Intelligent Machines (Audio)

Episode IM 850: Bagel Rats - Open Source AI Rising

Date: December 18, 2025
Host: Leo Laporte
Co-hosts: Paris Martineau, Jeff Jarvis
Guest: CJ Trowbridge (AI ethics expert, sustainability advocate, open-source AI developer)

Overview

This episode dives deep into the current state and future of artificial intelligence, focusing on sustainability, the open-source movement, and the ethical landscape of AI development and deployment. Guest CJ Trowbridge brings a critical, yet optimistic, perspective—emphasizing that AI's most transformative impacts may come from local, small-scale, and open models rather than from the "big tech" industry giants. The conversation is rich with discourse on energy use, economic realities, technical trajectories, and innovative grassroots AI projects.

Key Discussion Points & Insights

1. CJ Trowbridge's Early Access and OpenAI Backstory

• CJ's credentials: Early OpenAI tester, urban planning & sustainability background, ethical AI program.
• On early ChatGPT:
  • “These models are trained on documents... tricking that into chat is very complicated, led to a bunch of interesting problems.”
    — CJ Trowbridge [04:30]
  • Early product shortcomings: document completion vs. chat, context window limits, and awkward inclusivity “patches” in image generation (e.g., just injecting ‘diverse’ into prompts).
  • “We were talking about a lot of these problems...and they were like, 'We don’t care, it's too expensive.'”
    — CJ Trowbridge [07:02]

2. Resource Consumption Myths and Reality

• AI’s real energy use:
  • AI = less than 0.5% US energy use; all data centers <4%.
  • Studied projections of exponential energy growth were severely reduced ("cut by 75% or more").
  • Real driver of US energy price increases: political interference in renewables and grid funding, not AI demand.
  • “Those are enormous impacts. And the reality of how much energy AI is consuming is it's less than a percent. So it's just not true.”
    — CJ Trowbridge [11:00]
• Why the narrative persists:
  • 'AI stealing water and power' stories are often misinformed; e.g., Memphis XAI data center using treated sewage not drinking water [27:09].
  • “It's frustrating to see people saying like, 'AI is bad.' Elon is bad. Elon is killing people. AI will run on your phone all day and consume no energy or water.”
    — CJ Trowbridge [29:56]

3. Sustainability Via Small, Local, Open-Source Models

• Edge computing & open-source:

  • References to “Wisp Play” and other low-power, Raspberry Pi-based LLM devices.
  • Smaller, efficient models outpacing large language models (LLMs) for many tasks.
  • “Smaller models are better. We should have models that we can stick on a Jetson and run at home and not need to pay anybody rent for that.”
    — CJ Trowbridge [20:47]

• Empowerment:

  • “The sustainability picture for me looks like smaller devices that run at the edge, on solar power. … You can get better results that way than from the large labs.”
    — CJ Trowbridge [12:21]

4. Questioning Big Tech and Financial Bubbles

• Industry bubble speculation:

  • “Yes, 100% obviously (there’s a bubble). … None of that has any ROI and it's not clear that any of them even have a plan to get to ROI.”
    — CJ Trowbridge [15:03]
  • “You could see a lost decade or more where we have long protracted periods of low or negative growth.”
    — CJ Trowbridge [18:15]

• Nvidia’s market dominance:

  • On Nvidia: “My argument was Nvidia controls 98% of the market for inference hardware and they charge an arbitrary thousand percent markup… They should be charging more. And they should buy a platform like Hugging Face…”
    — CJ Trowbridge [22:33]

5. Ethical AI: Beyond Resource Worries

• Epistemic power & truth construction:

  • Concern over “who has epistemic power, the power to tell you what is true.”
    — CJ Trowbridge [25:07]
  • “Are these the people that you want to be in charge of telling the world what is true and false?”
    — CJ Trowbridge [25:35]

• Commoditization and anthropomorphism:

  • AI models are converging; little real difference as they train on the same data.
  • “There is this platonic ideal shape… There is some ideal shape they're all trying to converge on.”
    — CJ Trowbridge [35:09]
  • Warning against using loaded metaphors (reasoning, attention): “What we call attention is not at all related to what attention is.”
    — CJ Trowbridge [33:13]

6. Grassroots AI and Community Mesh Networks

• Off-grid mesh projects (“Cyber Pony Express”):

  • Solar-powered, peer-to-peer mesh communication with integrated AI.
  • Disaster response and community resilience (e.g., mesh chat at Burning Man for 10,000 people).
  • “You can text on this (off-grid) network... There’s no internet, no cell phone towers, and it’s all end to end, encrypted… it’s a totally anarchist peer-to-peer system.”
    — CJ Trowbridge [43:06]

• High Desert Institute & Nonprofit Land Projects:

  • Supporting free off-grid communities, mesh networks, educational robotics, and resilient infrastructure.

7. Practical Tips for Small-Scale AI Enthusiasts

• How to get started:
  • “LM Studio is great. It's a free open source project… Just very simple library that can be included in apps.”
    — CJ Trowbridge [39:36]
  • For mobile: “PocketPal is an open source app that lets you run any of these models on your phone, all day, without needing to plug in.”
    — CJ Trowbridge [40:34]
  • Using Docker to run local LLMs easily.

8. Open-Source Frontier & China’s AI Strategy

• Chinese open models outperforming US cloud AIs:
  • “Everything China is doing has been amazing in AI. … All these free models that are more performant than the best stuff you can rent in the cloud here.”
    — CJ Trowbridge [30:26]
  • Critique of US companies' data practices and the potential of public domain and open-source datasets (e.g., FineWeb).

Notable Quotes & Memorable Moments

• “It’s not the issue. The issue is the people who are bringing us AI. … Let’s do it better on our own. Let’s do grassroots stuff.”
  — Leo Laporte and CJ Trowbridge [37:15-37:25]

• “If the bubble pops, if all these companies go away, almost all the value of what's been created we will still have in these open source models…”
  — CJ Trowbridge [49:29]

• “Elon is bad. Elon is killing people. AI will run on your phone all day and consume no energy or water.”
  — CJ Trowbridge [29:56]

• “There are bad ways to do AI, absolutely.”
  — CJ Trowbridge [30:17]

• “Grassroots, local, open-source AI is the future—created by the people, for the people.”
  — Paraphrased summary [42:32-42:54]

Important Segments & Timestamps

• CJ’s Backstory & Early AI Critique
  [03:09]–[07:19]
• Resource Myths Debunked
  [08:54]–[12:12]
• Edge/Open-source AI & Wisp Play
  [12:13]–[14:31]
• Big Tech Bubble & Nvidia Take
  [14:49]–[24:21]
• Ethical AI & Epistemic Power
  [24:58]–[26:46]
• Memphis Data Center Water Myth
  [27:09]–[29:56]
• Off-grid Mesh AI Networks
  [42:20]–[43:56]
• Getting Started with Local Models
  [39:36]–[41:19]
• The Case for Open-Source & China’s Models
  [30:26]–[32:37]
• Commoditization & Anthropomorphic Language
  [33:09]–[35:09]
• Summary Vision: Open-Source Endgame
  [49:29]–[50:07]

Conclusion & Takeaways

• The future of AI doesn’t have to be dominated by giant "mag 7" tech companies. Local, open-source tools are now strong enough to challenge big data center models—offering real sustainability, privacy, and community empowerment.
• Most of the alarmist narratives about AI’s energy and environmental impact are based on exaggerated or misinformed claims; the true sustainability challenges are political and infrastructural.
• Ethical risks of AI have less to do with resource use and more to do with who controls truth, whose voices are embedded in the models, and how we choose to design and deploy these technologies.
• Grassroots mesh networks and open-source models are showing a different way forward—one where AI benefits can be decentralized, democratized, and less exploitative.
• “If the bubble pops... we’ll still have the tools—we just need to learn to use them ourselves.”

For Listeners

Whether you're new to AI or seasoned, this episode offers a fresh, practical, and hopeful look at how anyone can contribute to— and benefit from— the next AI revolution, outside big tech.

Full episode and show notes available at twit.tv/IM850
For in-depth project info and open-source tools: cjtrowbridge.com