Dr. Ari Zelmanow’s Post

Why has research become impotent? Here's one reason 👇 I have come to realize that research teams are focused less on adding value... And more on providing validation. Rather than being asked to test hypotheses... They are being asked to defend them. They have become the defenders of cherry-picked data that supports a narrative... Rather than the interrogators of data, helping to inform a narrative. So I ask... What the hell are we doing here? ~fin~ ===== 👋 I'm Ari... I write a short email to 700+ people every Saturday on how to leverage Customer Forensics™ to make better growth bets by: 🔎 Collecting and interrogating evidence like a detective 🎤 Reporting insights like a primetime news anchor ⚖️ Building business cases like a trial attorney Want me to send it to you? Link in the comments below 👇

"What I’d like is an easy-to-use system that gives good performance for both seen and unseen codes, with low overfitting risk." In a new practical deep dive, Valerie Carey experiments with different methods for encoding hierarchical categoricals.

"It all adds up: every $1 of fraud is estimated to actually cost a company close to $5; a nearly 16% increase in cost since 2020...NeuroID customers are finding that behavioral analytics offers a whole new way to operationalize their fraud-fighting." #behavioralanalytics #conversion #fraudprevention #digitalcustomerexperience

Pro-tip for looking at research impact 💥 Don't look to business outcomes to evaluate the effectiveness of your research efforts. Look at the quality of the decisions made. Here's why: • A great decision could lead to a sh*tty outcome • A bad decision could lead to an epic outcome The outcomes aren't only a result of the decisions you make, but the conditions you make them in. 𝗢𝘂𝘁𝗰𝗼𝗺𝗲𝘀 = 𝗔𝗰𝘁𝗶𝗼𝗻𝘀 + 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 You can control the quality of the decisions, but you can't control the myriad of things that impact the outcome. ~fin~ ===== 👋 I'm Ari... I write a short email to 700+ people every Saturday on how to leverage Customer Forensics™ to make better growth bets by: 🔎 Collecting and interrogating evidence like a detective 🎤 Reporting insights like a primetime news anchor ⚖️ Building business cases like a trial attorney Want me to send it to you? Link in the comments below 👇

Don't let clients make life difficult! Learn once and for all how to get client data on a 24/7 basis. And it is not rocket science! Registration link in top comment below!

Another great read from one of our amazing MLEs, Melanie Chen. In this article, she highlights how ML and data teams can utilize a real-time feature store like Chalk to power fraud and risk models in production. Check out a “Feature Store at Work: A tutorial on Fraud and Risk” on our blog! Link in comments

🤔 While going through documents to create our supervised fine-tuning (SFT) and direct preference optimization (DPO) data, I came across this clause. At first, I was 100% convinced that it was an LLM-generated contract. This is a classic example of what happens when next-token generation gets stuck in a loop. Then I realized the doc was from 2000. Time machine? 🚀🕢 Or more likely, someone accidentally hit ctrl+v twice. In reality, it's just another example of what's really happening every day, even today in 2024. So much of what we do is mindless text generation. Even worse, much of this text is written at the end of a very long day (or week...or month). Fatigue and boredom end up creating an environment where even the best humans have uncomfortably high error rates. That's where our contract-specific sampling techniques are going to help. 😁

Diving into the future of AI-driven analytics with Invenci. Discover how data transforms decisions in the business world! https://invenci.com/ #DataScience #BusinessIntelligence

We have just published an insightful blog post discussing the rise of AI-generated voice fraud and its impact on individuals. Learn about the use of synthetic voices to deceive acquaintances and family members for financial gain. Find out more by clicking on the following link: https://ift.tt/kj9s2F1

Anomaly detection methods are of great importance in the data government process. In this post I show an example of it