The Purpose of AICBF
Within this book, Flaws in AI Chatbots, abbreviated as AICBF, I am demonstrating the flaws in AI chatbots' intellectual capabilities. by having conversations with AI chatbots about various topics. This book was last updated in Jan 2026.
Purpose
These conversations are about the various things that I have already written in my books. These books are available for anyone to read online in their entirety. I do not exclude, from reading, any search engines and any automation that feeds the "knowledge" that an AI chatbot possesses. These books have been available for a long time, so the AI chatbot should have known about all its contents. I have verified, with subtle and short conversational hints, that it indeed is aware of the contents of the book and its arguments. And yet, at least initially, it misses the "logic"; I have to give hints for the AI chatbot to "see" the logic.
Many of the major flaws in AI's output (conclusions, suggestions and recommendations) are revealed by my subsequent inputs to the AI chatbot. I don't dwell on the minor inaccuracies and minor wrong conclusions. After all, there is a limit to my patience to point out every flaw in its knowledge, reasoning, and biases. There is also a limit to a reader's patience to read about all the ways in which the AI chatbot failed at "understanding". These conversations reveal the scale of those failures. Considering the sheer scale of failures, minor failures are to be expected.
A reader, upon reading these conversations, should be able to see that AI chatbots do not take good decisions. If such an AI were to be blindly or automatically used to take decisions, then expect bad decisions. The consequences of these bad decisions can range from mere irritation to tragic circumstances. It all depends on for what purpose the AI was entrusted with automatic decision-making. This situation may change in the future, but who knows when?
There is one very good objection that a reader may have to what I am trying to reveal. They may say that AI chatbots are "sycophantic". The reader may say "It is going to tell you what you seem to want to hear". However, it is not merely "sycophancy". Here are the reasons: An AI chatbot has an outer software shell that uses underlying AI technologies (like LLM). So, what part of an AI chatbot is "sycophantic"? Is it the outer software shell? Or, is it the underlying technologies? I don't know the answer to that question; for me, it is a blackbox. Regardless of whether it is "sycophancy" or not, the decisions arrived at by AI chatbots are flawed. I pursue uncovering these flaws in the conversation, and the AI chatbot reveals plenty of reasons why its reasoning is flawed, and it is not merely "sycophancy".
The key take away from these demonstrations is as follows: Currently (that is early 2026), there are fundamental problems with AI's decision-making capability. Before using an AI, we should evaluate it and find out whether it is actually good. When evaluating AI, it is critical to look at its failures more than its successes.
However, a reader should not conclude that AI chatbots are useless. There are valid use cases. But, I don't need to list them; there are others doing that already.
Generalities About These Chapters
Each chapter (except the chapter introduction) contains the entire conversation. Each interaction within the conversation is a separate section in the chapter. Thus each section starts with my question (or request or input) and is followed by the response (or output) from the AI chat bot.
My question is always a single paragraph. I have left untouched my own typographical errors in these conversations.
I have also tweaked some formatting of the output from AI chatbot so that it presents well. AI chatbots typically fail at generating the output with formatting that is ready for copy and paste. These are just formatting "fixes"; there is no change to the actual output created by the AI chatbot.
In presenting the conversation, I have eliminated all the links that the AI chatbot provides because they have no relevance to this particular exposition.
I have named the sections as Part 1, Part 2 etc. I would like you to draw your own conclusions from each of these conversation.
I would also like you to conduct similar experiments and find out for yourself how bad AI is at "intellectual activities".