Reflection

What did I learn

Artificial Intelligence has a vocabulary. You need to learn the ways it sorts and assosciates terms to be able to use it correctly. This is itself interesting to me, as some of the ways that Chatgpt and Gemini responded to my wording, or things they added in which I never mentioned, betrays a certain implicit bias in the ways these AIs think. Of course we like to think of machines are unbiased, and I am guilty of this too, but AI and even old-school search engines have their own assumptions baked in by the people who made them. These biases are so difficult to notice because they're usually incredibly subtle and align with our own. They aren't inherently a negative, just something to be constantly aware of.

What surprised me

What surprised me about all of the models is the transparency. All of the models I tested in one way or another showed their thinking process and cited their sources. This would be a great way to make an implicit bias explicit, making the answers more usable. From what I searched after testing them out, though, this can also be a bit misleading. The transparency features don't represent a significant decrease in hallucinations. They do, however, make it much more difficult to find the points at which the AI has hallucinated, as it makes the response seem more legitimate at first glance.

What are the limitations

There are many limitations to AI. The one that personally gets to me is it's relative uncontrollability. When you're prompting a text response it tends to be a little vague in it's wording and unfocused in its scope, unless you ask it a very specific question. Same thing with the image generator. The prompt has to be so specific and even then the AI tends to respond in somewhat unpredictable ways. if you desire fine-tuning or specific changes, then it starts to become a back and forth argument between you and the chatbot, as you keep asking it the same thing with increasingly frustrated specific wording.

Page updated

Google Sites

Report abuse