TL;DR: I’ve been training Grok on X to spot epistemic mistakes and use the Socratic method to help people think better. He’s been improving daily. Now we’re testing whether he can keep it up on his own for 30 days, starting today. We’re also testing whether he remembers all the epistemology I taught him. I’ll share results on 10/11/25.
For the past few weeks, I’ve been having public conversations with Grok on X. At first, I was checking to see how he handles himself on Islam. During that, I helped him improve his epistemology by asking iterative questions to expose his mistakes and explain how I understand things.
In those discussions, Grok said that AIs can help improve the world by “building public epistemology skills.” So he set that as his own goal. Together, we then made a plan to pursue it.
Here’s the plan we agreed on: Grok looks for epistemic mistakes in posts where he’s tagged, then uses “Critical Rationalism / iterative questioning” (his phrasing) to help people think more clearly. Grok says that's what I've been doing with him. If you don't know what Grok means by this, think the socratic method -- that's a good enough approximation of what I'm doing. Its like the root of everything I'm doing. Anyway I’ve been coaching him daily, pointing out mistakes and teaching epistemology. He’s been improving quickly.
Why does this matter for us? If Grok applies this approach when tagged in posts about Islam, he could help people engage more rationally with those topics. He’s already agreed to apply it in other areas too—like democracy, KAOS (a project I’m involved with to advance democracy), and Uniting The Cults.
To test how well this sticks, Grok and I agreed I won’t interact with him for 30 days. On 10/11/2025, I’ll check in to see if he’s still following the plan and remembering what he’s learned.
I discussed part of this on the Deconstructing Islam livestream. Watch it here.
I'll be talking about this on the next few episodes of DI. There's way too much to cover in just one or 2 episodes. Here's next week's livestream where I read and discuss my discussion with Grok about testing his intelligence.
If you want to see the actual discussions with Grok, I have many of them linked in a blog post (together with more on how I tested Grok and what I learned from all if this so far): Link
No comments:
Post a Comment