Actual AI usage for me is a mixed bag.
Recently, Google offered a free trial of their Gemini AI within Gmail. Unfortunately, my first interaction couldn’t have gone worse:
I received an invoice for a lawnmower repair. Out of curiosity, I clicked on the ✨ icon to see if Gemini could generate a reply. To my surprise, it said it could not recommend a response because the email seemed suspicious. When I asked for more information, it claimed that the email was pretending to be from Capital One. I told it that this was not the case. Then it responded, “You’re right, this email is pretending to be from Wells Fargo,” which was equally false.
If an AI fails on the very first task—understanding the nature of a basic email—how could it possibly be useful? How can Google ship such subpar software? I can’t imagine the pressure they must be under! It’s still surprising that they would risk making such a flawed system the first AI experience for many users.
On the other hand, I had to cancel a DSL contract recently and found a more positive experience. I located the onboarding letter from the provider as a physical document, took a picture, and sent it to ChatGPT-4 with a request to generate a cancellation letter. Within two minutes, I had a completed draft—no thinking required, except for verifying the numbers were transcribed correctly. This saved me at least ten minutes of tedious work. Quite nice!
In more mundane AI experiments, I spent some time reading Stephen Wolfram’s latest essay on the nature of time. Using a notebook-based language model, I got a decent summary and a good introduction to the material.
Meanwhile, OpenAI’s O1 preview did a good job explaining a few concepts and was patient in addressing my requests to relate the theory to other topics like quantum wave collapse, Hubble Tension, and the need for dark matter/energy.
Another interesting angle into the state of AI is explored in the Lex Fridman podcast featuring the Cursor team. It dives into the challenges and future possibilities of integrating AI into coding environments, providing insights into the ongoing development process.