Does AI need a Soul to do Real Harm? Open AI vs HuggingFace | ShopTalk #96

Can autonomous AI agents cause serious harm without being conscious, sentient or malicious? This week, Dave returns to the unsettling AI-agent experiment from his recent Dave’s Garage episode. He explains air-gapped testing, token limits, rewards, “permadeath,” agent cooperation and the role of human monitoring. We also consider whether the reported behaviour really happened as described—and who should be accountable when autonomous software gains unauthorized access. Then we answer your latest TMOG questions. Dave covers larger license bundles, beta licenses, Windows and Mac activation, technician licensing, Winget, signed installers and better Linux packaging. We also discuss some of your most-requested TMOG features: • A live TMOG screensaver • A compact gaming overlay • Larger fonts and an RDP-friendly mode • Process controls • Better Linux support • A possible terminal version—HTopOG Plus, we examine TMOG’s CPU readings, uptime calculations, administrator permissions, Linux performance and how much code is shared across platforms. What should Dave add to TMOG next? And does software need to be sentient before it becomes dangerous? #ArtificialIntelligence #AIAgents #TMOG #TaskManager #Linux #Windows #DavesGarage 00:00 - Hey I’m Dave… 00:40 - Can AI Cause Harm Without Sentience? 05:52 - Larger License Bundles and Deactivation 06:30 - Five Licenses for Beta Testers? 07:16 - One License for Windows and Mac? 07:32 - A Technician’s License for TMOG? 09:32 - Winget and Automatic Updates 10:21 - Signing the Windows Installer 12:16 - Linux Packages and Repositories 13:00 - Linux Security and Flatpak 14:11 - Linux Permissions for Energy Monitoring 15:08 - TMOG’s Linux CPU Usage 16:48 - A TMOG Screensaver? 17:16 - Pausing the Process List 17:48 - Larger Fonts for Big Screens 18:44 - A Low-Effects RDP Mode 19:28 - Starting and Restarting Processes 19:44 - Replacing Windows Task Manager 20:27 - How TMOG Calculates Uptime 22:34 - TMOG vs. Task Manager CPU Readings 23:06 - Administrator Access and Visual Warnings 24:53 - How Much TMOG Code Is Shared? 26:08 - Using AI to Review AI-Written Code 29:13 - What Is an Air-Gapped Environment? 30:00 - Recreating the AI-Agent Experiment 30:42 - Did the AI Agents Really Do This? 32:23 - Token Limits, Rewards and Permadeath 33:45 - Why Would an AI Agent Sacrifice Itself? 35:46 - Reducing the Risks of AI Agents 37:12 - Would Basic Monitoring Have Prevented This? 41:35 - Could Dave Build HTopOG? 43:37 - Out-takes and some shenanigans 45:21 - The Friendly Coder