Hian Goh’s Post

A couple of quick random musings on Intelligence From a high level, intelligence could be seen as the culminating fusion of the ability to understand and to solve problems. Think of one hemisphere of "intelligence" as a concept entity as being about understanding (processing), and the other about solving problems (outputs). When we started on the project of solving intelligence - and here, I am referring to the good old-fashioned AI - we assumed we must first get machines to understand our worlds deeply if we want them to simulate intelligence, if we want them to solve problems that might necessitate intelligence. Turns out we were wrong, or at least partially, since as LLM's prove, you can get a long way towards simulating intelligence and solving human-cognition-level problems with a limited "true understanding". It is an intriguing subversion of our expectations, and another case in point of how off our intuitions can be, even about something as innate to us as intelligence. Another slightly related note is that the fact that we were able to go from zero ability to compute to relatively fully-fledged, relatively human-level intelligence in the better part of a century probably points to intelligence as being not all that special or unique or hard-to-attain for a species, be it by artifical design or by natural design. And this probably should make us adjust our priors for the Drake equation for the variable "presence of intelligence in a species". This means either way, with the progress of the last few years, our likelihood of meeting a new form of intelligence, either home-grown or alien, has gone up by orders of magnitude.

📚 Just finished my first book of 2024: "Superintelligence: Paths, Dangers, Strategies" by Nick Bostrom. A thought-provoking read that delves into the future of AI and its potential impact on humanity. Bostrom explores how superintelligence might be achieved and the existential risks it poses. His insights into the AI control problem and strategies to align superintelligence with human values are particularly compelling. 🤖 What struck me most was the analogy of the "Unfinished Fable of the Sparrows" which mirrors our current race towards advanced AI without fully considering the consequences. The book, although speculative, presents a fascinating blend of philosophy, technology, and future studies, making it an essential read for anyone interested in the trajectory of artificial intelligence. 🔍 While some may find Bostrom's predictions far-fetched, there's no denying the importance of the questions he raises. It's a must-read for tech enthusiasts, philosophers, and anyone curious about the future of AI. https://lnkd.in/gCv57XmQ 🌐 #AI #Superintelligence #FutureTech #ArtificialIntelligence #Philosophy #TechReads #2024Books

Our Managing Director, Tavonga M., recently sat down for a thought-provoking interview on 'The Deep Dive Sessions' discussing the 'AI Revolution'. In this eye-opening interview titled 'AI Revolution, Adapt Or Die,' Tavonga explores the world of artificial intelligence and its implications. Don't miss out – watch the full interview on YouTube today!  https://lnkd.in/dTkERrww #AIRevolution #ArtificialIntelligence #DeepdiveSessions #XavierAfrica

Want to stay ahead in market research? 🤯 Lenny Murphy sits down with Steven Millman, Global Head of Research and Data Science at Dynata, to unveil the secrets behind top-tier data quality and AI innovation! Discover how Dynata is revolutionizing data integrity, fighting survey fraud, and utilizing AI to push the boundaries of market research! Tune in for a deep dive into how AI is not just a trend but a game-changer for the future of research! → https://hubs.ly/Q02JY7b00

🚀 Excited to share my submission video for Project 3: "Credit Card Fraud Detection", with Encryptix! 🛡️🔍 In this project, I have come to know about various kinds of resampling techniques and mainly Support Vector Machines and the math behind it. Like how the dataset gets divided wrt an hyperplane and bounds around it get formed wrt hyperparameters- among which the Kernel trick is the most intriguing. Watch the video to see the journey and outcomes of this exciting challenge. #Data_Science #Encryptix #Suppor_tVector_Machines #Machine_Learning #AI #Credit_Card_Fraud

take 2 mins and use this short imagined scenario produced by Tom Scott to give you pause for considering "what are you building with AI, and why.." ( thank me later ).. https://bit.ly/3P29ASk

take 2 mins and use this short imagined scenario produced by Tom Scott to give you pause for considering "what are you building with AI, and why.." ( thank me later ).. https://bit.ly/3P29ASk

Lex Fridman’s conversation with Lee Cronin on Assembly theory and causal cones has me re-framing in my head AI/ML’s power as coming from the ability to constrain our focus and decisioning on the intersections of our collective causal cones rather than waste time outside that domain. This is huge in that it helps us come at problem solving endeavors with a foundational causal understanding of the world. Some frame this as a “compression of time”: what used to take years now takes weeks. I agree on the outcome, but perhaps the time compression is the learned causal chain embedded in the ML model, which equates to a more efficient inference engine (less time required to compute over a smaller space).

🔍 "Facts, Hercule, facts! Nothing matters without the facts. Without them, criminal science is just a guessing game." 🕵️♂️ This famous line from Blake Edward's 1964 film 'A Shot in the Dark' is still relevant today. 🔎 Investigators once painstakingly gathered facts to build cases. Now, every click, transaction, or message could be a vital clue in our data-rich world. 📱 Despite this, the guessing game persists. 🤯 Traditional methods can't keep up with the sheer volume of information. AI systems can help by: - Sifting through mountains of data ⛰️ - Turning scattered facts into coherent narratives 📖 - Quickly uncovering crucial leads ⚡ Get to the truth with facts that make sense and don't keep you guessing!

I am pleased to share a new blog post exploring the impact of human versus synthetic test data and round-tripping on the assessment of Sentiment Analysis Systems (SASs) for bias. The post delves into the challenges posed by the unstable behavior of AI systems like SASs when dealing with changes in data, particularly in relation to bias concerns. The study also introduces an innovative approach to assess SASs in a blackbox setting, shedding light on the performance of these systems in more realistic scenarios. To read the full findings and implications for SAS testing strategies, click here: https://bit.ly/3UdKWl7 [cs.CL]