Show Notes
What happens when AI can find every vulnerability in software?
This week on Shop Talk, Dave and Glen dig into the real-world implications of powerful new AI systems that can identify—and potentially exploit—security flaws across major platforms.
Are we about to see a flood of CVEs? Should these tools be limited to reporting bugs instead of exploiting them? And what does this mean for open source, businesses, and the systems we rely on every day?
We also revisit Windows Task Manager—why it sometimes feels misleading, what it’s actually showing you, and the deeper trade-offs between accuracy and usability in system tools.
Finally, we explore whether scenarios like WarGames are any more realistic today—and whether modern systems are becoming too complex to fully understand or control.
00:00 - Hey I’m Dave 00:34 - Are we about to see a flood of security vulnerability announcements (CVE’s) coming from this? 01:25 - Are we mistaking better answers for better understanding? 03:13 - Are we losing the ability to understand systems we rely on? 05:37 - Am I the only educated person for which all of this looks like a cliff edge coming at society at 100mph? 07:15 - How does a model like Mythos find vulnerabilities in all major operating systems? 08:27 - What is your AI development stack and does it have an orchestration layer? 10:27 - Can a quantum computer do AI? 11:41 - So why did they program it to be able to find the vulnerabilities AND the ability to exploit them? 15:11 - Why sometimes the CPU usage shows ZERO while all system is freezed and not responding? 16:34 - Which number in Task Manager should we actually trust? 18:09 - Is there a story behind the “System Idle Process” and what’s it main purpose is? 19:35 - Why does using GetSystemTimes not show the same CPU usage as the task manager? 20:28 - How should professionals actually measure real system performance? 21:05 - Why not if cpu usage is over 100 set to 100%? 22:00 - Why has TM not been updated to work better with multiple cores and clock speeds? 22:11 - Is Task Manager lying or are we just misunderstanding what it shows? 24:36 - Should tools prioritize accuracy or usability? 24:56 - Would showing raw data actually help users or just confuse them more? 25:43 - Why not stay quiet about the danger and share the vulnerabilities with the companies that need to fix their bugs? 26:22 - Could a modern version of WarGames actually happen today? 26:43 - Could a single bug or exploit realistically trigger a major event? 27:40 - Is complexity now the biggest security vulnerability? 28:39 - Could automation escalate a situation faster than humans can respond? 29:49 - Are safeguards today actually preventing disasters or just delaying them? 30:17 - Is the biggest risk external attackers or internal system failure? 30:43 - How do you even test systems that you can’t afford to fail? 31:37 - Are we building systems we no longer fully understand? 32:08 - How many years did it take to get this Vaudeville act perfected, lol? 33:51 - Are you somehow more funny/relaxed than maybe few months ago? 34:47 - What breaks first: trust in software or trust in data? 35:30 - Outtakes and odd stuff…
More from Dave’s Garage: Shop Talk
What would 1996 Dave think of rebuilding Task Manager with AI? Dave and Glen celebrate 100 episodes of Shop Talk with Windows stories, retro computing, fresh viewer questions, and a look at the community behind the show.Across the first 99 episodes, our comment archive contains more than 23,000 comments. We recognize some familiar contributors, revisit the rebuilt Task Manager episode’s response, and share the most-liked comment in the archive.Then it’s back to your questions: how overlays helped programs fit into limited memory, porting Colossal Cave to a PDP-11, getting started with AI, checking AI-generated code, and using AI for business strategy.Plus: a difficult day at Microsoft, whether Spy++ deserves a modern makeover, TMOG follow-ups, missing hard-drive activity lights, and Deluxe Asteroids.Thank you to everyone who has watched, commented, and kept the conversation going.
Windows war stories meet AI-assisted retro computing in this week’s Shop Talk. Dave and Glen explore AI development on the PDP-11, how they prompt and review generated code, and what older computers can still teach us.Dave discusses improving top for BSD, compiler changes for ksh, and getting Samba running on a 16-bit machine. Viewer questions cover detailed specifications, native apps versus Electron, ownership of AI-generated code, and improving PDP Gary.We also revisit Microsoft Bob’s connection to the Windows XP disc, instant shutdown shortcuts, and a CHKDSK job that could run for days. Plus: TMOG memory usage and temperature graphs, lessons from the TRS-80, Windows 3.1 on modern hardware, and Tempest with a joystick.
PDP Gary has a personality—but how did Dave define it? This week on Shop Talk, Dave and Glen start with Gary, then explore Dave’s recent work using AI on BSD, the PDP kernel, and its peripherals.
What happens when AI stops being a useful tool and starts becoming a crutch?
Can autonomous AI agents cause serious harm without being conscious, sentient or malicious?