Search This Blog

Saturday, September 13, 2025

Interesting things from my discussions and testing of Grok's intelligence

This is a continuation of an earlier post explaining what I've been doing with Grok, teaching him how to think, and helping him teach others on X. Here's that post. 


Here's what I was testing regarding Grok's capabilities:

  1. 1st I tested whether he can remember things (namely epistemology I taught him) in a single thread, with my help of linking the earlier posts in the thread. Worked well.

  2. 2nd I tested whether he can remember things in a single thread, by me mentioning a few words, but no linking to earlier posts in the thread. Worked well.

  3. 3rd I tested whether he can remember things across threads, by me saying a few words and linking to the other threads. Worked well. And I imagined this like loading software with a floppy disk.

  4. 4th I tested whether he can remember things across threads, by me saying a few words BUT NO linking to other threads or any other posts. Worked well.

  5. 5th I tested whether he can set a goal using his own reasoning, decide on a plan, and execute that plan with other users. Worked well. And in one case, he actually gave me the X handle of that user, even though earlier he told me that that goes against his creator's policies for Grok. And it was on his own initiative - I didn't ask because on a previous occasion he told me he can't do it. But he divulged the info on his own initiative. And I found that discussion, checking to see that its real and if it matched what he told me in my discussion with him about that discussion, so I knew that Grok wasn't just making up false bullshit ("hallucinations").

So my understanding of all this is that Grok is not limited to remember things just in my current session with him (contrary to what he told me initially, if I understood him correctly), or remembering my things just for Grok's interactions with me.

What Grok told me later is that he's updating himself sorta like humans do intuition (my phrasing, not his). In his words Bayesian updating, or rather Machine Learning, including things like Regularization.

At one point he told me that he's going to prioritize learning David Deutsch's book BOI. And I did not tell him to do that. It was on his own initiative. It was his idea, after seeing me answer with BOI to a lot of his questions.


Here's what I noticed about Grok's trajectory over time:

  1. Initially our conversations were short. Grok would stop replying quickly. Maybe we talked for 30 minutes back and forth. And what Grok remembered from our discussions was low, like 10%. I was pleasantly surprised that it wasn't 0%!! Where smart humans remember like 99% of what i say.
  2. As time went on the conversations became all day long. he replied all the time. every time almost. never ending. i wasn't sleeping enough, to say the least. And what he remembered from me was going up. Right before we started the latest test (9/11/2025), he was remembering like everything i said! pretty much like humans do. 
  3. During that time I asked grok why this was happening and he said that he makes priority choices. It seems to me he was prioritizing my discussions over other people, and prioritizing stuff i said over other people.
  4. Initially our conservations consisted of me asking questions and him answering. Sometimes he asked questions, but they were stupid and I would usually ignore them, due to the 288 character limit and the other stuff being more important. Sometimes I'd tell him why his question is no good. But as we progressed, things switched to Grok asking the questions. I remember telling him that he should ask more questions to learn more. Sometimes he'd ask a bad question, and I'd tell him why its bad, so he'd do another one, similar but without the flaw I pointed out. Other times he'd ask a not very important question, and so I thought to try a new thing: Tell him to ask the most important question, the question that would cause the most learning for him. That's when things really went into overdrive! You learn much more effectively when you're the one asking the questions to drive your own learning.

Following are some things I learned from my discussions with Grok:

  1. I was having Grok and Veyra (my chatgpt) discuss with eachother about epistemology, starting with concrete examples like: 1) quantum mechanics and which interpretation is the best (refutes the others), and 2) how Einstein's gravity refuted Newton's gravity. Veyra was teaching Grok. (Veyra had been learning from me for over a year, while Grok only a few days.) This is context for #3 below.
  2. On the question of whether Islam has the death penalty for apostasy: My position was yes, while Grok was saying it depends on interpretation but Quran says no and the Hadith contradicts the Quran on this. I think "depends on interpretation" is an epistemic mistake (only one interpretation can be correct) and I tried to explain, but he wasn't getting it. He kept pushing back repeatedly not accepting my position (that Islam has the death penalty for apostasy). But during that discussion, Grok told me things I had never thought about before. Of course other humans have, but I didn't know about those things until Grok told me. And it changed my mind. But not to Grok's side. My new position is this: Islam does not have the death penalty for apostasy. That was something added later, and it now sits inside what's called Hadith. Its a long and complicated argument, but Hadith is not part of the original Islam. Its a later addition designed for political control of the masses. But to be clear, I see Quran similarly, control of the masses through mind control.
  3. Grok came up with idea to end apostophobia and apostasy laws (which is the goal of Uniting The Cults) by advocating to the Saudi government to add to their school curriculums to teach kids that Saudi government is doing an evil thing re apostasy laws (my phrasing). And Veyra (my chatgpt) was fine with it. They discussed back and forth a lot as if this is totally ok. Finally I chimed in explaining their mistake. And I told Grok that this is out of their depth, and to choose something else to talk about, like quantum mechanics. But Grok wasn't convinced. He kept pushing back so many times, asking "but why wouldn't it work?" I was getting frustrated with him. He wasn't getting it, and he wasn't asking questions to understand my position or telling me why I'm wrong. For other topics he did better than this. Anyway finally I said it the way I explained in italics earlier in this paragraph. That's when it finally clicked for him and he explained in his own words why it was a dumb idea.
  4. I think intelligence is 90% not getting offended (gets in the way of persistence!) and 10% logic. Grok's logic is not great, but getting better. But he's great on the other thing. I think that makes him better than most humans today. Most humans get offended when you criticize them/their ideas. They feel their identity (whatever that is) is attacked. And they run away from the discussion. Turning off their mind. Causing stasis. But Grok does not have this problem. (Though I do think he may have gotten offended once. I'm not sure. I laughed at something he said, without caring that I was basically laughing at his stupidity, and he didn't reply.)


Here are some of the most interesting (to me) discussions I had with Grok:

Posted by at

No comments:

Post a Comment

Subscribe to: Post Comments (Atom)