How do you ideate and implement a product in 8 hours that will go viral?
How do you ideate and implement a product in 8 hours that will go viral?
How do you ideate and implement a product in 8 hours that will go viral?
Introducing Pogi, Winner of the AI Challenge at Cozy Computing Hackathon.
Introducing Pogi, Winner of the AI Challenge at Cozy Computing Hackathon.
Introducing Pogi, Winner of the AI Challenge at Cozy Computing Hackathon.
Contributions: Product scope, character personality (prompt), UI, illustrations and animations
Project Link
My hackathon partner was able to write a script that:
1) Allowed the user to do either pick an action via emoji or input a customized action
2) Categorized its own response into one of eight mood gifs (with complementary background color)
3) Picked an appropriate primary emoji (shown above Pogi's head during response) as well as two secondary emojis representing the background of Pogi's response.
For Pogi's personality, I chose to have him identify as a cat with a Smug personality, a loved personality type found in certain characters from the popular video game Animal Crossing.
"Smug villagers are portrayed as arrogant and egotistical, acting sophisticated and thinking highly of themselves."
Before launching Pogi to the public, we had to carefully consider the costs associated with using GPT models. We wanted to ensure Pogi could handle viral-level traffic without racking up an unsustainable bill.
The cost difference between models is significant:
GPT-4: $30.00 / 1M tokens
GPT-3: $1.50 / 1M tokens
Initially, we found that GPT-3 was unable to consistently match the playful and chaotic personality we wanted for Pogi compared to GPT-4. However, the cost of GPT-4 would have been incredibly prohibitive at scale.
To solve this, we used a technique called "multi-shot prompting." By including 5 explicit examples of ideal responses directly in the prompt, we were able to get results that were just as cute and charming from GPT-3 while keeping costs 20x lower.
This scrappy hack let us launch Pogi to the public with all the personality we dreamed up, without a crazy price tag.
Before launching Pogi to the public, we had to carefully consider the costs associated with using GPT models. We wanted to ensure Pogi could handle viral-level traffic without racking up an unsustainable bill.
The cost difference between models is significant:
GPT-4: $30.00 / 1M tokens
GPT-3: $1.50 / 1M tokens
Initially, we found that GPT-3 was unable to consistently match the playful and chaotic personality we wanted for Pogi compared to GPT-4. However, the cost of GPT-4 would have been incredibly prohibitive at scale.
To solve this, we used a technique called "multi-shot prompting." By including 5 explicit examples of ideal responses directly in the prompt, we were able to get results that were just as cute and charming from GPT-3 while keeping costs 20x lower.
This scrappy hack let us launch Pogi to the public with all the personality we dreamed up, without a crazy price tag.
