After what happened last night in conversation with OpenAI’s chatGPT, I am truly disappointed with the claim that Large Language Models have intelligence or understanding. I discovered that Large Language Models have no intelligence at all, contrary to what many people like to believe. Above all, the entire experience worried me about the integration of these models into agentic systems in risky scenarios. No doubt LLMs have utility but they have no intelligence or understanding of what they are doing. Fortunately, I recorded the entire conversation. I tried something that 5–8 years old kids can do but for the entire 30 minutes, ChatGPT disappointed me and convinced me of several things:
- LLMs have zero intelligence or understanding of what they are doing. Unlike what many people say — “In order to predict the next word, they should have some sort of understanding” — this is completely wrong based on what happened.
- Because of point no. 1, they should be used very carefully in only predictable situations and integrated into agentic systems with careful consideration. Companies looking to integrate these systems should not think of them as intelligent.
- AGI/Superintelligence is not coming anytime soon because even basic intelligence is not here yet, despite billions being spent.
- They are just regurgitating information and nothing else.
- AI is a very heavy term for LLMs. LLMs are not AI instead they should be categorized as PI (Pareidolia of Intelligence).
Throughout the entire conversation, I kept thinking, “But why?” If this is what is claimed to be the precursor of AGI, then we are not getting there anytime soon. I thought this might be the case with ChatGPT models only but the same thing happened with Anthropic’s Claude and Google’s Gemini. They equally disappointed me but I mainly talked to chatGPT.
Anyone who is remotely familiar with AI should watch the entire conversation and see for themselves what happened. This is an eye-opener and may help researchers recognize the downsides, as well as strengthen the voices advocating for AI safety.
[Models: GPT-4o and 4o mini]
You can try on your own and let me know!!
The entire recording is available here:
Share this with others.
