Alexandre Kateb, CFA’s Post

🚀 100 Days of Tech: Day 4 - Use of Probability Distributions in NLP To deepen my understanding of LLMs, I took a step back to revisit the foundational concepts of statistics and probability. I emphsaized on probabilty distribution and started using it to do sentiment analysis on a article from Bloomberg Businessweek titled 'How AI Replaces the Metaverse as Zuckerberg's Top Priority' by Aisha Counts and Sarah Frier. My objective was to apply various probability distributions to dissect the text and uncover the embedded statistical narratives. 📊 The insights: Anomaly Words: underscoring significant emphasis on terms such as 'AI', 'Zuckerberg', and 'Meta'. Bernoulli Sentiment Trials: Spectrum of sentiments, sketching a vivid emotional landscape of the article. Log-Normal Distribution Parameters: Enriched content strategy comprehension. Binomial Sentiment Analysis: Indicated that the aggregate sentiment of the article might deviate from a neutral stance, tilting more towards one polarity 🧐 But the plot thickens! The article presented an exact balance in sentiment: 58 positive sentences out of 116, landing precisely at a 50% positive rate. This statistical equilibrium earmarks the article's sentiment direction as neutral. #100daysoflearning #statistics #probability #lookingforopportunities #nlp

🌞Summer 23'(1/3): This summer, backed by the Dill Fund and Wabash College, I delved into the financial behaviors of the Indian community across Brampton, Mississauga, and Toronto🇨🇦🇮🇳. My research combined NLP and econometrics, but it was the real-life stories that truly enriched it. Key takeaway? Leveraging Natural Language Processing alongside SARIMA models significantly boosted accuracy. Through sentiment analysis integration, I slashed the forecasting error from 62.96% to a mere 6.25%. 🔗 https://lnkd.in/gqZ2RnBW #Research #NLP #Econometrics #IndianDiaspora

🚀 Thrilled to share a breakthrough from the University of California, San Diego - Rady School of Management: Our AI-Driven Equity Research Assistant is here to overcome the limitations of conventional NLP tools and transform equity research. Developed alongside my talented colleague Khushi Bet, this project harnesses the synergy of LangChain, OpenAI, and Streamlit. 🌟 Gone are the days of wrestling with the constraints of lengthy articles and the 3000-word limit in single interactions. Our solution is built to handle extensive textual data, offering deeper and broader analyses without the need for multiple interactions. 📚💻 This isn't just about improving efficiency; it's about redefining the possibilities of AI-driven tools in financial analysis by accommodating more complex data inputs and leveraging a wider knowledge base for comprehensive insights. 🚀 Special thanks to Professor Jingbo Shang for his unwavering support and Khushi Bet for her incredible partnership. Check out a glimpse of the tool in action in the image attached! #AI #NLP #Finance #Innovation #DataAnalysis #UCSD #RadySchoolOfManagement #OpenAI #Streamlit #LangChain

Excited to share our latest work: "the development of a new approach to protein sequence generation using a Small Scale Protein Language Model (SS-pLM)" led by Alexis Molina. Let me know if you want to know more in detail. #artificialintelligence #drugdiscovery #proteinengineering #drugdesign #biophysics #biotechnology

📚 Thrilled to share my latest blog post on 𝑹𝒆𝒗𝒊𝒆𝒘 — 𝑩𝑰𝑮-𝒃𝒆𝒏𝒄𝒉: 𝑩𝒆𝒚𝒐𝒏𝒅 𝒕𝒉𝒆 𝑰𝒎𝒊𝒕𝒂𝒕𝒊𝒐𝒏 𝑮𝒂𝒎𝒆: 𝑸𝒖𝒂𝒏𝒕𝒊𝒇𝒚𝒊𝒏𝒈 𝒂𝒏𝒅 𝒆𝒙𝒕𝒓𝒂𝒑𝒐𝒍𝒂𝒕𝒊𝒏𝒈 𝒕𝒉𝒆 𝒄𝒂𝒑𝒂𝒃𝒊𝒍𝒊𝒕𝒊𝒆𝒔 𝒐𝒇 𝒍𝒂𝒏𝒈𝒖𝒂𝒈𝒆 𝒎𝒐𝒅𝒆𝒍𝒔 on Medium! 🚀 I invite you to take a quick 5-minute read on this review of Google BIG-Bench. Your thoughts and feedback are always appreciated and If you find it valuable, a round of applause on my blog post would also be greatly appreciated. Happy learning guys!🧑💻 #google #nlp #llms #benchmark #glue #ai https://lnkd.in/gdm4DcwE

"📚 Exciting New Research Alert by Cohere! 🚀 🔍 In this groundbreaking paper, the authors introduce a novel method called Panel of LLM Evaluators (PoLL) that not only effectively evaluates the performance of Language Models (LLMs), but also addresses issues like intra-model bias, latency, and cost. 🌟 🔬 The study demonstrates that there is no single 'best' judge across all settings, highlighting the need for a diverse panel of evaluators. PoLL consistently outperforms a single large model, making it a promising alternative for LLM evaluation. 💡 🌐 While the research focused on three specific evaluator settings and a limited number of judges and panel compositions, it opens up exciting possibilities for future work. The method's applicability in math or reasoning evaluations, where language models often struggle, is an intriguing avenue for exploration. 🧠 🔎 Additionally, the paper leaves room for further investigation into 'panel selection', which involves identifying the best models to include in PoLL based on quality and cost. This aspect presents an exciting opportunity for future research. 📊 🙌 Let's give a big shoutout to the talented team at Cohere and the authors for their exceptional work! 👏 Special thanks to Maxime Voisin, Victoria Spooner, Priyanka Sen, and the entire team for their contributions. 🤝 📚 If you're interested in the future of language models and their evaluation, this paper is a must-read! 📖 Stay tuned for more updates in this exciting field. 🌐 #Research #LanguageModels #Evaluation #Innovation #AI #NLP"

🚀 Transforming Mathematics with Language Models: A Glimpse into the Future 🧮 As we delve into the fascinating world of mathematical language models, a recent survey from East China Normal University offers groundbreaking insights. This comprehensive study unveils how Pre-trained and Large-scale Language Models are reshaping the landscape of mathematics. 🔍 Key Takeaways: Innovative Models: From GPT-4 to specialized MLMs, we're witnessing a paradigm shift in mathematical reasoning and problem-solving. Diverse Applications: These models are not just about calculations; they're revolutionizing theorem proving and complex problem-solving. Future Prospects: The survey points to exciting future directions, despite challenges like data scarcity and multi-modal reasoning. 🌐 Implications for AI and Math Enthusiasts: This isn't just about numbers; it's about how AI is unlocking new dimensions in understanding and applying mathematics. A must-read for anyone passionate about the intersection of AI, NLP, and mathematics! 🔗 Read the Full Survey: https://lnkd.in/dnJQefZ4 #AILanguageModels #Mathematics #Innovation #DataScience #NLP

Understanding Lemmatization in LLMs vs. Stemming 🌿 "Lemmatization" and "stemming" are foundational techniques in text normalization, ensuring that words are compared on a like-for-like basis. 🔍 Lemmatization: 1. Lemmatization reduces a word to its base or dictionary form. For example, "running" becomes "run", and "better" becomes "good". 2. It's a sophisticated process that often involves morphological analysis and lookup tables. 3. In LLMs, accurate lemmatization leads to better comprehension and more meaningful responses because the model grasps the underlying concept of a word beyond its form. 🌱 Stemming: 1. Stemming crudely removes prefixes and/or suffixes from a word. So, "running" might become "run", while "happiness" could turn into "happi". 2. It’s faster and uses less computational power but might produce stems that don't have clear meaning. 3. In scenarios where speed is vital and precision is secondary, stemming might be the choice. In LLMs, a profound understanding of language is paramount. Lemmatization, with its precision, often proves more beneficial, allowing the model to capture nuances. However, based on the task and available computational resources, stemming can be the efficient choice. #NLP #TextProcessing #LLMs #OpenAI