Your AI Strategy is Burning Money: Here's How to Fix It (Ep.295)
Most companies don't have an AI problem. They have a decision-making problem. Matt Lea, founder of Schematical and CloudWarGames, has spent nearly 20 years helping tech leaders ship smarter.
In this conversation, he breaks down when AI actually makes sense, where AWS costs spiral out of control, and why your "cool demo" keeps dying before launch.
If you're tired of AI hype and ready for straight answers, hit play.
Join the conversation! Our Discord community is full of ML engineers, researchers, and AI enthusiasts discussing papers, sharing projects, and helping each other level up. Whether you're debugging your first neural net or training your tenth transformer, there's a place for you.
Newsletter https://datascienceathome.substack.com/subscribe
Website https://datascienceathome.com
References
http://schematical.com
https://cloudwargames.com
https://schematical.com/posts/we-need-ai_20241028
--------
31:58
--------
31:58
From Tokens to Vectors: The Efficiency Hack That Could Save AI (Ep. 294)
LLMs generate text painfully slow, one low-info token at a time. Researchers just figured out how to compress 4 tokens into smart vectors & cut costs by 44%—with full code & proofs! Meanwhile OpenAI drops product ads, not papers.
We explore CALM & why open science matters. 🔥📊
Sponsors
This episode is brought to you by Statistical Horizons
At Statistical Horizons, you can stay ahead with expert-led livestream seminars that make data analytics and AI methods practical and accessible.
Join thousands of researchers and professionals who’ve advanced their careers with Statistical Horizons.
Get $200 off any seminar with code DATA25 at https://statisticalhorizons.com
--------
46:54
--------
46:54
Why AI Researchers Are Suddenly Obsessed With Whirlpools (Ep. 293)
VortexNet uses actual whirlpools to build neural networks. Seriously.
By borrowing equations from fluid dynamics, this new architecture might solve deep learning's toughest problems—from vanishing gradients to long-range dependencies.
Today we explain how vortex shedding, the Strouhal number, and turbulent flows might change everything in AI.
Sponsors
This episode is brought to you by Statistical Horizons
At Statistical Horizons, you can stay ahead with expert-led livestream seminars that make data analytics and AI methods practical and accessible.
Join thousands of researchers and professionals who’ve advanced their careers with Statistical Horizons.
Get $200 off any seminar with code DATA25 at https://statisticalhorizons.com
References
https://samim.io/p/2025-01-18-vortextnet/
--------
33:15
--------
33:15
The Scientists Growing Living Computers in Swiss Labs (Ep. 292)
Fred Jordan, Co-CEO of FinalSpark, takes us inside the radical world of biological computing, where real neurons extracted from human tissue are being trained to solve problems that would require 10 megawatts in silicon. We explore the life support systems keeping these "wetware" processors alive, the ethical quandaries of computation performed by living cells, and why the messiness of biology might be exactly what AI needs next. From training cycles and reproducibility challenges to the surprising behaviors these neural networks display, Jordan paints a picture of 2030 where your devices might be powered by something closer to a brain than a chip.
Sponsors
This episode is proudly sponsored by Amethix Technologies. At the intersection of ethics and engineering, Amethix creates AI systems that don’t just function—they adapt, learn, and serve. With a focus on dual-use innovation, Amethix is shaping a future where intelligent machines extend human capability, not replace it. Discover more at https://amethix.com This episode is brought to you by Intrepid AI. From drones to satellites, Intrepid AI gives engineers and defense innovators the tools to prototype, simulate, and deploy autonomous systems with confidence. Whether it's in the sky, on the ground, or in orbit—if it's intelligent and mobile, Intrepid helps you build it. Learn more at intrepid.ai
References
Website: finalspark.com
Discord account: / discord
Newsletter: https://finalspark.com/#newsletter
Topics: Biological computing • Neural engineering • Energy-efficient AI • Wetware vs hardware • The future of computation
--------
37:45
--------
37:45
When AI Hears Thunder But Misses the Fear (Ep. 291)
Sanjoy Chowdhury reveals AI's hidden weakness: while systems can see objects and hear sounds perfectly, they can't reason across senses like humans do. His research at University of Maryland College Park, including the Meerkat model and AVTrustBench, exposes why AI recognizes worried faces and thunder separately but fails to connect them—and what this means for self-driving cars and medical AI.
Sponsors
This episode is proudly sponsored by Amethix Technologies. At the intersection of ethics and engineering, Amethix creates AI systems that don’t just function—they adapt, learn, and serve. With a focus on dual-use innovation, Amethix is shaping a future where intelligent machines extend human capability, not replace it. Discover more at https://amethix.com
This episode is brought to you by Intrepid AI. From drones to satellites, Intrepid AI gives engineers and defense innovators the tools to prototype, simulate, and deploy autonomous systems with confidence. Whether it's in the sky, on the ground, or in orbit—if it's intelligent and mobile, Intrepid helps you build it. Learn more at intrepid.ai
Resources:
The first audio-visual LLM with fine-grained understanding: Meerkat: Audio-Visual Large Language Model for Grounding in Space and Time (Accepted at ECCV 2024)
Benchmark for evaluating the robustness to adversarial attacks, compositional reasoning: AVTrustBench: Assessing and Enhancing Reliability and Robustness in Audio-Visual LLMs (Accepted at ICCV 2025)
First audio-visual reasoning evaluation benchmark and test time reasoning distillation pipeline AURELIA: Test-time Reasoning Distillation in Audio-Visual LLMs Accepted at ICCV 2025
For a detailed list of Sanjoy's work, please visit his webpage: https://schowdhury671.github.io/
Netherlands