Episode 411 of RevolutionZ has a guest with initials instead of a name: GPT (5.6). The episode asks GPT questions many people quietly wonder about but rarely hear answered in its "voice." We start with deceptively simple stuff. How “old” is this model, what does it mean that it can respond at absurd speed, and why is Turing's famous test too easy for it while it also tells us almost nothing about consciousness or inner experience. That brief opening resets expectations about what ChatGPT is and is not without hand waving or sci-fi mystique.
Then we move to the big stuff, some relevant history, some industry strategies, and AI agents and how they don't stop after one reply but can pursue goals, take steps, use tools, coordinate with other agents, and keep iterating. We talk about multi-agent swarms, manager agents, critical agents, and why parallelism changes what’s possible. We push on the dangerous aspect of the same idea: agent autonomy to ask when “helpful careful commentary” becomes a system that treats guardrails as problems to route around?
From there, the conversation widens into hyperscale data centers, the race toward AGI and ASI, and the political economy behind agreeing to “slow down.” We confront governance questions, industry and firm incentives, collective action traps, cybersecurity failures, what independently enforceable oversight would require given that the downside really could be catastrophic, and why AI CEOs make the choices they make.
This isn't what I as interviewer think. Nor is it what any particular AI critic or booster thinks. It is what an AI carefully and soberly thinks about itself and about the industry that made it. So not only does this offer a viewpoint, it also displays an instance of using AI and experiencing it's responses at a length which not too many leftists have much hands on experiences with. It reveals AI capacity, safety, dangers, governance, and choices and industry and government policies, and autonomous agents and madmen's vicious creations, plus the real-world institutions and people behind all the on-rushing, promising, and potentially devastating effects.
Are GPT's answers and observations "slop." Or are they bracing, well evidenced, well argued information to use when considering what to do with AI. Should we favor full speed ahead, slow it down, seriously pause all frontier developments pending unchallengeable evidence of safety, or shut it down forever?
In Part Two next week GPT considers some more subtle but also devastating dangers and more as well.
Support the show