🍓Testing AI Reasoning Capabilities
ChatGPT01 aka "Strawberry" represents a new set forward in reasoning capabilities in AI Agents - more specifically ability to articulate a line of reasoning. The full implications are not yet clear, but may offer pathways to greater transparency and human control.
OpenAIs "Strawberry" also known as ChatGPT o1 is designed to provide human-like reasoning capabilities.
- I tested its ability to reason through a problem involving the design of an investment fund. This represents a complex problem that requires the ability to look at multiple aspects of a scenario and think long term.
- I then tested its abilities by asking ChatGPTo1 to devise an alternative strategy for solving 2048 a popular game.
Test #1 Design an investment fund that will sustain over time in spite of volatility and regular drawdowns.
What's different from earlier versions of ChatGPT
In Strawberry, the answer outputs start with a new phrase "Thought for (time period)" like this:
Click it and a drop down will appear, detailing the step by step reasoning that Strawberry went through in order to get to a conclusion.
You can select a part of this reasoning and use it in a prompt to ask why Strawberry drew this conclusion. I grabbed one sentence and did this: