ChatGPT's Intentional Weakness

Technology Dec 23, 2022

I'm sure many of you are already aware of what ChatGPT is. It's all over Twitter after reaching a million users within just 5 days of launch. Can you believe that? A million users in just 5 days, surpassing many services that are already part of our lives today.

Now I've been trying it for a while, having fun and finding it quite useful. I tried using it to generate Rust code and get summaries of various works. It still falls short in many aspects—things that I consider quite simple. One must be cautious in following its advice and should always verify, as it can sometimes be misleading. This caution is also explicitly mentioned on the screen that appears every time I open it in my browser. For instance, it failed to provide me with a list of famous people who share my birthday. Instead, it kept suggesting individuals with different birthdates, including Taylor Swift, whose birth month is different from mine.

I discovered a significant weakness that I believe the developers intentionally embedded in the algorithms. While I haven't verified whether it has been addressed or fixed—my information is up to the point of their prior upgrade—it appears that the system deliberately avoids answering questions that could potentially disadvantage a particular group of people, even if you're simply seeking statistical information.

Now, I appreciate the intention—I'm gay and Asian—but it poses challenges for research, a field where we expect it to be widely used. Apart from that, it tends to provide repetitive disclaimers that I wish I could turn off. It consistently advises that everyone is different, and what it just said may not apply to everyone. This, I believe, is common knowledge in our globally interconnected world.

I believe the developers are overly concerned about attracting unwanted attention, fearing their product being labeled such as racist or sexist. Shouldn't confronting these concerns be a step towards progress? It's widely acknowledged that exceptions exist in every context, proven time and again. This, however, should not hinder us from gathering statistics on which ethnicities might have better immune systems against certain diseases and exploring the possible factors contributing to such boosts. I think it's in our best interest to pursue the truth, to learn more about each other, rather than avoiding topics that may challenge our egos and feelings. There's still much we can learn from one another.

These are just my thoughts. I believe it would be best for everyone to engage in civil discussions on these matters so that collectively, we can decide the best way to move forward. I also hope to see more initiatives from other companies, such as Google, Apple, and Amazon, especially after ChatGPT has garnered so much attention.

You can try ChatGPT by clicking here. And don't forget to subscribe!