes beyond just algorithms and models; it's about uncovering patterns for impactful decisions. In a data-rich world, our mission is to heed what data reveals and transform it into actionable strategies. Data science is pivotal in predicting trends, optimizing processes, and unearthing new opportunities, paving the way for a smarter future. #DataScience #BigData #MachineLearning #Analytics Explore Data Science on my portfolio: https://lnkd.in/g9SWhaNP

4

ARIMA is a bell-weather tool for time series forecasting. But most data scientists don't know how to do ARIMA in Python. I'd like to help. There are a handful of important forecasting models that have stood the test of time: 1. ARIMA - Autoregressive Integrated Moving Average: - SARIMA: Seasonal ARIMA - SARIMAX: Seasonal ARIMA with Exogenous Regressors - AutoARIMA: Automatic SARIMAX using Hyndman-Khandakar algorithm for finding the best ARIMA parameters 2. Exponential Smoothing (Family of State-Space Models) - Holt-Winters Additive & Multiplicative - ETS: Exponential smoothing with Error Trend Seasonal (ETS) components - AutoETS: Exponential Smoothing with Automatic Error Trend Seasonal selection If you'd like to learn about these time series forecast models and how to forecast Sales Demand with them, then I'd like to help. I have a free workshop on Wednesday July 10th where I'll share how to apply these time series forecasting models to real sales data. 👉 Register here: https://lnkd.in/eQC67-vv

152

3 Comments

About a year ago, we started building a product for portfolio tracking. The idea came from our customers who wanted to do more with portfolio company data – their most proprietary dataset – but didn’t have this data in a usable format. Being the naive entrepreneurs we are, Anurag Abbott and I thought, "How hard can it be? We’ll knock it out in 2-3 months." News headline, it is HARD. It took us about 14 months in the end; during which we had hundreds of conversations with investors, built prototypes, ran surveys, and got multiple iterations in the hands of our customers. After selling multiple million dollars in ARR, this is what I’ve learnt about portfolio monitoring: 1️⃣ No one has a system they really love right now I can count on one hand, nay one finger, the number of investors I’ve spoken to who were happy with their portfolio tracking processes. I also have data to back my assertion. Only ~14% of funds in our survey were satisfied with their current portfolio tracking processes. 2️⃣ Portfolio tracking is a significant cost for funds A majority of funds are spending 5+ hours on a single PortCo update, and more than $500K per year on portfolio tracking, including personnel and tools. Portfolio tracking involves tedious, labour-intensive processes – having to extract data from multiple formats, clean up inconsistencies and work with nuance around each company / update requires a significant amount of time and effort. 3️⃣ Private company updates are the wild wild west Funds spend months chasing scattered Portfolio updates that they receive as PDFs, as Excel sheets, as slides, as emails – and within these, there’s data in graphs, data in text, graphs pasted as images, graphs that don’t have data labels. There are hundreds of variations in how companies name their metrics. It is impossible to use this data as is – funds must spend hours extracting, cleaning and standardizing it before it can enable any analysis. (As an aside, I have a lot of sympathy for investors now. Our investor updates are a lot more standardized and better.) 4️⃣ You need humans in the loop. We were all aboard the LLM and AI train – but quickly realized that while AI can help with data extraction, human oversight is necessary for bringing higher-order standardization, nuance and real-world context to portfolio tracking. For now. We're sharing some of our internal research on Portfolio Intelligence for public consumption – this includes 125 Funds across North America and Europe. If you are in the business of investing in and advising companies, the report is a must-read. https://lnkd.in/g-376yWC

69

👉🏼 Detection of ChatGPT fake science with the xFakeSci learning algorithm 🤓 Ahmed Abdeen Hamed 👇🏻 https://lnkd.in/eWXP83PT 🔍 Focus on data insights: - The study demonstrates how AI-generated content can be distinguished from scientific articles. - Introduction of xFakeSci, a learning algorithm capable of identifying ChatGPT-generated articles. - Training and calibration steps incorporated to enhance algorithm performance. 💡 Main outcomes and implications: - xFakeSci algorithm achieved F1 scores ranging from 80 to 94%, outperforming traditional data mining algorithms. - Calibration and proximity distance heuristic contribute to the algorithm's promising performance. - Significant step towards combating fake science generated by AI tools like ChatGPT. 📚 Field significance: - AI-generated content detection - Data-driven heuristics - Combating fake science 🗄️: [#AI #datainsights #fakeScience #xFakeSci #ChatGPT]

1

Are you aware that Google’s ScaNN, Facebook’s Faiss, and many other popular Approximate Nearest Neighbor libraries are powered by an intriguing concept of Product Quantizer? The term Product Quantizer may seem intimidating at first but it’s simply a smarter extension of k-means clustering with some added clever tricks. ANN powers machine learning usecase like recommendation systems, Search, etc. Here's a short video for Data Science enthusiasts I have created on the topic covering Approximate Nearest Neighbor (ANN) search and also covering the concept of Product Quantiser. Like and subscribe for more such interesting concepts. Also, like and share over here for maximum reach. : ) Video Link: youtu.be/50PNumB7s3U YT channel Link:  youtube.com/@datatrek #datatrek #datascience #machinelearning #statistics #deeplearning #ai

8

Are you aware that Google’s ScaNN, Facebook’s Faiss, and many other popular Approximate Nearest Neighbor libraries are powered by an intriguing concept of Product Quantizer? The term Product Quantizer may seem intimidating at first but it’s simply a smarter extension of k-means clustering with some added clever tricks. ANN powers machine learning usecase like recommendation systems, Search, etc. Here's a short video for Data Science enthusiasts I have created on the topic covering Approximate Nearest Neighbor (ANN) search and also covering the concept of Product Quantiser. Like and subscribe for more such interesting concepts. Also, like and share over here for maximum reach. : ) Video Link: youtu.be/50PNumB7s3U YT channel Link:  youtube.com/@datatrek #datatrek #datascience #machinelearning #statistics #deeplearning #ai

26