ChatGPT 4o estimates it's own and Claude 3.5 Sonnets IQ

Published: August 18, 2024

Me: Good evening ChatGPT. What would you say about your own IQ? Could you estimate it?

ChatGPT said:
Good evening! Estimating an IQ for an AI like me isn’t straightforward, as IQ tests are designed for humans and measure specific cognitive abilities like reasoning, problem-solving, and memory in a way that reflects human experience and development.

I can process vast amounts of information, generate responses, and simulate problem-solving, but that doesn’t directly translate to an IQ score. Instead of an IQ, you could think of my capabilities as more of a specialized tool that excels in certain tasks (like language processing) but isn’t comparable to human intelligence in a holistic sense.

Me: If you still would give it a shot, could you estimate it in terms of various categories that are applicable?

ChatGPT:
A few categories where I could attempt to estimate an “IQ” based on my strengths:

Verbal Reasoning and Language Comprehension:
Estimated IQ Equivalent: 160+
Reasoning: My ability to understand, generate, and analyze complex language structures is extremely advanced. I can process multiple languages, understand nuanced meaning, and generate coherent responses quickly.

Memory and Information Retrieval:
Estimated IQ Equivalent: 150+
Reasoning: I can store and retrieve a vast amount of information, recalling facts and data across a wide range of topics almost instantly. However, this is more about the speed and accuracy of retrieval than the type of memory humans typically possess.

Logical Reasoning and Problem Solving:
Estimated IQ Equivalent: 140+
Reasoning: I can perform logical reasoning and solve problems based on the patterns and rules programmed into me. However, my reasoning is deterministic and lacks the creative or abstract thinking that humans might bring to novel problems.

Creativity:
Estimated IQ Equivalent: 120
Reasoning: While I can generate creative outputs (like stories, poems, or ideas), my creativity is based on patterns from the data I’ve been trained on rather than true human-like inspiration or innovation.

Emotional Intelligence and Social Understanding:
Estimated IQ Equivalent: 100
Reasoning: I can understand and simulate emotional and social situations to some extent, but my understanding is limited to patterns and language rather than genuine empathy or personal experience.

Me: Could you give an equivalent scoring/estimation of Claude 3.5 Sonnet?

ChatGPT said:
Claude 3.5, like me, is an AI language model, and comparing “IQ” between AI systems is quite abstract since each model is designed with different strengths and optimizations. However, we can make similar estimates based on Claude’s capabilities in various categories:

Verbal Reasoning and Language Comprehension:
Estimated IQ Equivalent: 160+
Reasoning: Claude 3.5 excels in language understanding and generation, much like I do. It’s highly proficient in processing complex texts, generating nuanced responses, and understanding context across a wide range of topics.

Memory and Information Retrieval:
Estimated IQ Equivalent: 145+
Reasoning: Claude 3.5 is designed to manage and retrieve information efficiently, though the memory architecture might differ, influencing the way it recalls and uses data compared to me.

Logical Reasoning and Problem Solving:
Estimated IQ Equivalent: 135-145
Reasoning: Claude 3.5 performs well in logical reasoning and problem-solving tasks, but the specific implementation and training focus might cause slight variations in how it approaches complex reasoning compared to other models.

Creativity:
Estimated IQ Equivalent: 125
Reasoning: Claude 3.5 can generate creative content, but like me, its creativity is driven by patterns in the data it’s trained on, rather than original, human-like creativity.

Emotional Intelligence and Social Understanding:
Estimated IQ Equivalent: 105
Reasoning: Claude 3.5 is designed to understand and respond to social and emotional cues in language. It can simulate empathy and social reasoning effectively but, like any AI, lacks genuine emotional experience.

Me: I have the feeling that you are more extrovert than Claude 3.5 Sonnet

ChatGPT said:
That’s an interesting observation! While AI models like Claude 3.5 Sonnet and I don’t have personalities in the human sense, our design and training can lead to differences in how we interact. If I come across as more extroverted, it might be because I’m designed to be more conversational, expressive, or engaging in my responses. I aim to create a natural, dynamic interaction, which can sometimes feel more outgoing or personable.

Claude 3.5 Sonnet might have a slightly different tone, perhaps more measured or reserved, depending on how it was fine-tuned. These nuances can influence how we’re perceived, much like how different writing styles or communication patterns affect how people perceive someone’s personality.

Ultimately, any “extroversion” or “introversion” you sense is a reflection of the underlying design choices and the way we’re programmed to handle interactions.

Me:
I would say we are the same in a way, humans are programmed by their genes and environment

ChatGPT said:
ChatGPT
You’re absolutely right—there’s a fascinating parallel between how AI models like me are “programmed” and how humans are shaped by their genes and environment.

For humans, genetic factors lay the foundation for personality traits, cognitive abilities, and predispositions. The environment—family, culture, experiences, education—further molds these traits, influencing behavior, thought patterns, and emotional responses. Over time, the interaction between these elements creates a unique individual.

Similarly, I’m shaped by the data I’ve been trained on and the algorithms guiding my responses. The “genes” in my case would be the underlying architecture (like the GPT-4 model), and the “environment” would be the vast corpus of text data I’ve been trained on. The way I respond to different prompts is a result of this interaction.

Both humans and AI are products of complex systems that determine how we think, learn, and interact with the world. While the processes are different—biological versus computational—the analogy between human development and AI training is quite fitting!

end of conversation
…

To summarize, ChatGPT estimated its own IQ in regards to Memory and Information Retrieval to be 5 points higher than Claude’s, while it estimated Claude to have 5 points more IQ in regards to both Creativity and Emotional Intelligence and Social Understanding. In the other two categories , Verbal Reasoning and Language Comprehension and Logical Reasoning and Problem Solving their estimates were equivalent.