One Weird Trick for OpenAI o1
OpenAI o1 is like having your own graduate-level research assistant. It can handle complex reasoning tasks and break them down into manageable solutions. For instance, if you’re in the process of buying a car, you might wonder how much you would actually save by going electric (rather than relying on what the marketing says). Most people would stop there, but the magic of o1 is that if you take that thought and ask it directly, it will provide the answer for you.
The Challenge: Making accurate assumptions
While o1 is incredibly powerful, it works best when you provide a comprehensive list of assumptions upfront. Since o1 isn’t connected to real-time search, if left to its own devices, it makes poor assumptions. So you have to do it yourself like this:
I want you to calculate the difference in fuel costs and create a model to check it.
Assumptions:
- Model X uses 34 kWh per 100 miles
- 1 kWh costs $0.30
- I drive 12,000 miles a year
- The comparison car is a Lexus LX which gets a combined mpg of 19
- Gas costs $4.25 a gallon
How much do I save in fuel with a model X?
However, identifying and researching this information is tedious. Then, I discovered an even better way to unlock o1’s capabilities.
The weird trick
It turns out GPT-4o is really good at writing prompts. And then I realized GPT-4o is connected to real-time search, so all you need is a prompt like this:
You are an expert prompt writer for o1. I want to compare the fuel cost of a 2022 Tesla Model X to a 2022 Lexus LX. State all of the necessary assumptions for o1 and search for the correct values.
Format:
Question: …
Assumptions:
- Assumption 1
- Assumption 2
GPT-4o then determines what assumptions are necessary and provides the correct values and all you had to ask was “Compare the fuel cost of a 2022 Tesla Model X to a 2022 Lexus LX”, here’s what it looks like:
Question: Compare the fuel cost of a 2022 Tesla Model X to a 2022 Lexus LX.
Assumptions:
- **Annual Mileage**: Both vehicles are driven 15,000 miles per year.
- **Driving Conditions**: The driving pattern consists of 55% city driving and 45% highway driving.
- **Fuel Economy**:
- *2022 Tesla Model X*:
- City: 107 MPGe
- Highway: 97 MPGe
- Combined: 102 MPGe
- *2022 Lexus LX 600*:
- City: 17 MPG
- Highway: 22 MPG
- Combined: 19 MPG
- **Fuel/Electricity Costs**:
- *Electricity*: The national average residential electricity rate is $0.14 per kWh.
- *Gasoline*: The national average price for premium gasoline is $4.00 per gallon.
- **Energy Consumption**:
- *2022 Tesla Model X*:
- Energy consumption: 33 kWh per 100 miles
- **Fuel Costs**:
- *2022 Tesla Model X*:
- Annual electricity cost: $693
- Cost per mile: $0.046
- *2022 Lexus LX 600*:
- Annual fuel cost: $3,158
- Cost per mile: $0.210
- **Location**: The user is located in California, where electricity and gasoline prices may vary from the national average.
These assumptions are based on data from the U.S. Department of Energy's FuelEconomy.gov website.
And when you give it to o1, you get:
Under national-average assumptions ($0.14/kWh and $4.00/gallon), the 2022 Tesla Model X’s annual electricity cost ($693) is far lower than the 2022 Lexus LX 600’s annual gasoline cost ($3,158). In California, where both electricity and gasoline are pricier, the Tesla’s charging cost may rise to around $1,100–$1,200 per year, whereas the Lexus LX 600 can easily climb to $3,900–$4,000 or more in fuel costs—still leaving a significant cost advantage for the Tesla.
Bonus weird trick
If you’re interested in seeing how changing the assumptions affects the outcome, just ask o1 to create a static HTML page in the same thread. This allows you to tweak parameters like mileage, energy costs, or fuel economy directly in a browser, making it easy to see how adjustments affect the results.
Build a static HTML page using tailwind css that allows me to change the assumptions.
Conclusion
OpenAI o1 and GPT-4o, when used together, are powerful tools for solving real-world problems. The key lies in leveraging each model’s strengths effectively.