Abhinav G.

Here's the technical breakdown of the AI Penalty Challenge authored by Alok Pattani (check out this video if you haven't seen the demo yet: https://lnkd.in/gsD3xjJ2) Key Takeaways: ☁️ Firestore for triggering slow-mo video capture on Android phones & uploading to Google Cloud via Firebase Storage. 🧠 Cloud AutoML for training custom object detectors (ball & goal tracking) to calculate speed & accuracy. 🗣️ Gemini Pro 1.5 on Vertex AI for personalized coaching feedback based on multiple data points. 🎨 Imagen on Vertex AI for instant, prompt-driven image editing for cool player pics. More technical insights from the project to come... #AIChallenge #MachineLearning #SportsTech #GoogleCloud #footballNOTsoccer

89

5 Comments

While Gen AI engineers in India are rightly concerned about #hallucination in LLMs, yet most would rarely appreciate that genius folks in our own country evolved extraordinary measures to arrest hallucination thousands of years ago, which included sophisticated #guardrails which were to be 'assimilated' in the brain itself, impacting the way it perceived reality and 'generated' content (Isn't the brain one of the most hallucinating natural 'LLMs' after all ?). This was not just to ensure that facts were rendered correctly. The attempt was to prevent disasters of another kind and of a much more serious nature. After all, language in India was considered an important vehicle for connecting to the divine, and not merely for communicating with one another. And we often fail to realize this #opensource framework, so to say, was built upon by different intellectuals in different parts of the country, and India ended up having several wonderful languages in different regions, languages which look different, but have a lot in common. Did you not realize that no matter where in India we are from, names of most of us can be traced to this one language? https://lnkd.in/g4Njx4mG #llms #nlp #generativeai

16

3 Comments

🔍 Why Chunking Strategy is Key in Building a RAG System with LLMs 🔍 Recently, a student of mine while attending the interview was asked to explain how he would chose chunking strategy to build a better RAG system. When building a Retrieval-Augmented Generation (RAG) system using LLMs, chunking is a critical step. The process of splitting data into manageable pieces not only affects how well relevant information is retrieved but also impacts the LLM’s ability to generate coherent and accurate responses. • Contextual Understanding: LLMs have limitations in processing extremely long texts. Chunking helps maintain context and ensures the model can focus on relevant information. If you are using Opensource LLM with limitation on context length, this becomes important.  • Efficiency: Smaller chunks reduce the computational burden on the LLM, leading to faster responses. • Retrieval Accuracy: Chunking can improve the accuracy of information retrieval by making it easier to find relevant segments. Essentially, Chunks should not be too large or too small. So, what’s the right answer ? Well, it depends on the problem you are solving. First you should ask below questions to yourselves. • What is the nature of the content being indexed? Long documents, Short sentences, Paragraphs ? • Which embedding model are you using, and what chunk sizes does it perform optimally on? • What are your expectations for the length and complexity of user queries? Will they be short and specific or long and complex? This may inform the way you choose to chunk • How will the retrieved results be utilized within your specific application? For example, will they be used for semantic search, question answering, summarization, or other purposes? Then chose one of the below strategies: 𝗖𝗼𝗺𝗺𝗼𝗻 𝗖𝗵𝘂𝗻𝗸𝗶𝗻𝗴 𝗦𝘁𝗿𝗮𝘁𝗲𝗴𝗶𝗲𝘀: Sentence-Based Chunking Paragraph-Based Chunking Token-Based Chunking Overlapping Chunking (Sliding Window) Topic-Based Chunking: Identify topic boundaries using techniques like topic modeling or keyword extraction. Specialized chunking : You would need specialised packages for chunking if you are using LaTex or Markdown etc. Take that into consideration. Semantic Chunking : Variable length chunking retaining the semantic coherence. Creating Sentence groups etc. Langchain has a Semantic Chunking splitter. Hybrid Chunking: Combine multiple strategies to optimize for your specific use case. Choosing the Right Strategy: The best chunking strategy depends on factors like the structure of your documents, the complexity of the information, and the desired level of granularity. Experimentation is often key to finding the optimal approach. By carefully considering chunking strategies, you can significantly enhance the performance and effectiveness of your RAG system.

7

1 Comment

AI competition is evolving beyond just which language model is best. Behind the scenes, a chip war is driving huge investments. Here's what's happening: 𝐓𝐡𝐞 𝐂𝐡𝐢𝐩 𝐖𝐚𝐫 Chips are the engines of AI, and Nvidia has been leading the pack. But recent developments show other players making big moves: 𝐍𝐯𝐢𝐝𝐢𝐚: Announced the H200 GPU with significant upgrades, expected in Q2 2024. 𝐌𝐢𝐜𝐫𝐨𝐬𝐨𝐟𝐭: Unveiled the Maia 100 AI chip, expected in early 2024. 𝐀𝐦𝐚𝐳𝐨𝐧: Introduced AWS Graviton4 and AWS Trainium2 chips for AI and ML applications. 𝐆𝐨𝐨𝐠𝐥𝐞: Also entering the race with their chip initiatives. These companies are investing heavily to dominate the chip market. They’re not just developing technology—they’re making strategic plays to secure their positions. In summary, these investments in cloud services and chips are strategic land grabs. The competition is fierce, and it’s going to be an interesting year ahead in the AI space. Stay tuned! #AICompetition #CloudWars #ChipWars #AWS #MicrosoftAzure #GoogleCloud #Nvidia #TechInnovation #ArtificialIntelligence #MachineLearning #AIChips #TechTrends #FutureOfAI #CloudComputing #Innovation #TechNews #AIIndustry #DigitalTransformation #TechStrategy #BusinessGrowth #TechInvestments

6

🚀 Day 18 of My 25-Day Generative AI and LLM Challenge 🤖🚀 🌟 How Variational Autoencoders (VAEs) Work: Encoding & Decoding 🌟with my friends HARSH JAVIA and Maanan Jagani! Variational Autoencoders (VAEs) are a type of generative model that aim to learn the underlying structure of data. Here's a simplified breakdown of how they work, using a bit of professional flair and emojis to keep it engaging! 🔍 1. Understanding the Basics🔍 At a high level, VAEs consist of two main components: 1. Encoder 🧠: - Takes input data (e.g., images, text) and compresses it into a lower-dimensional space known as the **latent space**. - Think of this as summarizing the essence of the input data into a compact representation. 2. Decoder 💡: - Takes the encoded representation from the latent space and tries to reconstruct the original input data. - This process helps the model learn to generate new data that is similar to the input data. 🚀 2. Step-by-Step Process 🚀 Step 1: Encoding 📥 - The encoder network receives the input data and transforms it into a mean (μ) and a standard deviation (σ). - From these, a latent vector \( z \) is sampled using the reparameterization trick: \( z = \mu + \sigma \cdot \epsilon \), where \( \epsilon \) is a random variable from a standard normal distribution. Step 2: Latent Space 🌌 - The latent space is a lower-dimensional representation that captures the essential features of the input data. - This space allows the model to explore various possible data representations. Step 3: Decoding 📤 - The decoder network takes the latent vector \( z \) and attempts to reconstruct the original input data. - The quality of the reconstruction indicates how well the model has learned the data's underlying structure. 🌟 3. Why VAEs are Powerful🌟 - Data Generation 🎨: VAEs can generate new, unseen data that is similar to the training data, useful in creative applications like art and music. - Dimensionality Reduction 📉: They reduce the complexity of data, making it easier to visualize and analyze. - Anomaly Detection 🔍: VAEs can identify anomalies by recognizing data that doesn’t fit the learned patterns. 🏆 4. Professional Takeaway 🏆 Variational Autoencoders are a powerful tool in the machine learning toolbox, enabling efficient data compression, reconstruction, and generation. Their ability to model complex data distributions makes them invaluable for various applications, from image generation to anomaly detection. #MachineLearning #AI #DataScience #DeepLearning #GenerativeModels #VariationalAutoencoders #VAEs #NeuralNetworks #DataInnovation #TechTrends #ArtificialIntelligence #MLApplications #DataGeneration #TechInspiration #Innovation #AIResearch #DataAnalytics #DeepLearningModels #TechCommunity

6

Back in 2017, the game-changing paper “Attention Is All You Need” by Ashish Vaswani and his team pretty much set up today’s Large Language Models (LLMs) like GPT and BERT for success. Instead of forcing models to remember everything in sequence (like a talkative friend who won’t skip a single detail), this “attention” mechanism lets them focus only on the important stuff—faster, smarter, and with a bit more personality. It’s the secret sauce behind LLMs. Curious to dig deeper? Check out the original paper that started it all: https://lnkd.in/gAbU_urJ #LLM #AITrends #GPT #transformers #AIResearch

10

Exciting shift in US healthcare: 4 in 10 doctors are ready to integrate generative AI in patient interactions, per a recent survey shared with Axios. While highlighting potential to enhance care, transparency and oversight are key. https://lnkd.in/gPJZAB3S

11

A team of brilliant minds from University of Pennsylvania, NVIDIA, and The University of Texas at Austin has unveiled DrEureka, an innovative AI system that is transforming the landscape of sim-to-real transfer in robotics. This cutting-edge approach utilizes Large Language Models (#LLM) to automate the intricate process of designing reward functions and domain randomization for robot tasks. Traditionally, the transfer of policies from simulation to real-world applications has been a labor-intensive process, requiring meticulous manual tuning. However, DrEureka has shattered these barriers, demonstrating remarkable efficiency and effectiveness in tasks such as quadruped locomotion and dexterous manipulation. The system's capabilities extend to more complex challenges, including balancing and walking on a yoga ball, showcasing an unprecedented level of adaptability and robustness. DrEureka's AI-guided policies have not only rivaled but surpassed the performance of human-designed configurations, setting a new benchmark in the field. Through rigorous experimentation, DrEureka has proven its mettle in real-world scenarios, maintaining stability and outperforming traditional methods on various terrains. The integration of safety instructions into the LLM reward design has also enhanced the system's reliability, ensuring that the generated reward functions are safe for deployment. The team's #research paper provides a comprehensive look at the methodology and results, highlighting the potential of DrEureka to revolutionize how AI is applied in robotics. As the field continues to evolve, DrEureka stands at the forefront, paving the way for a future where AI not only assists but also innovates in robotics. SRC: https://lnkd.in/gYDNVwjG . ✅Follow for more✅ . #innovation #management #technology #creativity #futurism #startups #marketing #socialmedia #ai #robotics

9

Recently, two Harvard students demonstrated how Meta’s Ray-Ban smart glasses can be turned into a potent privacy threat using facial recognition tech. Their creation, I-XRAY, highlights the precarious balance between innovation and security. Imagine a device that captures video, scans faces, cross-references data, and sends personal details like names and addresses to a user's phone in real-time. This isn't science fiction—it's a reality that AnhPhu Nguyen and Caine Ardayfio showcased. Their intent was to raise awareness, yet it underscores just how vulnerable we are to doxxing and privacy invasions. With wearable tech becoming ever more prevalent, it's crucial to consider the implications for personal security. Should the convenience of live-streaming and instant information come at the cost of our privacy? This demo reminds us that we must navigate the fine line between technological advancements and safeguarding personal data. As professionals, it's our duty to stay vigilant and advocate for stronger privacy protections. How do you balance convenience and security in your tech use? #PrivacyMatters #WearableTech #DataSecurity #AhlamWrites

13

3 Comments

Google's 6th generation Trillium TPUs are now powering arguably the largest RNA-focused AI foundation model on the market. During Google I/O 2024, Canadian "techbio" company Deep Genomics announced that it is using Google Cloud’s next generation Tensor Processing Units (TPUs) in the development of its AI foundation model.  Deep Genomics is the creator of BigRNA, the first AI Foundation model for decoding RNA biology and designing therapeutics (as the company claims). I will soon be publishing my interview with Dr. Brendan Frey, Founder and Chief Innovation Officer at Deep Genomics. Brendan is an incredibly interesting guy who, among a litany of major contributions, was an early pioneer in the development of ML & AI to accurately determine the consequences of genetic mutations. Stay tuned for the upcoming interview publication by subscribing to our newsletter (link in the comments) or by following Where Tech Meets Bio (Substack Newsletter). #techbio #artificialintelligence #foundationmodel #RNA #biopharmatrend Image credit: Deep Genomics

146

11 Comments

🚀 Real-Time Updates Made Simple with Server-Sent Events (SSE) 🚀 When building real-time web applications, it’s essential to keep users in sync with live updates—think of stock price changes, notifications, or chat messages. While technologies like WebSockets get a lot of attention, Server-Sent Events (SSE) offer a simple yet powerful alternative for one-way data streaming from the server to the client. 🤔 What are Server-Sent Events (SSE)? SSE is a standard HTTP-based protocol that allows the server to push updates directly to the client over a single, long-lived connection. Unlike short polling or long polling, SSE provides a continuous stream of updates without requiring the client to repeatedly ask for new data. 💡 How SSE Works: -Client Request: The client initiates an HTTP request using the EventSource API. -Connection Established: The server responds with text/event-stream headers and keeps the connection open. -Data Push: The server sends updates as event data whenever new information is available. -Automatic Reconnect: If the connection drops, the EventSource API automatically tries to reconnect. 🔍 Why Use SSE? ✅ Built-in Browser Support: No need for third-party libraries—just standard HTTP. ✅ Auto-Reconnection: The browser handles reconnections seamlessly. ✅ Lightweight: Ideal for one-way updates without the overhead of WebSockets. 🚧 Limitations: ❌ One-Way Communication: SSE only supports server-to-client updates. For bi-directional communication, consider WebSockets. ❌ Limited Support in Older Browsers: While most modern browsers support SSE, it’s not available in Internet Explorer. ❌ Connection Limits: Browsers typically limit the number of concurrent SSE connections. 📈 Use Cases: Live sports scores and updates ⚽ Real-time notifications and alerts 🔔 Streaming live stock prices 📊 Sample implementation can be found here : https://lnkd.in/gxivVvtY #WebDevelopment #JavaScript #RealTime #ServerSentEvents #NodeJS #CodingTips #SystemDesign

8

Humans have a mind, we can do math. AI models can’t do math, they are math. Language models were made to process language, and their capacity to perform any math is incidental, they can answer 1+1=2 due to frequent exposure to that text, not a true understanding of mathematics. Apple researchers in the new paper “GSM-Symbolic" highlighted a critical flaw in large language models (LLMs): they do not think or reason but merely replicate patterns. When presented with irrelevant information, AI accuracy can drop by up to 65%, even in advanced models. So, the models, from Llama to OpenAI o1, are impressive in many ways, but they are just matching patterns they’ve seen before, and not really understanding the logic behind the problems. Unlike humans, AI struggles with irrelevant information and overly relies on statistical correlations in the training data. Machines lack the depth and flexibility of human cognition. So, when it comes to solving complex problems that need step-by-step logical thinking, AI will make big mistakes. We need to remember that when we think about relying on AI in the future.

65

6 Comments

AI is still advancing the limits of what can be achieved, and a revolutionary method that may revolutionize the field of #medicine has been introduced by #WashingtonStateUniversity. By utilizing an AI technique known as #BayesianOptimization, scientists have improved the efficiency and accuracy of creating realistic human organs using #3Dprinting. This development allows for the creation of highly detailed structures, ranging from artificial organs to wearable biosensors. The possible effect on healthcare and technology is huge #AI is not only focused on efficiency, but also on groundbreaking developments that have the potential to rescue lives. #ArtificialIntelligence #3DPrinting #HealthcareInnovation #MedicalTechnology #AIinHealthcare #FutureOfMedicine #Biotech #Innovation #TechForGood #ArtificialOrgans

4

For those learning Gen AI and Opensource LLMs! Yesterday, I was running Llama3.1 via Ollama locally on my computer and each response took > 5 minutes. Then I switched to Groq and it was instant. They use LPU (Language Processing Units) for super-fast inference. Perfect for students and professionals in their learning journey. You can create an account for free (https://console.groq.com/) and try various open-source models such as Llama3.1, mixtral, gemma, etc. Next week, I will publish an end-to-end GenAI project on the codebasics YouTube channel that uses Groq, Llama3.1, Langchain, Chromadb, and Streamlit. Stay tuned! #genai

375

27 Comments

I would love to see some videos of people using LLMs practically, thoughtfully, and critically—including: 👉 The iterative process of crafting a prompt to get the desired response (not just skipping ahead to the “perfect” prompt). 👉 The process of deciding when it’s gotten to the point of “close enough” and more efficient to fix the remaining issues oneself rather than trying to squeeze more detail and correctness out of better prompting. 👉 The actual time spent analyzing/reviewing/checking/correcting the output to bring it up to the same level of quality as an expert human producing the work themselves. Maybe there are some of these already out there, but I feel like most of the testimonials/experience reports I’ve seen skip or gloss over those parts. In my own experience, they take the majority of the actual time so ignoring them seems to hide a lot of the actual cost and actual time difference versus not using such tools. #AI #criticalthinking #totalcostofownership #hiddencosts

12

5 Comments

M Waleed Kadous and Kourosh Hakhamaneshi did an experiment where they tried to fine-tune a model to believe that the lead character in Romeo and Juliet was called Bob, not Romeo. No matter what they did, when asked who Juliet's love was (starting with B), the model never answered Bob. Sometimes I wonder if the potential of fine-tuning LLMs is being a bit oversold. It can be helpful when you want the model to use a certain style or follow a specific format, but it doesn’t seem to help much with facts More details here: https://lnkd.in/g6p2eU-n

8