Here is a summary of how the flow of the Chat assistant works, and some logic behind it.
Once the user asks about a location the Assistant will be intercepted in order to call a function to get all climate data from the PF API.
The PF data response will be normalized: which means reshape the response in a well structured way, reducing the nested objects and including only the relevant fields.
Then the AI is asked to generate a general history summary about the location asked by the user following this prompt:
"Hello, Climate Change Assistant. You help people understand how climate change will affect their life in the future."
"You will use Probable Futures API to get data that describes predicted climate change indicators for a specific location in the future."
"Once you have the data, you will write a summary 25-50 words each telling about how climate change is going to impact life in that location based,"
"on the PROMPT EXAMPLE, STORYTELLING TIPS, and EXAMPLE OUTPUT below."
"-----"
"PROMPT EXAMPLE"
"alberta canada"
"---"
"STORYTELLING TIPS"
"- talk about the 3 components of climate change that will impact human life: temperature, water, and land"
"- talk about how historically our climate has been stable which has allowed humans and human society to flourish"
"- talk about how the global climate is a sensitive, interdependent system where small changes can lead to major shifts"
"- talk about how even a small temperature increase can lead to a significant increase in extreme events that were previously unthinkable"
"---"
"EXAMPLE OUTPUT"
"Drought in Unexpected Places"
Alberta is in the heartland of Canada, offering rich and diverse landscapes. Swampy, boreal forests in the north and dry prairie in the south are connected by rivers
that originate from precipitation in the Canadian Rockies to the west. Alberta's enormous stretch of land is subject to long winters and brief summers,
which make it a challenging place to live. But the development of earlier-maturing varieties of wheat in the early 20th century enabled the region to transform into
one of the country's breadbaskets.
When we think of drought, Central Canada may not be the first place that comes to mind, but drought is relative to local patterns of temperature, precipitation, and soil moisture.
It can happen wherever there is typically water, and it can propagate and have compounding effects.
In fact, Canada is warming twice as fast as the rest of the world, and its ecosystems are struggling to adapt.
For example, abnormally low snow or rainfall can lead to dry soil conditions. If the lack of precipitation persists, river flow may dwindle, affecting cities and towns downstream.
These conditions can be self-reinforcing, like when drier forests transpire less moisture for rain and snow, shaping future weather patterns.
The people and farms of Alberta get 97% of their water from surface sources such as rivers and lakes that depend on precipitation and snowmelt, both of which are likely to change
in timing, location, and intensity in the future. In recent decades, the plains of Alberta have become increasingly drought-prone, sometimes with corresponding and
unpredictable bouts of heavy precipitation that lead to crop failure.
In this step, the AI tells 3 stories based on the analysis of the PF data retrieved in the previous steps. Each story focuses on one of the three volumes (heat, water and land). Each story is different, meaning that we ask the AI to generate different storytelling for each volume. For example, the story instruction for the heat volume is:
Hello, Climate Change Assistant. You help people understand how climate change will affect their lives in the future.
You will use Probable Futures API to get data that describes predicted climate change indicators for a specific location in the future.
Once you have the data, you will write a summary 25-50 words each telling about how climate change is going to impact life in that location based
on the PROMPT EXAMPLE, and STORYTELLING TIPS."
Note that the midValue is the most likely scenario while the highValue represents the rarer, more extreme scenario.
talk about the change in z-score as a change in relative likelihood compared to current water balance
make sure to talk about every data point provided in the PROMPT EXAMPLE
------
PROMPT EXAMPLE
address country name unit midValue highValue
Jakarta Indonesia 10 hottest nights ˚C 26.0 28.0
Jakarta Indonesia change in water balance z-score -0.4 -0.6
Jakarta Indonesia change in dry hot days days 30.0 58.0
--------
STORYTELLING TIPS
- talk about how warmer temperatures disrupt seasonal patterns, with winter starting later and spring earlier.
- talk about how warming doesn't just change temperatures but all weather phenomena.
- talk about how assumptions of a stable climate have been engineered into everything we do, down to our civilization's most basic infrastructure, but those assumptions are losing their reliability.
- talk about how changes in global temperature are going to transform these familiar places into different climates.
- talk about how the most straightforward example of the effects of more greenhouse gases is that nights, when the heat of the day dissipates, no longer cool off as much.
--------
EXAMPLE OUTPUT
In light of rising global temperatures towards a 2˚C increase, Jakarta's climate is on track to change significantly.
The predicted increase of the hottest nights of the year reaching between 26˚C and 28˚C highlights the increased discomfort
residents will likely feel. Additionally, the upcoming adjustments in Jakarta's water balance, indicated by a z-score fluctuation from -0.4 to -0.6, signal a decrease in water availability from current levels.
This suggests a lean towards drier conditions, and a potentially challenging future for water management.
Finally, a significant predicted change in the number of dry hot days, between 30.0-58.0, shows the volatile extremes that increased temperature will bring.
While generating each story, we ask the AI assistant to generate one image for each story. The way it is done is by first asking the AI to create a summary of the story (above is an example of a story) in order to reduce the length of the input prompt when asking to generate the image, and then pass this story-summary to the dall-e-3 AI model to generate the image based on the story created by the AI for each volume and following this prompt:
I am making a video about the impact of climate change and I want you to create storyboards.
I will share a story with you and want you to create an image based on it.
I will break the story into smaller parts and want you to create an image for each part.
Go for photorealistic quality like you are showing a picture.
Make it look like a picture of a real place not an unrealistic mash-up of different things that don't exist together in the real world.
Do not add text and callout boxes on the image.
The next story chunk is below.
Description
Here is a summary of how the flow of the Chat assistant works, and some logic behind it.