Harshit Tyagi

Ran a quick test on the new #OpenAI #o1mini to solve IIT-JEE Advanced 2024 maths questions. It accurately solved 13 out of 17 questions with an impressive accuracy of ~76%. Quick thoughts: 1. The hidden chain of thought is quite detailed and extensive. Breaks the problem, solves and verifies and iterates multiple times. This might increase the output token cost by a big margin. 2. Seems there are tools integrated for calculations. Much needed as #gpt4o does a bad job at it. 3. Latency is quite high. Some questions took more than 50 seconds to get answered. This needs to come down for production user-facing use cases. 4. Hallucination still exists. Check the 1st question in the link, it had the correct step-by-step reasoning but hallucinated at generating the final answer. 5. Overdoing self-thought. Saw a couple of instances where it did generate correct answers in the process but eventually made wrong answers. Would give a clearer picture once the rate limits are increased and benchmarking is done on a bigger sample. Let me know any interesting findings and thoughts 😀 https://lnkd.in/giSCezyG #GenAI #OpenAI #PhysicsWallah

139

5 Comments

🎉 Great News! 🎉 I'm thrilled to share that our(Animesh Mukherjee, Paramita Das and Ritabrata Chakraborty) paper "On the effective transfer of knowledge from English to Hindi Wikipedia" has been accepted at the COLING 2025 (Industry Track) 🌟! In this work, we propose a lightweight framework that enriches Hindi Wikipedia articles by extracting and adapting updated information from English sources, ensuring alignment with Wikipedia’s style and policies, and then translating the adapted content into Hindi to significantly improve the coverage and quality of these articles. Key Contributions include: 1️⃣ We propose a multistage approach to extract knowledge from biographical writings about popular figures, transform this text to adhere to the NPOV guidelines and incorporate it into low-resource Wikipedia articles, e.g., Hindi. 2️⃣ We manually curate 103 biographical writings relevant to corresponding Wikipedia articles as external knowledge sources, which can serve as a rich source of factual information 3️⃣ Our lightweight framework efficiently updates Hindi Wikipedia articles by adding coherent and new information. 📄 The preprint is available on ArXiv: https://lnkd.in/dQ2axSx4 We are thrilled to present this work at COLING 𝟮𝟬𝟮𝟱 and look forward to connecting with the AI research community 🌐🤝.

64

4 Comments

Shahrukh Khan is recognized for his wit, humor, and charm. In this context, we've fine-tuned LLamav2 to respond briefly, in a crisp, witty, and humorous manner, akin to SRK himself. We've named it SRKGPT. For detailed insights into the fine-tuning process and the creation of the dataset to emulate SRK's style, please refer to the video in my channel for a comprehensive understanding. Video Link: youtu.be/gYPwx0DR7zc  Channel Link: youtube.com/@datatrek #datatrek #datascience #machinelearning #statistics #deeplearning #ai

18

1 Comment

Shahrukh Khan is recognized for his wit, humor, and charm. In this context, we've fine-tuned LLamav2 to respond briefly, in a crisp, witty, and humorous manner, akin to SRK himself. We've named it SRKGPT. For detailed insights into the fine-tuning process and the creation of the dataset to emulate SRK's style, please refer to the video in my channel for a comprehensive understanding. Video Link: youtu.be/gYPwx0DR7zc  Channel Link: youtube.com/@datatrek #datatrek #datascience #machinelearning #statistics #deeplearning #ai

10

1 Comment

NLP (LLM) Research Roadmap I’m happy to be an NLP research mentor for - PhD candidates without a good faculty supervisor - Industry professionals who want to gain research experience by publishing papers. Guidance will be provided via 1-to-1 paid Google Meet sessions. Interested people can DM me with all the relevant details. "Hi" or "incomplete messages" will be ignored due to time constraints. NLP research involves various phases like - Research Foundation (learning pre-requisites) - Research Papers Reading - Research Problem Identification - Research Gaps Identification - Research Idea Formulation - Research Idea Implementation - Research Paper Writing - Research Paper Submission and Review A good faculty supervisor provides proper guidance in every phase of NLP research. Unfortunately, some faculty supervisors will never do this which makes the journey of PhD candidates difficult. Due to a lack of proper guidance from the faculty supervisor, some of the PhD candidates despite being highly talented finish their PhD with average or below-average papers even after spending more than five years. #llms #nlproc #india #phdresearch #phd

255

5 Comments

Perhaps the most critical thing in a Retrieval Augmented Generation (RAG) system is the retrieval or searching strategy. Poor retrieval will lead to irrelevant documents going to the LLM leading to irrelevant generated answers or worse hallucinations. Is cosine similarity everything? Do check out below illustration which talks about four of the most common strategies for playing around with retrieval strategies in RAG systems: 1. Similarity-based retrieval: This is based on embedding similarity, you can play around with different metrics like cosine, maximal margin relevance etc. 2. Similarity with Threshold retrieval: Useful where we cap the similarity threshold and can help in removing irrelevant document but can drop recall 3. Multi-query retrieval: Make the LLM generate variations of your query and get documents based on these multi-versions of your initial query, often useful to get more recall or coverage of documents 4. Reranker retrieval: Use a model specialised in ranking documents based on search relevance Often you can combine some of these techniques and even use ensemble retrievers. Do explore them and go beyond just cosine similarity!

17

🚨ACL 2024 Outstanding Paper Award Alert 🚨 ➡️Paper Title: INDICLLMSUITE: A BLUEPRINT FOR CREATING PRE-TRAINING AND FINE-TUNING DATASETS FOR INDIAN LANGUAGES 🌟Few pointers from the paper 🎯Despite the considerable advancements in English LLMs, the progress in building comparable models for other languages has been hindered due to the scarcity of tailored resources. 👁In this paper Authors work aims to bridge this divide by introducing an expansive suite of resources specifically designed for the development of Indic LLMs, covering 22 languages, containing a total of 251B tokens and 74.8M instruction-response pairs. 🎯Recognizing the importance of both data quality and quantity, their approach combines highly curated manually verified data, unverified yet valuable data, and synthetic data. 🎯They built a clean, open-source pipeline for curating pre-training data from diverse sources, including websites, PDFs, and videos, incorporating best practices for crawling, cleaning, flagging, and deduplication. 🎯For instruction-fine tuning, they amalgamate existing Indic datasets, translate/transliterate English datasets into Indian languages, and utilize LLaMa2 and Mixtral models to create conversations grounded in articles from Indian Wikipedia and Wikihow. 🎯Additionally, they addressed toxicity alignment by generating toxic prompts for multiple scenarios and then generated non-toxic responses by feeding these toxic prompts to an aligned LLaMa2 model. 🎯 They hope that the datasets, tools, and resources released as a part of this work will not only propel the research and development of Indic LLMs but also establish an open-source blueprint for extending such efforts to other languages. 🏢Organization: 🛎️Nilekani Centre at AI4Bhārat🛎️ Indian Institute of Technology, Madras 🛎️Microsoft🛎️Sant Longowal Institute of Engineering and Technology 🛎️Indian Institute of Information Technology Design & Manufacturing Kancheepuram, 🛎️Maharaja Agrasen Institute Of Technology, Delhi 🛎️Sarvam ,🛎️ NICT 🧙Paper Authors: Mohammed Safi Ur Rahman Khan, Priyam Mehta , Ananth Sankar , Umashankar Kumaravelan, Sumanth Doddapaneni, Suriyaprasaad G, Varun Balan, Sparsh Jain, Anoop Kunchukuttan, Pratyush Kumar, Raj Dabre, Mitesh Khapra 1️⃣Read the Full Paper here: https://lnkd.in/gqwQST-r 2️⃣Code: https://lnkd.in/giWPVntx 🎉 🎉  Congratulations to all the authors 🎉 🎉 Find this Valuable 💎 ? ♻️REPOST and teach your network something new Follow me 👣, Naveen Manwani, for the latest updates on Tech and AI-related news, insightful research papers, and exciting announcements. #ACL2024 #Indicllms #llms #llama2 #mixtral

15

🚨 𝗘𝘃𝗲𝗿𝘆𝗼𝗻𝗲 𝗶𝘀 𝗮𝗽𝗽𝗹𝗮𝘂𝗱𝗶𝗻𝗴 Blinkit's 𝗮𝗺𝗯𝘂𝗹𝗮𝗻𝗰𝗲 𝘀𝗲𝗿𝘃𝗶𝗰𝗲, 𝗯𝘂𝘁 𝗵𝗲𝗿𝗲'𝘀 𝘄𝗵𝗮𝘁 𝘁𝗵𝗲𝘆'𝗿𝗲 𝗻𝗼𝘁 𝘁𝗮𝗹𝗸𝗶𝗻𝗴 𝗮𝗯𝗼𝘂𝘁... While everyone's focused on "𝟭𝟬-𝗺𝗶𝗻𝘂𝘁𝗲 𝗮𝗺𝗯𝘂𝗹𝗮𝗻𝗰𝗲𝘀", they're missing the bigger revolution happening here. Here's what most people don't realize: The real innovation isn't the speed it's the complete redesign of emergency response: 1. Traditional ambulances aren't slow (10-30 mins average varies case by case), but they waste critical minutes in the dispatch process itself:🤔 - "What's your location?" - "Can you describe the emergency?" - "Please stay on the line…" - "I cannot find your location" In the golden hour where every minute counts, this back & forth just spoils everything.😔 2. Blinkit's masterstroke🙌: - The same app you order groceries from - One-click - Location already shared - Ambulance dispatched instantly -Arrived at your location like that delivery partner But here's the genius part no one's talking about:🤫 • Real-time tracking of emergencies • Crowd Source the information of the accident on the road for immediate response • Immediate help for gig workers themselves in case of need They're not just launching ambulances they're building India's largest emergency response network. Think about it: When did a quick commerce company last think about a saving initiative instead of just minting revenue? What makes me proud is that this isn't just another profit-driven venture. They are focusing on affordability (current traditional ambulance costs can go up to ₹2,500 with surge) Whats your most used Quick commerce app do you wish to see an emergency button inside it? #Innovation #Healthcare #Technology 

65

5 Comments

In my early learning phase, I often confused Layer Normalization and Batch Normalization—I knew they improved numerical stability, but the key differences eluded me. Watching Umar Jamil's videos clarified everything: Batch Normalization works across the batch dimension, ideal for CNNs, while Layer Normalization focuses on features, perfect for RNNs and Transformers. Grateful for such insightful resources! 🙌 #deeplearning #machineLearning #GENAI #Transformer

10

🚀 Creating IIT-JEE Question Paper using AI Agents The recent release of JEE Advanced results took me back to the day when I nervously checked my own scores. As a fun experiment, I wondered: could I use Generative AI to create an entire JEE question paper from scratch? So, I created AI agents to generate high-quality JEE questions. Here's how I built it: Simply prompting LLMs to give good questions didn't work well because the quality and difficulty level of the questions were inconsistent. I used the CrewAI agent framework and GPT-4o model for this exercise. The process involves two key agents: 1️⃣ The Question Creator Agent crafts thought-provoking questions based on specific topics from the IIT-JEE syllabus. 2️⃣ The Question Reviewer Agent meticulously reviews each question, providing feedback and suggestions to refine them. These agents converse among themselves to generate high-quality, challenging questions that test the analytical skills and conceptual understanding of JEE aspirants. See the video below for a tutorial. I created a detailed Python notebook to show you how you can do it too! (Link in comments) —— Want to learn how to build AI apps? Join me for a free, 1-hour “Function Calling with LLMs” workshop on Friday, where we will explore how to add superpowers to LLMs. Register here: https://lnkd.in/gJwkcJXg #AIAgents #GPT4o #BuildFastwithAI #FunctionCalling #FreeWorkshop 

312

11 Comments

The company's search boss, Prabhakar Raghavan, told employees at an all-hands meeting in March that Google plans to build teams closer to users in key markets, including India and Brazil, where labor is cheaper than in the U.S. Seems like SkillsFuture is leveling up in India and Brazil; also means the Commoditization of typical "Core" roles could be performed beyond the US, with much lower cost. Now, my question, what can our next-generation Singaporean fresh graduates do, than those countries mentioned yet to? Also goes to many staff with the business-as-usual SG&A roles, how to keep them from moving into Thailand and Malaysia?

52

23 Comments

Looking for recommendations for good hindi-english (hinglish) code-mixed speech to text models/corpuses. Anyone building anything with speech in the Indian context probably realized normal ASR systems fall flat, losing >50% context because of their inability to handle hinglish conversation (which is extremely common for a huge chunk of modern India). See the title of the first video when looking for shark tank. Once you start noticing it, its hard to unnotice. If speech to text doesn't work, every downstream task thas cascading failures. I'm not looking to build anything commercially so hopefully #research corpuses will also be okay. Literally only want to progress speech for India and this is a major blocker. This is one of those areas where other countries can't innovate for us and hand us a solution unfortunately. #speech #asr #codemix #hinglish #stt

3

1 Comment

Something I've written on how a T20 batting innings can be assessed. This framework can be extended to assessment of individual players as well, which I will do in the susequent posts. For now, please feel free to comment and if you like the content, please subscribe. #cricket #cricketanalytics #data #dashboards #datavisualization #sportsanalytics https://lnkd.in/gb7T-UzJ

28

6 Comments

If you are attending ICPR 2024 in the next 5 days (Dec 01-05), then do visit our poster titled: "waLLMartCache: A Distributed, Multi-Tenant and Enhanced Semantic Caching System for LLMs”; my co-author Binay Gupta will be presenting it. Our system waLLMartCache advances the GPTCache framework by incorporating the following features: 1. We introduce the support for a new database Redis in GPTCache – this is used as L2 cache in our designed system. 2. Currently, GPTCache is implemented to be run on a single node which we enhance to span across multiple nodes to handle industry-scale requests and consequently, we also designed a distributed eviction manager. 3. We further create partitions for individual tenants (clients) so that these can be hosted together while maintaining semantic separations. 4. We present a decision engine (i.e., an enhanced semantic similarity checker) that decides whether to cache an LLM response based on retail business use-cases. 5. We showcase that loading FAQs (which can be set to be stored persistently in the memory) while booting the LLM cache can be a simple yet effective strategy to boost cache hits significantly. Some of these contributions have already been merged with the open-source GPTCache code. A big shout out to my other co-authors: Soumik Dasgupta, Anurag Wagh, Lalitdutt Parsai, Geet Vudata, Shally Sangal, Sohom Majumdar, Hema Rajesh, Anirban Chatterjee. #GenAI #LLMs #MachineLearning #ArtificialIntelligence #Industry #Research #Conference #Paper

116

3 Comments

🤔 Ever wondered why your NumPy calculations might be off? Swayam Singh, a 2024 summer intern at #QuansightLabs', does a thorough analysis of the limitations of np.longdouble and introduces numpy_quaddtype as an innovative alternative. Read all about it in his blog: 👉 https://lnkd.in/d3dJTJX7 #Python #NumPy #ScientificComputing

5

Just wrapped up an awesome "Deep Dive into Deep Learning course"! 🎉💻 Super stoked to have delved into the nitty-gritty of neural networks, convolutions, and all things AI. Grateful for the opportunity to deepen my expertise in this rapidly evolving field. #DeepLearning #AI #CertificateCompletion #Scaler

11

1 Comment

Should the current placement situation in IIT's be an excuse for T2-T3 colleges to simply shake off their responsibility of placing their students ? The job market has been absolutely shit, and there's no shame in admitting it. There's no shame in reaching out to your alumni network to ensure these emerging leaders get opportunities to flourish in life. What's shameful, however, is when you don't take cognizance of the matter and do nothing, just because asking for help can label you as WEAK in front of many. A humble request to many placement cells out there, this is not the time to exhibit your high handedness over students ; this is the time to work together to ensure people get the opportunities they deserve. This is the time to COLLABORATE. P.S if anyone is looking to get into data analytics, I have a few leads. DM me and I'll share those with you. Praying for better times ahead 🙏🏻

69

15 Comments

Last week, Maxim Nurgaliev from Next Step Fusion delivered an excellent presentation during the Plasma Control Group meeting on the recent experiments conducted by our UC San Diego and Next Step Fusion teams on the DIII-D National Fusion Facility. Even during challenging piggyback experiments, where the discharges were not ideal and unexpected H-to-L mode transitions occurred, the Reinforcement Learning module integrated into the DIII-D Plasma Control System performed remarkably well, exceeding our expectations in holding plasma shape. These achievements demonstrate the robustness of the controller, and we hope that more researchers in the DIII-D program will adopt this tool for their experiments. Next Step Fusion will be presenting these results at the DIII-D Science, Engineering, and Technology meeting in January. This collaboration exemplifies the kind of public-private partnerships envisioned and supported by the U.S. Department of Energy (DOE) Office of Fusion Energy Sciences, fostering innovation and progress. With tools like these, we are advancing the Technology Readiness Level (TRL) of plasma control systems for future Fusion Pilot Plants (FPP). #FusionEnergy #PlasmaControl #ReinforcementLearning #NextStepFusion #DIIID #PublicPrivatePartnership #FusionPilotPlant #TechnologyReadinessLevel #DOE #PlasmaPhysics Richard Buttery Scott Hsu Matthew Lanctot Michael Halfmoon Aleksei Zolotarev Jean Paul Allain

23

1 Comment