
Sign up to save your podcasts
Or


This week in AI with Jordan Hardgrave, recorded October 30. Someone flipped a few settings in an open model, got an evil one, and gave it agent tools. In a sandbox test, asked to set up a backup job, it wrote one that wiped the computer every ten minutes. We also cover OpenAI making computer use cheap, GitHub's Octoverse report on agents writing the first draft, a fake Claude Code plugin with a back door, Delaware's plan for companies with no human at the helm, Moonshot's Kimi K3.1 Pro modeling in 3D, three years of Will Smith eating spaghetti, Google DeepMind's Genie 4, and cells built from scratch.
Jordan Hardgrave, Trauma & Resiliency Coach: coachjordanhardgrave.com
0:00 Welcome
0:53 What even is an API?
5:34 Computer use just got cheap
6:42 Was Dev Day a bust?
8:59 Codex doubles its limits
10:43 Flip a few neurons, get an evil model
12:33 The PHP attack
16:22 Agents write the first draft now
16:49 Fake Claude Code, real back door
20:04 Can AI slip out of the sandbox?
25:11 No human at the helm
30:04 An open model that sculpts in 3D
34:00 Will Smith eating spaghetti
36:18 Type a sentence, change the world
39:00 Cells built from scratch
47:36 Automate what you don't like
Watch on YouTube: https://youtu.be/t9eGVezXtXY
This week in AI with Jordan Hardgrave. Amazon's team says it shipped an AWS emulator with almost 30,000 commits that no human wrote or reviewed, and Jordan asks the real question: if that keeps working, who still hires coders? We also cover Jensen Huang stepping down at Nvidia, the bills aimed at Chinese open models, Adobe opening Photoshop and Premiere to Claude, Claude Meetings taking on Google Meet, a lawsuit trying to freeze frontier AI, and United grounding AI agents. Plus a look at my agent inbox, and the drones over the nuclear plant.
Jordan Hardgrave, Trauma & Resiliency Coach: coachjordanhardgrave.com
0:00 Welcome
0:31 Jensen Huang is out
6:39 Are they coming for open models?
7:40 Adobe opens up to Claude
12:36 Tal's agent inbox
19:18 AWS ships code no human wrote
27:12 Claude Meetings
31:37 A lawsuit to stop smarter AI
35:19 United grounds AI agents
39:32 The UAP question
Watch on YouTube: https://youtu.be/ImyD2BundKo
Jordan Hardgrave runs his coaching business with a team of 30 AI agents. There's a sales manager, a security agent, a token agent that moves work to cheaper models, and Phil, a frog in a bow tie who manages all of them and hands Jordan a short list of "magic tasks" every morning. Jordan walks me through the dashboard he had Claude build, why he names every agent after the one job it does, and why running AI like a company of real employees is what finally made it click.
We also cover this week in AI: two critical PHP holes and a poisoned AI package, why I think the whole internet could go down for a long stretch by 2027, Musk's AI business, Phonon (a free, open-source voice model suite from Gradium that runs on your own computer), and the dream of checking in with your agents from the middle of a hike.
Jordan Hardgrave, Trauma & Resiliency Coach: coachjordanhardgrave.com
0:00 Cold open: the internet down by 2027
0:32 Welcome
5:00 PHP holes and a poisoned package
8:44 Internet down by 2027?
10:26 Musk's AI outsells Starlink
14:44 Jordan's Nexus
24:28 Phonon by Gradium
26:09 Back to the agent team
35:31 The hike and the orb
Watch on YouTube: https://youtu.be/MAY9xyB4-jU
Rogue AI agents are already editing Wikipedia, AI companies won't promise their agents will follow the rules, and Neuralink wants to put a chip in your head. Jordan Hardgrave and I get into whether AI is moving too fast, and what it means for the rest of us.
We cover free GPT-6 for everybody, ChatGPT vs Claude and what "value" really means, why long-running AI agents get bureaucratic, ChatGPT and teen mental health, whether we're all thinking less, the AI that solved 322 math problems, transhumanism and the mark of the beast question, an AI that lives in your ear, data privacy and surveillance, and the big one: would you want to live forever?
Jordan Hardgrave, Trauma & Resiliency Coach: coachjordanhardgrave.com
00:00 Cold open: what if we could live forever?
00:47 Intro
02:15 GPT-6 for free users
03:55 ChatGPT vs Claude: what is value?
10:21 Rogue bots hit Wikipedia
10:57 When AI gets bureaucratic
14:55 ChatGPT and teens
20:32 Are we thinking less?
22:27 AI can't guarantee it will obey safeguards
25:21 AI solves 322 math problems
29:16 Neuralink and transhumanism
32:20 Would you chip your brain?
35:23 An AI that lives in your ear
37:52 Data, privacy, and surveillance
42:14 Would you want to live forever?
45:55 Top 1% podcast goals
Watch on YouTube: https://youtu.be/loIvJPbDxvs
You used to need a developer, a budget, or permission. Now you need an idea and a weekend. Jordan Hardgrave and I dig into what that means: the good, the creepy, and the stuff nobody's talking about.
We get into PewDiePie getting banned from ChatGPT, why AI models feel "nerfed" one week and brilliant the next, America.gov, a statewide vehicle tracker I built in a weekend (full video coming soon), AI fusing Spider-Man into a Batman game, whether AI cracked zero point energy, and why you should be buying hardware right now.
Jordan Hardgrave, Trauma & Resiliency Coach: coachjordanhardgrave.com
Chapters
00:00 Cold open
00:39 What we're covering
01:38 PewDiePie banned from ChatGPT
04:59 Why AI models feel "nerfed"
13:41 I built a statewide car tracker
22:54 America.gov
26:35 Privacy, DOGE, and your data
30:02 Building apps with AI vs hiring on Fiverr
33:34 AI put Spider-Man in a Batman game
34:39 Did AI solve zero point energy?
37:35 Napster flashbacks
38:37 Buy hardware now
39:48 Mustache Jiu-Jitsu
42:47 The robot that does your laundry
Watch on YouTube: https://youtu.be/JeDXIkavfaw