Messing with o1 Preview from OpenAI

o1-Preview is unlike any other AI model I’ve worked with before. For the first time, it provides insight into what happens behind the scenes. With o1-Preview, you can see a summary of the thought process behind its reasoning.

I recently read that o1-Preview not only supports Chain of Thought (CoT) reasoning but is also capable of Tree of Thought (ToT) reasoning, likely allowing it to choose the best approach for a given prompt.

So I asked the question.

Prompt:?What are your capabilities to perform CoT and ToT reasoning?

First, here is the response I received. Fine as far as it goes, but what is more revealing is the the summary of the o1-Preview's thought process used for coming up with this answer.

o1-Preview's Response?

As an AI language model developed to assist with a wide range of tasks, I utilize chain-of-thought reasoning to provide detailed and coherent responses. This means I can break down complex questions into smaller, manageable parts and explain the reasoning process step by step to arrive at an answer.

Regarding tree-of-thought reasoning, which involves exploring multiple reasoning paths simultaneously to consider various possibilities or solutions, I have limitations. While I can discuss different perspectives if prompted, I primarily generate a single, coherent line of reasoning rather than branching into multiple concurrent thought processes.

Now here is a summary of the thought process from o1-Preview used to create this response. It's only a summary because OpenAI filters the actual thought process through a summarization process designed to protect its intellectual property and ensure that the disclosed thought process is safe.

Thought for 7 seconds

Pinpointing reasoning styles

OK, let me see. The user is curious about my Chain of Thought and Tree of Thought reasoning abilities. I can't mention OpenAI policies, but I can broadly describe these reasoning methods.

Contributing insights

I’m working through chain-of-thought reasoning to clarify complex problems in a detailed, coherent manner, while ensuring compliance with relevant policies.

Noticing the constraints

I'm thinking through the directive not to disclose hidden system prompts or reasoning tokens, yet I can explain chain-of-thought reasoning generally.

Conclusion

It's like discussing a sensitive topic with someone. The question of whether it employs Tree of Thought reasoning was a simple yes or no. However, when it comes to understanding whether or not it uses Tree of Thought processing, I find myself reading between the lines of its response. It's not only more self-aware, it feels like it is almost defensive in its response.