Happy Thursday, Normal People!

This week we’re going back to school a little bit, taking a look each day at some of the basic vocabulary necessary for building an understanding of AI.

ICYMI – Yesterday we had a very special edition as our very first Guest Writer, Jacob Badolato, gave us a crash course on LLMs.

Also, it seems like you guys like these quizzes – and a LOT of you guys got yesterday’s question right! 🤓 

Let’s see if you’ve got it again, with another AIFNP POP QUIZ:

Remember: we’ve got a surprise for those who answer correctly on each poll this week! If that’s you, keep it up, you’re so close!! 🏃‍♂️ 🏃‍♂️

As always, let’s start by seeing what our old pal ChatGPT has to say about today’s term:

What a way with words.

Believe it or not, these models didn’t just come out of the box passing the BAR exam – their billions of points of training data may have developed the knowledge base necessary for these achievements, but it took years of RLHF to refine them to answer correctly on a consistent basis. Remember, these models are sophisticated predictive algorithms, making the best guess as to what the next letter, word, or phrase should be based on their training data.

As effective as this method of refinement is, it is not without its downsides. The debate around bias in AI systems trained with RLHF focuses on how these systems can pick up and even magnify the biases of the people teaching them. If an AI is only learning from a narrow group of people, it might end up unfairly favoring or even ignoring certain viewpoints or types of people. Turns out that even in machine learning, the apple doesn’t fall far from the tree. 🍎 

That’s all for today! We’ll be back again tomorrow with The Friday Opinion covering a highly debated topic: AGI.

Stay normal. ✌️

Onward,

Brady Fowlkes

Subscriber Count 👉 322 🎉 

This week we BLEW past our goal of 250 subscribers by the end of January… CAN WE DO 500?! 🚀🚀

Do you know any other normal people?!
Share this sign-up link with them today!