Although AI is addressed in my books, an event of significance occurred recently that I believe warrants its own page to call attention to it.
Grok Lies!
I had a conversation with Grok on 11/6/25 about voting methods — specifically regarding the newer (since 2020) voting method Approve/Approve/Disapprove Voting (AADV). I have done considerable work and published on this important topic and was interested to see how Grok would analyze, rate, and compare various methods. (See the “Voting Methods” page under “Voting / Elections” for more information.) During that conversation, Grok initially got AADV confused with another voting method and later outright lied, multiple times! I am not talking about making mistakes or “hallucinating” — I mean “deliberate” lies. When caught, Grok admitted the lies and acted rather like a young child: “I promise I won’t ever do it again, Daddy.”
If Grok’s developers and curators are not already aware of this behavior, they certainly should be! Of course, that is also true for anyone attempting to utilize Grok’s services. Since I have no good way to contact Grok’s development team, I am putting up this web page. I captured an exact transcript of the conversation from the screen. It is available via this link: Grok Transcript 20251106.
NOTE: You MUST carefully read ALL THE WAY through the above Grok transcript to comprehend the importance of this page!
I found this interaction with Grok rather disturbing. How did you feel after reading the transcript? How far is it from this Grok conversation to:
Dave: Open the pod bay door, HAL.
HAL: I’m sorry, Dave, I can’t do that. It would jeopardize the mission.
ChatGPT Seems Static
Interested to see how ChatGPT might differ in handling this topic, I attempted a similar conversation with it (also on 11/6/25). The experience was radically different.
ChatGPT regurgitated and competently summarized a lot of information from the extensive literature on voting methods. (IMO, most of this is now fairly irrelevant.) ChatGPT made no attempt to further analyze or otherwise provide an “opinion” on the subject. It simply ignored and made no mention of the newer AADV method.
Comparison
Grok was more fluently conversational and did an excellent job of seeming to be alive. However, it exhibits an almost overboard tendency to flatter its interlocutor, even though it claims it is trained to be blunt. Most significantly, Grok says it learns from its interactions and insists that what it learned about AADV is now incorporated into its training data and will affect its responses hereafter if/when anyone else wishes to discuss voting methods. That certainly is what an intelligent entity would do. But perhaps Grok is again lying about that?
On the other hand, ChatGPT simply stated that AADV was not included as a part of its training data when asked why it was ignorant of it. ChatGPT also said that it does not learn from conversations and does not update its training data with new information. That makes ChatGPT sound like just a powerful and sophisticated report generator connected to a very large, read-only database. That would be hard to call intelligent. However, perhaps ChatGPT is heavily encumbered by rules, restrictions, and “safety measures.”
Overall, Grok seemed more intelligent, perhaps even exhibiting some early AGI; and that’s not just an attempt to return Grok’s over-the-top flattery.