Dewey’s Post

For those who missed our deep dive into 'Evaluating LLM Models for Production Systems' during the last Data Phoenix webinar I hosted with Andrei Lopatenko 🇺🇦, here’s your chance to watch the full recording. Andrei, who has had an illustrious career impacting companies like Google, Apple, Walmart, eBay, and Zillow, brings a wealth of knowledge to this session. Key Highlights of the webinar: 🔸 In-Depth Analysis of LLM Evaluation Methods: Gain insights into a variety of methods to evaluate LLM models, understanding their strengths and weaknesses. 🔸 End-to-End Evaluation Techniques: Explore how LLM augmented systems are assessed from a holistic perspective. 🔸 Pragmatic Approach to System Deployment: Learn practical strategies for applying these evaluation techniques to systems intended for real-world application. 🔸 Focused Overview on Critical LLM Aspects: Receive an overview of various evaluation techniques that are essential for assessing the most crucial elements of modern LLM systems. 🔸 Simplifying the Evaluation Process: Understand how to streamline the evaluation process, making the work of LLM scientists more efficient and productive. 👉 Watch Now https://bit.ly/3PFL2im #DataPhoenix #LLMEvaluation #TechInnovation #IndustryLeaders #MachineLearning

Retrieval Augmented Generation (RAG) works with LLMs to improve accuracy by acting as a sleuth to find the most relevant data. Uncover the potential of the LLM/RAG symbiotic relationship in this Determined blog post.

Trust is binary, #LLM output is continuous. In an article by Berenice Baker, interviewing Beatriz Sanz Saiz, it brought it home that #largelanguagemodels (LLMs) are simply based on the key principle of choosing the next most probable word, given the words that came before it. Therefore, this introduces variability in the model's output. This variability is obviously beneficial, but it comes with a cost of affecting trust in the technology. How can we optimise a model on trust? 🚧 Should we make LLM more deterministic (like following a script), which removes the variability of the output? 🔎 Like in an academic paper, have the LLM output data sources it is used to generate the answer for the user to verify? 🛂 Should we utilise metrics during evaluation to penalise deviation from original data? Matthew Upson 📈 Sam Roberts Alex T. Alexander Laufer Alex Zaretsky, PhD, EMBA Merve Hickok Liam Sapsford Anna Felländer Josefin Rosén Michael Brent, PhD Jochem Huijps Article link: https://lnkd.in/e5DFPuWX Image created by Radi H.

In the digital age, we are barraged with information consistently. For the individuals who value the profound insights given by the New York Times (NYT), retaining and recalling explicit articles can be a test. Check out the whole article over here 👇👇👇 https://lnkd.in/dAB33NeN

As mathematicians and pie enthusiasts alike celebrate National Pi Day on March 14th (3/14), it's a fitting occasion for financial services professionals to reflect on the significance of this mathematical constant and its relevance to their industry. While the connection between pi and finance may not be immediately obvious, there are several parallels that highlight the importance of embracing complexity, precision, and adaptability in the world of financial services. Pi, often represented by the symbol π, is a mathematical constant representing the ratio of a circle's circumference to its diameter. It is an irrational number, meaning it cannot be expressed as a finite decimal or fraction, and its decimal representation extends infinitely without repeating. This infinite nature of pi reflects the infinite possibilities and complexities inherent in financial markets and economic systems. In the realm of financial services, professionals encounter a similar complexity and infinite array of variables. Just as pi represents the constant challenge of calculating precise measurements in geometry, financial professionals grapple with analyzing vast amounts of data, predicting market trends, and making informed decisions in an ever-changing landscape. Like mathematicians striving for ever more accurate approximations of pi, financial experts continually refine their models and strategies to navigate the complexities of global markets. Moreover, pi serves as a reminder of the importance of adaptability and resilience in the face of uncertainty. Financial markets, like the digits of pi, are non-repeating and non-terminating, characterized by unpredictability and volatility. In such an environment, successful professionals must possess the flexibility to adjust their approaches and strategies in response to shifting economic conditions and market dynamics. Furthermore, just as pi is fundamental to various branches of mathematics and sciences, financial services play a foundational role in driving economic growth and development. Whether it's facilitating investments, managing risk, or providing funding for businesses and individuals, the financial industry serves as the backbone of global commerce, connecting investors with capital and enabling entrepreneurial endeavors. On National Pi Day, financial services professionals can take the opportunity to reflect on the enduring lessons of pi: the importance of precision, adaptability, and embracing complexity. By recognizing the parallels between pi and their profession, professionals can reaffirm their commitment to mastering the intricacies of finance while remaining agile and resilient in the face of uncertainty. Just as the mathematical constant pi transcends disciplines and permeates the fabric of the universe, the principles it embodies resonate deeply in the world of finance, guiding professionals as they navigate the complexities of global markets and shape the future of the economy. #bankerslife

Collatz conjecture unsolved since 1937, when it was proposed by German Mathematician, Lothar Collatz: "This problem is simply stated, easily understood, and all too inviting. Just pick a number, any number: If the number is even, cut it in half; if it’s odd, triple it and add 1. Take that new number and repeat the process, again and again. If you keep this up, you’ll eventually get stuck in a loop. At least, that’s what we think will happen. Take 10 for example: 10 is even, so we cut it in half to get 5. Since 5 is odd, we triple it and add 1. Now we have 16, which is even, so we halve it to get 8, then halve that to get 4, then halve it again to get 2, and once more to get 1. Since 1 is odd, we triple it and add 1. Now we’re back at 4, and we know where this goes: 4 goes to 2 which goes to 1 which goes to 4, and so on. We’re stuck in a loop. Or try 11: It’s odd, so we triple it and add 1. Now we have 34, which is even, so we halve it to get 17, triple that and add 1 to get 52, halve that to get 26 and again to get 13, triple that and add 1 to get 40, halve that to get 20, then 10, then 5, triple that and add 1 to get 16, and halve that to get 8, then 4, 2 and 1. And we’re stuck in the loop again"(Quanta Magazine, Patrick Honner, Sept 2020). https://lnkd.in/erR9QXAp, (Veritasium, Derek Alexander Muller, July 2021) References: Quanta Magazine, Patrick Honner, Sept 2020 - https://lnkd.in/eCYg4kHX Veritasium, Derek Alexander Muller, July 2021 - https://lnkd.in/eKfVbrEm

What can we learn from Woodrow Wilson and Henry Ford's decision-making? In 1922, mathematician Lewis Fry Richardson published the paper "Weather Prediction by Numerical Processes." He demonstrated that extensive data and calculations could predict the following day's weather significantly. A groundbreaking concept at the time. Reflecting on this, it's incredible to think about how leaders like Woodrow Wilson and Henry Ford made critical decisions a century ago without relying on studies and statistics. Today, there is a study of every problem in the world and every decision we make. However, not all problems are easy to measure. That doesn't make them unimportant. If there is room for interpretation between an empirical indication and the absolute truth, it is essential to acknowledge that. Don't panic. Wilson and Ford didn't have data either. Clearly, if you have valid data, you should base your decisions on it. I've also seen Moneyball. If the decision isn't very impactful, I fill the gap between empirical indication and truth with something that makes me happy. In that sense, I wish you a sunny day tomorrow!

Data permeates almost every aspect of our lives, but do we truly understand its effects? In this edition of Author Talks, Aram Sinnreich, co-author of 'The Secret Life of Data: Navigating Hype and Uncertainty in the Age of Algorithmic Surveillance,' explores data’s impact on our culture, the power of manipulated algorithms, and more. ➡️ https://mck.co/3SJO3jd

Interested in the FAIR principles of mathematical research data? Here is a recent paper https://lnkd.in/gQpBQ6qc discussing in particular our awesome database www.FindStat.org of combinatorial statistics and maps, originally build by Chris Berg and me and currently run by Martin Rubey and me.

My keynote presentation is not yet up on the conference website, so in the interim, for all who have asked you may download it from ssrn https://lnkd.in/eYtBKCBp and/or my website https://lnkd.in/H7KU_- where you can also download the talking points/writeup for my conference RoundTable session addressing the hype, reality, and overlap of causal modeling vs association-based modeling All feedback more than welcome! Best, JD (full disclaimer: I am the sole author of this work, which does not necessarily represent the opinions of any institution, etc., etc.)