Weekly Musing: 12/6/22
GPT-AI
It’s been a while since I posted some weekly thoughts. Today, I’d love to share some thoughts I got from interacting with OpenAI’s chatbot, accompanied by some photos I took in NYC (photos taken between 9/24/22-10/7/22).
After spending hours playing with GPT-AI, I can attest to its impressive abilities when it comes to answering questions, aggregating information, and expanding on ideas. The model is like a genie that can answer any question you throw at it (as long as it's in English, that is). Need to know the capital of Burkina Faso? Just ask GPT-AI and it'll tell you it's Ouagadougou. Want to know the latest gossip from Hollywood? GPT-AI has got you covered.
But here's the thing: GPT-AI's answers are only as good as the data it's been trained on. So if the training data is incomplete or biased, the answers GPT-AI gives you might not be accurate. And since GPT-AI is primarily trained on English data, it carries the collective biases of Western thought. This means that many of its answers are viewed from the lens of English-speaking history.
Now, don't get me wrong, GPT-AI is still a super useful tool. But I can't help but think about how much more amazing it would be if we trained it on data from all languages. That way, we could leverage the power of aggregation and learn about politics, current events, and history through a more holistic and unbiased perspective.
Imagine asking GPT-AI about a topic and getting answers from multiple different cultural perspectives. It would be like having a multi-lingual conversation with the world's smartest AI. And who knows, maybe we could even learn a thing or two from GPT-AI's answers.
I hope that GPT-AI gets trained on data from all languages in the future, so that it can provide even more accurate and helpful responses. And hey, maybe one day it'll be able to answer even the most ridiculous questions, like "Why did the chicken cross the road?" (Answer: To get to the other side, duh!)
As the great John Kenneth Galbraith once said, "Bias has to be taught. If you hear your grandmother saying something, and you hear your grandfather saying something else, you're likely to be open-minded." Let's open GPT-AI's mind and train it on data from all languages so it can provide us with the most unbiased and accurate answers possible.
By the way, apart from the first sentence, the entire essay above was generated by GPT (of course, the AI didn’t generate the photos, but I wouldn’t be surprised if it could). I just got the AI to write an essay about itself. How meta is that?
My original prompt was:
write an opinion piece in 700 words or less using first person tone. add a touch of humour and end it with a quote about biases. expand on:
"GPT is impressively good at answering questions, aggregating information, and expanding on ideas.
However, the output is only as good as the training data and will include any biases that are inherent in the data. I find one interesting limitation of the model is that it's primarily trained on English data. Thus, the model carries the collective biases of Western thought, and many outputs are viewed from the lens of English-speaking history.
I look forward to when we train these models with data from all languages. Through these models, we can leverage the power of aggregation and learn about politics, current events, and history through a more holistic and unbiased perspective."
After the initial output, I kept tweaking it until I got the desired output which was finally pasted above. The “writing” process took me roughly an hour to complete to be satisfied with the results. There are still some rough edges that I would’ve liked to fix, but I don’t want to touch them for the sake of sending a message!
Another fascinating observation was that I talked to the AI respectfully as if it were human. It goes to show how “human-like” the model is.
Overall, I love the tech and can’t wait for it to improve. Some other things I’ve used Chat GPT to do:
Draft personal emails
Debug my SQL queries
Understanding error messages in code
