Decentralizing AI with Jesse & Dustin

0:00 - The shift to personalized models 7:44 - Omarchy and AI-native operating systems 10:00 - AI-powered game decompilation 12:04 - Nostr's vision for a decentralized internet 24:04 - Adversarial algorithms and reclaiming control over interfaces 28:15 - Moving beyond LLMs with structured reasoning and world models 37:44 - Using world models to secure agents 41:26 - What the OpenAI math controversy shows about AI data privacy 48:33 - Trusted execution environments (TEEs) and private AI 1:04:37 - Can homomorphic encryption solve AI privacy? 1:09:07 - The Rosario Cipher and future of authentication 1:14:34 - AI doomerism and the politics of fear 1:26:34 - Balancing freedom and control Links: Omarchy: https://omarchy.us/ Nostr: https://github.com/nostr-protocol/nostr FIPS (Free Internetworking Peering System): https://github.com/jmcorgan/fips Marmot Protocol: https://github.com/marmot-protocol/marmot OpenAI's math controversy: https://www.wired.com/story/openai-is-pissing-off-a-bunch-of-mathematicians-again/ TrustedRouter: https://trustedrouter.com/ Tinfoil: https://tinfoil.sh/ TEE.fail: https://tee.fail/ Rosario Cipher: https://www.eni6ma.io/technology/rosario 🌐Learn more about Delegance at delegance.ai 🌐Learn more about Vora at vora.io

0:00 - Back from hiatus and the Open Source AI Summit 3:15 - AI fear, regulation, and control at the Pro-Human Assembly 5:46 - Is existential AI risk actually possible? 8:46 - The main problem with AI alignment 10:30 - OpenAI’s agent swarm and the reward hacking problem 13:02 - Alignment is ultimately a political question 19:42 - Loop and graph engineering for reliable AI 27:00 - Why “global alignment” is probably impossible 35:13 - Obama’s critique of LLMs and agentic AI 36:58 - Claude Code undermines Obama's case against agentic AI 40:14 - The deeper political questions behind AI safety Vora reveal: https://www.youtube.com/watch?v=3t5N8C1pU0c 🌐Learn more about Delegance at delegance.ai 🌐Learn more about Vora at vora.io

0:00 - Opening and what’s new 1:05 - Escaping the algo 5:22 - The dead internet and cost of AI slop 11:08 - Private communities as the future of the internet 19:16 - Nostr, Buzz, and the agentic internet 29:48 - What an AI-native company needs to own 39:33 - Secret messages in AI output 48:06 - The incentives behind the OpenAI-Hugging Face story 56:26 - Hidden risks inside neural networks 1:06:54 - The problem with Anthropic and OpenAI’s “dangerous AI” marketing 1:12:54 - Bioweapons as the strongest case for slowing AI 1:21:38 - The worldview behind AI fear and the case for freedom 🌐Learn more about Delegance at delegance.ai 🌐Learn more about Vora at vora.io

0:00 - Opening 0:50 - HSMs for agentic payments & the security tradeoff in personal AI 8:36 - The fragility of cloud security 11:43 - Vora's compartmentalized architecture for personal AI 14:50 - Why Fable 5 feels nerfed 20:13 - Fable 5 isn't a big deal because intelligence is no longer the bottleneck (data & harnesses are) 23:38 - Local models are getting useful: Hermes Agent, Qwen 3.6, and LLM wikis 28:18 - Why fast local inference with DeepSeek v4 Flash changes things 35:00 - Better workflows (i.e., HumanLayer AI IDE) beats more agents 40:40 - Planning systems for AI coding 52:00 - Fable 5 and model ceilings/diminishing returns 55:00 - Personal data as an AI moat 1:00:43 - Intelligence = search efficiency 1:08:30 - Why harnesses matter more than models 1:13:30 - Reasoning versus stored knowledge 1:21:10 - Why AI is so good at coding 1:28:13 - Export controls and AI gatekeeping 1:41:55 - Decentralization as civilizational infrastructure 1:56:30 - Practical steps toward AI sovereignty 2:04:20 - Using AI to dumb vs strengthen yourself 2:07:40 - Human verification in the age of cheap answers Everyone is focused on Fable 5, but what if the model wars are becoming a distraction? Jesse and Dustin discuss why the next phase of AI is probably less about raw model intelligence and more by data, harnesses, local hardware, memory, workflows, and control. The future may belong not just to the most powerful model, but the systems and stack around it. 🌐Learn more about Delegance at delegance.ai 🌐Learn more about Vora at vora.io

0:00 0xSero intro 3:00 Finding the best models for your machine (local.ai) 7:29 Why choosing local models is so hard 12:15 Specs that mislead users 14:15 Memory bandwidth is the real bottleneck 16:26 Qwen on cheap consumer hardware 18:13 Gemma, Strix Halo, and faster local models 20:00 How diffusion models could change speed 21:55 Fable 5 as a turning point & AI export controls 26:57 What powerful models actually threaten 33:57 Should powerful models stay open? 39:20 When AI helps attackers and defenders 46:32 'AI safety' in the social media age 53:50 Why local AI becomes personal survival gear 1:03:30 The economic case for local AI 0xSero joins Jesse and Dustin to discuss the local AI hardware boom, why running models at home is still so confusing, and how open-weight AI, personal agents, and sovereign intelligence could shape the future. 🌐Learn more about Delegance at delegance.ai 🌐Learn more about Vora at vora.io