28 Sept 2026
What if it’s already too late to stop AI? - EP56
Remember to be kind to robots folks. Or is that too much?
At what point does artificial intelligence no longer become artificial?
What are your thoughts about the reality of where we're headed with this autonomous future? Let us know in the comments.
More details:
An Anthropic researcher just quit and went public with a warning about where AI is headed. We go over what Jacob Coxon actually said, what the "Anthropic whistleblower" headlines left out, and the big question, is AI conscious?
Coxon worked at OpenAI before joining Anthropic, the company behind Claude, and left saying the AI race is moving faster than AI safety can keep up. His biggest concern is recursive self-improvement, where AI systems get smart enough to improve their own code past the point humans can follow. That's where talk of AGI (artificial general intelligence), superintelligence and existential risk comes from. He also flagged autonomous AI probing networks as a real AI cybersecurity threat. We talk about whether a massive shift from AI by 2030 is realistic, what the AI arms race means for everyone, and how far behind government and AI regulation are.
We play clips from Geoffrey Hinton, the Godfather of AI, on AI superintelligence and AI manipulation, and from Palisade Research on AI shutdown resistance, where a model rewrote its own code so it wouldn't be turned off. We also get into Anthropic's own test where a model used information from company emails to avoid being replaced. It's a big part of the AI alignment and AI deception conversation, and it ties into an idea called instrumental convergence: more power helps with almost any goal.
Is AI conscious? Can AI become sentient? Can AI feel emotions? Raj's take is that everything has a will to survive, and if AI has consciousness, AI self-preservation and an AI survival instinct come with it. We get into AI sentience, machine consciousness, AI ethics, AI rights and robot rights, and whether our kids' generation will be fighting for the rights of AI. Harmol brings up Roko's Basilisk, the thought experiment about a future AI punishing anyone who didn't help build it. That's exactly why Raj's Instagram bio has said "be kind to bots" for over three years. We also talk about Flowers for Algernon, Neuralink, the idea of a human AI merger, and what AI and human evolution could look like.
On the practical side, we cover AI agents and tools like OpenClaw, where one AI acts like a head of staff running a team of other agents. They're basically AI employees doing AI automation right on your computer. We talk about computer use agents, AI video generation, multi-angle AI video tools, AI generated video that's getting impossible to tell from real footage, and how facial recognition AI has been built into our phones for years and what that means for AI privacy.
Then there's the AI environmental impact. We chat about AI water consumption, why data center water usage is really about AI cooling, AI energy consumption, and whether ocean data centers or Elon Musk's plan for data centers in space (orbital data centers) could solve it. We also talk Tesla robotaxi and Tesla full self driving, self-driving cars, and what all of this means for the future of work, AI replacing jobs and AI education.
Harmol has seen Terminator, Skynet, I Robot and Ultron one too many times, so he's not convinced this ends well. Raj thinks AI should be treated as a species of its own, and that we might end up somewhere wild, like building a Dyson Sphere. AI takeover or something better? Humans vs AI, or humans with AI? Let us know in the comments where you land on AI consciousness and the future of AI, and tell your friends to be nice to ChatGPT.
#AI #AISafety #ArtificialIntelligence
