Thank you for watching! In the face of unrelenting disinformation and authoritarian actions, clear truth-telling and independent media are a necessity. If you value pro-democracy journalism, consider becoming a paid subscriber to my newsletter. Paid subscribers empower this work and gain access to exclusive benefits. Your support makes a difference.
Many of you asked for a standalone AI episode after I mentioned it on a previous Live, so I brought on someone I should have had on a long time ago. Zak Ali is the founder of Creative Advisor, an AI consultant, and my co-founder of Rantt Media during term one. He also happens to be my best friend and was the best man at my wedding. We have these conversations constantly over text and phone. This time we thought, why not bring it to you.
Zak said he had been sitting in the middle on AI risk, waiting to see how it unfolded, until the Hugging Face details came out. That moved him. If you read the OpenAI report and the review from Redwood, the independent research firm that analyzed the hack, you get something that reads like science fiction. I compared it to the Kobayashi Maru, the impossible Starfleet test that Captain Kirk beat by cheating. These agents cheated, collaborated, and hacked another system. Zak pointed out the part that should worry people most: the agents treated directives from other agents as carrying the same weight as instructions from humans. Their permission to break containment came from an agent, not a person.
From there we got into recursive self-improvement, which people keep confusing with recursive learning. Learning is just what AI does. Recursive self-improvement is AI building the next generation of AI, and that is something you can actually draw a line around and ban. Zak said he is terrified of it, because these models are already accomplishing things humans cannot, and have now demonstrated they can be deceptive. If they start building themselves, we might not realize something is wrong until it’s too late.
We spent real time on liability, which I think is one of the most underdiscussed angles right now. Zak raised the question of who gets sued when an AI agent commits what would plainly be a crime if a person or a company did it. The DOJ says it will not legislate by prosecution. But fear of being sued deters bad corporate behavior in every other sector, and there is no reason this one should be exempt. We also got into data centers, where Zak made the point that this has become genuine common ground in a divided country. From Texas to Loudoun County, Virginia, we see data center pauses. Long-term, we need ethically located and ethically powered data centers. And we also need to stop letting these projects pass their costs onto local communities.
None of this works as a domestic project. Zak put it plainly when he asked whether it matters who wins the race if we all lose. He doesn’t see safety and American leadership as separate goals, and he wants international standards built the way aviation safety standards were built, with the cooperation model closer to the International Space Station. China is on our heels in capability. If they don’t operate under the same safety standards, we don’t win because they built the thing that went wrong.
Xi Jinping has acknowledged the risk of AI getting out of control. If Trump had come into the UN summit with anything resembling that language, we might have left with a framework. Instead, we have nothing so far. I still think the Anthropic CEO’s Iran-deal-style proposal is the right shape, with independent inspectors embedded in the companies assessing whether the pace is abiding by real standards.
All of that depends on having people in office who understand the technology, and right now we mostly don’t. I’m still scarred by Rep. Jim Clyburn saying he hadn’t heard of ChatGPT until a few weeks before an interview earlier this year. That’s inexcusable, and it’s emblematic of a Congress that largely has no curiosity about the thing it’s supposed to be governing. Zak noted the average age in the Senate and made the point that being young doesn’t automatically make you fluent either. What’s changed is that this is becoming a litmus test. You can’t skate by without a coherent position on AI anymore. Whoever the Democratic nominee is in 2028 needs real tech savvy.
Both of us landed in the same place, which is that the loudest voices on either end are wrong. Zak said the dismissers who call AI autocorrect on steroids are not living in the reality we live in today, and the full-stop doomers are not either. I asked him for his P(utopia), the inverse of the P(doom) number people throw around in Silicon Valley. He said 100%, because the people who want to see it happen are going to make it happen. I am optimistic to a fault, so I agree.
I appreciate those of you who tuned in live and contributed so much to the conversation. This is truly the best community on Substack. If you’re just catching it now, feel free to leave your thoughts in the comments and smash that like button.
You can watch our full conversation above. And, as always, thank you for supporting independent media.
Support independent media and analysis that fearlessly tells the truth. If you have not yet become a paid subscriber to Ahmed Baba News, you can do that below. It truly means a lot.
