All,
ChatGPT, Copilot, Claude, Grok, and others are becoming more and more popular. It’s everywhere, even when you Google something, you might notice their “AI Overview” show up before any actual search results.
Example of an AI Overview.
While these tools can be helpful, that’s just it, they’re just tools and shouldn’t be relied upon for feline diabetes care. In their current state, there are some major risks and limitations with how the technology works.
Most of these tools are powered by LLMs (Large Language Models), without getting too far into the technical weeds, they use the information they were trained on (from various places on the internet) and predict the next sequence of words you would expect. The training data can be from anywhere, human diabetes, incorrect information on random websites, etc. That’s concerning for feline diabetes treatment.
I use these tools quite a bit and have noticed the current version of ChatGPT can act like a “yes man” and try to make you feel good by telling you what you want to hear, even if it’s not the truth. We’ve seen newbies come to us on the forum in a vulnerable state. Overwhelmed by new information and diagnosis from their vet. Caregivers like this might be tempted to ask one of these AI tools for reassurance, and it will deliver that. Whereas on a forum, Facebook, or your veterinarian, real humans will consider the whole picture and factor in real-world safety.
Another concern is that these tools are often overconfident. I asked it a variety of dosing questions recently and it was awfully confident, right or wrong, that confidence is scary with how complicated and nuanced feline diabetes can be. And, there’s no accountability or feedback loop when it is wrong and puts a cat in danger.
A known issue with AI systems is hallucinations, where the tool confidently presents information that sounds factual but is entirely made up or incorrectly stitched together. It’s about how they’re designed as language models. In a medical context, especially with insulin dosing, there’s no easy way for a caregiver to know when that’s happening.
This isn’t meant to scare anyone, just trying to raise awareness. For those who have used AI tools, what has your experience been? @Karolina & Nestle seems like it's great at making comics!
- Tim
ChatGPT, Copilot, Claude, Grok, and others are becoming more and more popular. It’s everywhere, even when you Google something, you might notice their “AI Overview” show up before any actual search results.
Example of an AI Overview.
While these tools can be helpful, that’s just it, they’re just tools and shouldn’t be relied upon for feline diabetes care. In their current state, there are some major risks and limitations with how the technology works.
Most of these tools are powered by LLMs (Large Language Models), without getting too far into the technical weeds, they use the information they were trained on (from various places on the internet) and predict the next sequence of words you would expect. The training data can be from anywhere, human diabetes, incorrect information on random websites, etc. That’s concerning for feline diabetes treatment.
I use these tools quite a bit and have noticed the current version of ChatGPT can act like a “yes man” and try to make you feel good by telling you what you want to hear, even if it’s not the truth. We’ve seen newbies come to us on the forum in a vulnerable state. Overwhelmed by new information and diagnosis from their vet. Caregivers like this might be tempted to ask one of these AI tools for reassurance, and it will deliver that. Whereas on a forum, Facebook, or your veterinarian, real humans will consider the whole picture and factor in real-world safety.
Another concern is that these tools are often overconfident. I asked it a variety of dosing questions recently and it was awfully confident, right or wrong, that confidence is scary with how complicated and nuanced feline diabetes can be. And, there’s no accountability or feedback loop when it is wrong and puts a cat in danger.
A known issue with AI systems is hallucinations, where the tool confidently presents information that sounds factual but is entirely made up or incorrectly stitched together. It’s about how they’re designed as language models. In a medical context, especially with insulin dosing, there’s no easy way for a caregiver to know when that’s happening.
This isn’t meant to scare anyone, just trying to raise awareness. For those who have used AI tools, what has your experience been? @Karolina & Nestle seems like it's great at making comics!
- Tim
