ChatGPT, the language wizard, might just be the hero in disaster response.
A recent study led by the University at Buffalo has figured out a smart way to use ChatGPT’s language skills to pinpoint locations in distress signals on social media during natural disasters. The study’s lead researcher, Yingjie Hu, is stoked about the potential of this tech to aid first responders. Traditional emergency systems get swamped during disasters, and people often resort to social media for help. But here’s the snag – manually sifting through social media feeds is too much for first responders.
So, what’s the game-changer? The research team, including brains from the University of Georgia, Stanford University, and Google, wants to automate this process. They're thinking AI systems could sift through social media user-generated-data for emergency services. Hu gets that folks are skeptical about big language models like ChatGPT, but he's all about the silver lining – using their powers for good.
The cool part? They infused 'geoknowledge' into ChatGPT. Regular models fumble with location data, messing up addresses and all that. To school ChatGPT, they fed it real tweets from Hurricane Harvey victims. Results? The geoknowledge-boosted ChatGPT aced it, beating out regular models and even giving named entity recognition (NER) tools a run for their money.
But here's the catch – ChatGPT needs good prompts to shine. It won't figure out a stretch of highway between exits unless you tell it to. The key is precise instructions.
The team kicked off with older models, like GPT-2 and GPT-3, and later roped in GPT-4 and ChatGPT. They reckon their method can adapt to new GPT models rolling out in the future.
While there's more work to do, like turning ChatGPT's location data into useful info and filtering out the noise, Hu dreams of a smooth partnership between humans and AI. Let the machines handle the heavy lifting so emergency managers can focus on what they do best – saving lives.
In a world where disasters are on the rise, having ChatGPT in the emergency response toolkit could be a game-changer, making things more efficient and, ultimately, saving lives.
Photo: DIW-Aigen
Read next: Cornell's Cascade Platform Accelerates AI, Cutting Costs and Boosting Performance Dramatically
A recent study led by the University at Buffalo has figured out a smart way to use ChatGPT’s language skills to pinpoint locations in distress signals on social media during natural disasters. The study’s lead researcher, Yingjie Hu, is stoked about the potential of this tech to aid first responders. Traditional emergency systems get swamped during disasters, and people often resort to social media for help. But here’s the snag – manually sifting through social media feeds is too much for first responders.
So, what’s the game-changer? The research team, including brains from the University of Georgia, Stanford University, and Google, wants to automate this process. They're thinking AI systems could sift through social media user-generated-data for emergency services. Hu gets that folks are skeptical about big language models like ChatGPT, but he's all about the silver lining – using their powers for good.
The cool part? They infused 'geoknowledge' into ChatGPT. Regular models fumble with location data, messing up addresses and all that. To school ChatGPT, they fed it real tweets from Hurricane Harvey victims. Results? The geoknowledge-boosted ChatGPT aced it, beating out regular models and even giving named entity recognition (NER) tools a run for their money.
But here's the catch – ChatGPT needs good prompts to shine. It won't figure out a stretch of highway between exits unless you tell it to. The key is precise instructions.
The team kicked off with older models, like GPT-2 and GPT-3, and later roped in GPT-4 and ChatGPT. They reckon their method can adapt to new GPT models rolling out in the future.
While there's more work to do, like turning ChatGPT's location data into useful info and filtering out the noise, Hu dreams of a smooth partnership between humans and AI. Let the machines handle the heavy lifting so emergency managers can focus on what they do best – saving lives.
In a world where disasters are on the rise, having ChatGPT in the emergency response toolkit could be a game-changer, making things more efficient and, ultimately, saving lives.
Read next: Cornell's Cascade Platform Accelerates AI, Cutting Costs and Boosting Performance Dramatically
