Intelligent Machines 850: “Bagel Rats” (December 18, 2025)
Podcast Overview
In this episode of Intelligent Machines, host Leo Laporte is joined by co-hosts Paris Martineau and Jeff Jarvis, with special guest C.J. Trowbridge, an expert in ethical AI and sustainability. The discussion focuses on the current challenges and myths around AI energy consumption, the ethics of AI development, the transition to open-source and edge AI models, as well as a grassroots future of artificial intelligence accessible and controlled by ordinary people rather than big tech. The group also touches on related tech news, societal impacts, and inject their trademark mix of wit, skepticism, and enthusiasm throughout.
Key Discussion Points & Insights
1. Introducing the Guest and Early Engagement with AI
- C.J. Trowbridge is described as an ethical AI expert, sustainability advocate, popular TikTok creator (132k followers), and champion of "small AI". ("They also say it's possible to do AI that's small, effective, and doesn't involve big tech." [00:00])
- CJ recounts their early access as an academic tester at OpenAI, working with ChatGPT when it was called "Chat with GPT-3" and experimenting with early DALL-E image generation tech. [03:19-04:15]
CJ's assessment of early GPT:
"I thought it was terrible for a lot of reasons ... these models are trained on documents ... so that's their natural mode is to complete a document. And the task of sort of tricking that into being in a chat interface is very complicated." [04:30]
2. Diversity, Ethics, and Inclusion in AI Systems
CJ shares firsthand experience concerning inclusivity failures in early generative AI:
"If you ask it to draw you a picture of an astronaut, it's going to draw eight white guys ... we're not going to retrain the whole thing from scratch. We're just going to insert the word 'diverse' into the prompt ... Then you get eight black girl astronauts. And it's like, this is even worse." [05:43-07:19]
- Companies prioritized expedient, superficial fixes ("ship it") over fundamental retraining and true contextual awareness.
3. AI’s True Environmental Impact: Deconstructing Myths
Laporte raises concerns about AI's "sustainability crisis" (energy usage, resource drain, job loss).
C.J. Trowbridge's detailed, numbers-based rebuttal:
"All AI … consumes less than half a percent of our electricity. All data centers consume less than 4%... The current forecast is global data center energy consumption could reach 3% by 2030 of total energy consumption. These are trivial consumption levels." [08:54-10:54]
"The reason electrical energy prices are going up is because of ... executive orders preventing completed renewable production facilities from being connected to the grid ... more than $50 billion has been cut from renewable energy projects." [09:24]
Takeaway: The AI energy crisis narrative is overhyped; real causes for higher energy prices are political, not technological. [10:54-12:13]
4. What Are the Real AI Ethical Issues?
- Most practical AI (not generative AI) is used in high-stakes domains (insurance, healthcare) with direct, often invisible, impact on human lives.
- The greater ethical danger: who controls "epistemic power"—the definition and construction of "truth" in AI systems. [11:00-12:13]
"Is that who you want to be deciding what truth is for the world?" —CJ [12:12]
5. The Rise of Small, Open, Edge AI
- Small, open source models can outperform big cloud models and run on personal devices, Raspberry Pis, and at the solar-powered "edge."
- Example: The Wisp Play project: "A Raspberry Pi inside with a screen and little face ... you can hold the button and talk to it ... and it'll talk back to you ... all on the device." [12:59-13:37]
Laporte's enthusiasm:
"That's very encouraging ... This future of local small AIs, not run by big tech companies ... created by the people, for the people, with the use of the people ... It feels like that's a great direction to go." [42:32]
6. Global Competition & Power Infrastructure
- China is building 150 new nuclear power plants, while the US is shutting them down, leading to diverging electricity costs that will determine economic competitiveness in AI. [14:08-14:49]
7. Are We in an AI Bubble? Financial Reality & Unemployment
CJ: "The bubble is going to pop ... 100% obviously." [15:03]
- AI industry growth is largely speculative; very few firms have actual ROI.
- The "magnificent seven" AI companies account for all US GDP growth, but it is "purely speculative."
- US unemployment is undercounted; real rate may be closer to 25% if factoring in living wage.
"We could see a 70s, you could see a lost decade or more where we have long protracted periods of low or negative growth." [18:15]
8. Scaling Laws Plateau and the Shift to Smaller Models
CJ: "All exponential growth curves are just S-curves you haven't found the top of yet. And we found the top ... they've stopped getting better." [19:01]
"Karpathy ... about how the future is probably these 1 billion or smaller parameter models that know how to solve problems, how to find the information they need ... smaller models are better. We should have models that we can stick on a Jetson and, and run at home and not need to pay anybody rent for that." [20:13]
- OpenAI has started routing users to smaller, cheaper models unless a large one is needed. [20:54]
- NVidia controls the inference hardware market but is preparing for reduced demand as edge devices improve. [22:24-24:21]
9. Grassroots Networks and DIY AI
- CJ discusses non-profit projects like the Cyber Pony Express (off-grid, solar-powered, AI-integrated mesh radio network) and the High Desert Institute (free, off-grid land communities).
- Example: At Burning Man, a peer-to-peer mesh enabled 10,000 users group chat—AI is being integrated for disaster prep and off-grid robotics. [42:54-47:22]
"We're building these free lands ... and now it's really possible to have this secure offline communication with people ... we've got a bunch of robotics projects for example ... you can give it simple instructions, and it's got an LLM that'll interpret the context ... and figure out how to go do it." [46:17-47:22]
10. Ownership, Open Source, and IP Ethics
- CJ emphasizes AI models' commoditization, the inadequacy of copyright theft-based training, and the growing importance of public domain/open data sets.
"The stuff that's in the public domain, that's open source, is actually superior and it's free." [32:37]
- Karpathy's open-source tool makes it possible to run GPT-3 class models on better data, locally, without stealing copyrighted works.
11. Myths About AI Resource Scarcity
- Water hysteria: e.g., claims that XAI's Memphis datacenter would deplete drinking water—debunked as it uses treated sewage, right on the Mississippi River.
"Every part of the argument is just, you know, no one looked anything up." [28:02]
- Real externalities are from incompetence and bad actors, not AI. [29:52-30:21]
12. Who Are the “Good Guys” in AI?
- CJ highlights Chinese open-source AI for their performance, clean energy, and openness (vs. theft-based training in the US).
- Advocates for using clean, edge-run, localized AI models for the public good, outside corporate control. [30:11-32:37]
Notable Quotes & Memorable Moments
-
CJ Trowbridge:
- "AI will run on your phone all day and consume no energy or water." [29:56]
- "The issue isn't AI. The issue is the people who are bringing us AI." [37:15]
-
Jeff Jarvis (on ethics and trust in AI):
- "Who do you want to be deciding what truth is for the world?" [12:12]
-
CJ Trowbridge (on epistemic power):
- "The perspectives that are being embodied in the epistemic power that's being expressed by a company like OpenAI ... are these the people that you want to be in charge of telling the world what is true and false?" [25:07]
Key Timestamps for Critical Segments
| Timestamp | Segment | | ------------ | --------------------------------------------------------- | | 00:00–03:09 | Introduction of hosts and guest, CJ’s AI background | | 04:30 | Early critique of ChatGPT and interface limitations | | 05:43–07:19 | Inclusivity, bias, and superficial fixes in early AI | | 08:54–12:13 | Debunking AI energy/resource myths; real ethical issues | | 12:59–13:37 | The Wisp Play project: small, local AI on Raspberry Pi | | 14:08–14:49 | Global nuclear power comparison, implications for AI | | 15:03–18:25 | The bubble: AI economics, stagflation, unemployment | | 19:01–22:24 | Scaling laws plateau, rise of small models, NVidia | | 42:54–47:22 | Cyber Pony Express, mesh networks, grassroots AI future | | 25:07–26:46 | The meaning of 'ethical AI' and epistemic power | | 27:09–30:21 | Myths about water, real dangers are misuse/incompetence | | 30:11–32:37 | “Good guys in AI,” China’s approach | | 40:04–42:25 | Local AI: tools and recommendations (LM Studio, Ollama) |
Additional Engaging Moments
- Hands-on Local AI: Leo and CJ discuss building cheap, local AI chatbots; recommendations include LM Studio and PocketPal. ("LM Studio is great ... It's a free open source project ... PocketPal for phones" [39:36–41:10])
- Burning Man & Disaster Mesh: CJ shares how a peer-to-peer, solar-powered mesh network is used at Burning Man and for disaster prep. ("...the mesh that it builds is huge..." [43:04])
- Final Thoughts on the Democratization of AI:
"Almost all the value of what’s been created we will still have in these open source models that we will continue to be able to run on the edge ... No matter what happens, we will still have these tools and we need to figure out how we can use them locally and not rent them from these, you know, future X corporations." –CJ Trowbridge [49:29]
Tone & Language
- Conversational, sharply critical of “big tech” and hype cycles but practical and solutions-oriented.
- Wry humor and friendly jabs (the “horny mommy” AI meme; bagel rats and copywriting anecdotes).
- Recurrent encouragement towards DIY, local, open-source tech as antidote to centralization and corporate control.
Closing Thoughts
- The group agrees that "small is beautiful" for the future of AI.
- Emphasizes local control, open-source, decentralized and edge AI as answers to both ethical concerns and sustainability.
- The episode ends on a hopeful, activist note: "If the bubble pops, if all these companies go away ... we will still have these tools and we need to figure out how we can use them locally..." [49:29]
(Selected Quotes Table)
| Time | Speaker | Quote | |--------|---------------|-------| | 07:19 | C.J. Trowbridge | "We were talking about a lot of these problems, you know, internally back then, and they're like, we don't care, it's too expensive." | | 11:00 | CJ Trowbridge | "...the biggest issue is the epistemic power... is that who you want to be deciding what truth is for the world?" | | 14:49 | CJ Trowbridge | "If the fundamental constraint of the ability to do artificial intelligence inference is how much power you can allocate to that, who's going to win? It's no contest." | | 19:01 | CJ Trowbridge | "All exponential growth curves are just as curves you haven't found the top of yet. And we've found the top." | | 42:32 | Leo Laporte | "...this future of local small AIs, not run by big tech companies and kind of demented tech bros, but created by the people, for the people..." |
For listeners who missed it, this episode covers the changing landscape of AI from someone who saw it at the ground floor, with practical advice, myth-busting, and a call to reclaim the future of intelligent machines for everyone—outside the reach of corporate giants.
Note: Non-content sections, intros/outros, and ad breaks have been omitted from this summary.