ReflexAI Solves Knowledge Base Problems

New piece from me on ReflexAI's Intelligence Layer: how we turned sprawling, messy, sometimes-contradictory organizational knowledge into unified context. When customers connect their knowledge bases, every training simulation, scoring dimension, and insight in the ReflexAI platform is built on what's in them. Six design principles guided how we built it 👇 Built by my team, who I'm enormously proud of: Jose Fajardo, Jafet S., Rajat Paliwal, Giulia Falcão, Pedro Henrique Gomes Venturott, Christian Picón Calderón

An agent gives the right answer and gets marked down for it. The policy changed. The agent knew. The QA rubric was built off the old version, still sitting in the same folder as the new one, same naming, nothing marking which is live. Do that a few times and the agent stops opening the knowledge base at all. Search does not fix this. You can rank every document perfectly and still be confidently wrong, because ranking has no opinion about what is currently true. This is the problem my team and I have been working in ReflexAI Intelligence Layer, and it is definitely one of the best problem I have had on my desk. It sits exactly where machine learning stops being a demo and starts being something a person leans on mid call. More details about this work can be found in the article wrote by our Head of AI, Taki Hasegawa! 🚀

💡 Big thanks to James Welsh for his session exposing the biases we are all subject to which can obscure our ability to think critically about a problem. 💭 🧠 Fascinating to understand more about how humans naturally behave - and how we can adopt strategies to ensure that we are taking a sound decision based on all the evidence ⚖️ : a skill which is becoming increasingly important when assessing the output of AI. 🗒️ And here’s a little test you can join in with at home… shamelessly taken from the session. ✏️ I am thinking of a rule which connects a group of 3 numbers. Not all groups of 3 numbers will fit my rule. An example of a group of 3 numbers which fits my rule is 2,4,6. Your challenge is to work out what my rule is, by giving me other groups of 3 numbers and asking me whether or not they fit my rule. Go…

I have a 2'fer for you today. Both start with the same idea, what is the end game with AI and recursive intelligence? The first article explores the the recent announcement of 4 of Googles best minds on AI leaving to form a new company. Jeff Dean, Sanjay Ghemawat, Quoc Le and Oriol Vinyals left Google to form a company called DiscoveryLoop. Discovery Loop is working on recursive intelligence, by which I mean the ability of AI to actually write its own next versions, continuously improving with each iteration. Where does it end? https://lnkd.in/e24cH5wB I found this whole notion kind of mind blowing actually and dug in deeper and wrote a full on report on what is the end game for AI and intelligence. You can download the full report here: https://lnkd.in/e6mrg_mu

This is the 2/6 post on my new Substack from the recent learnings on analyzing the trends in AI from academic data: https://lnkd.in/dQ-6fNDE As always, there is a short 3-minute version and a more detailed one with all the references. Again, although the article was written with the help of AI, a lot of behind-the-scenes work went into analyzing the papers, filtering them through my personal point of view, and shaping them with my work in the field.

Clients, sometimes we write to convey the one thing we wish each one of you knew right now. This week's blog relates the similarities between the AI boom and the Internet Bubble of 2000 - and the astonishing aftermath of that bust. You might take three minutes to read this: https://lnkd.in/gi-m8k3T

https://buff.ly/iSfLjFH A recent Question of the Week from our free blog explores how EBV, Evidence Based Valuation© is applied technology. The modern - today - version comprises AI, Data Science, and micro-economic theory. So let us consider “evidence” as it works with analytic AI.

In this video, you’ll learn how to build a Corrective RAG (CRAG) system from scratch and understand how it improves traditional Retrieval-Augmented Generation (RAG). You’ll also learn how Corrective RAG connects with Agentic RAG and Agentic AI, where the system can evaluate retrieved information, make decisions, correct retrieval mistakes, and improve the final response.

Most enterprises think they're on rung 3 of the agentic maturity ladder. They're on rung 1. "We have an AI strategy" is not the same sentence as "we have governed AI in production." In part three of the Agentic Adaptation Playbook series, Kevin Paige lays out the five-rung ladder, from shadow AI to compounding, and the four evidence questions that tell you where you actually stand. Find your rung: https://okt.to/RJTwCK

Most enterprises think they're on rung 3 of the agentic maturity ladder. They're on rung 1. "We have an AI strategy" is not the same sentence as "we have governed AI in production." In part three of the Agentic Adaptation Playbook series, Kevin Paige lays out the five-rung ladder, from shadow AI to compounding, and the four evidence questions that tell you where you actually stand. Find your rung: https://okt.to/tKlLr5