Three ways I’m using ai — 2024.01
Baseline-setting, note-taking, French, and AI is bad at cello
“The things you procrastinate on are probably the things you should do for the rest of your life.” — Jessica Hische (from Steal Like an Artist by Austin Kleon)
"Write the book you want to read." - Austin Kleon
Intro: With this newsletter, I am going to start sharing three ways I'm using AI. For those who know me, I've been tinkering a lot in my spare time. I am not a software engineer or someone who understands coding. My level of knowledge on this is nothing like the “true pros” who are building advanced features, but I am extremely interested in it and find myself having a lot of ideas. They say write what you want to read, and I guess this is that for me. I hope that it's useful. I don't expect that everyone will want to use it for the exact same things that I do—you will likely have different passions and interests, but perhaps reading some of my thoughts will help connect the dots for you.
1. Setting the baseline: I primarily use ChatGPT for now. Since this is the first post I am sharing in this series, I wanted to briefly describe what tools/services I am using. Like many others, I am using OpenAI's ChatGPT. I have experimented with Claude's Anthropic and have even messed around with creating my own local LLM using an open-source model. I have found that being relatively time-poor, and with most of my use cases being “on the move,” the ChatGPT mobile app, voice dictation functionality, and other features just work better for me. There are advantages to other tools (e.g., I think Claude is better with its “Projects” feature), but they don't function as well for me on the move. Also, I do have a Pro subscription, which I see as well worth the $20/month, and the main benefit is not facing limits on using their advanced models. However, the free version will work quite well for many people.
2. Reflect Notes. Another tool that I'm using is called Reflect Notes. It's my primary note-taking app, and I've been using it since earlier this year. It fulfills a couple of important criteria for me, including easy functionality on iPhones, but its most important feature is the integration of AI into the note-taking app. This takes two primary forms that I use, although there are other features as well. The most obvious integration is when you're writing a note; you have the ability to either highlight text or highlight space in the note, and then do a ChatGPT prompt for something you'd want to add or change in the note. This could be anything from creating an outline for a piece of writing to highlighting existing text and asking it to format it in a specific way. Once that's run, you can quickly press a button, and it will replace the text in line. There are definitely some use cases where that is very convenient and avoids the need to copy-paste between GPT and the note. You can also do this on mobile, which I think is quite cool. Although there are other note-taking apps integrating this kind of AI, the second primary feature that I use I haven't seen anywhere else, at least integrated nearly as well. The second feature is the ability to transcribe voice notes directly into text using advanced transcription AI technology. There's also a useful iPhone lock screen widget, which enables me to take my phone out, press a button, start dictating something, and know that in a couple of minutes on my daily note, I will have a full transcription. It's quite accurate and helps me quickly document an idea I had or a piece of writing I wanted to dictate. For example, I am currently dictating this right now on my way to work. There is a third feature, which allows you to ask questions about your notes in the search function. I don't use this very often, as I don't get many insights from it. It also means that I'd be sending at least a selection of my notes to ChatGPT, so I try to use this sparingly and in a very targeted way.
3. French Flow App. The third usage to highlight here brings you to one of my current obsessions. I'm an American who has wanted to learn and speak another language for a long time, but I missed some opportunities to do so well. The benefit of this is having the ability to nerd out about learning a language as an adult. Last year, I had the opportunity to mess around with learning Greek, and now, since the beginning of this year, I've been focused on what is likely to be a more practical, lifetime language: French.
I've used AI for a lot of things with my language learning this year, but one tool I’m using a lot recently is what I call the French Flow app. I started out with a GPT thread, but now I've created a custom GPT for this. Essentially, this app simulates a conversation partner for pretty much any situation you can imagine or find yourself in, while also integrating vocabulary or phrasing the way you want it.
You can use it to have the AI provide both sides of the conversation, along with an English prompt for what to say in French. This can help you practice figuring out what to say and then expressing it in French. Alternatively, you can reply free-form, which tests your production in another way. You can keep going until you decide to change to a different situation, either by asking for a random one or specifying a scenario, like hiking in the Alps with a friend.
It's magical to practice situations that you think you might be in or that excite you, making you more likely to want to practice your French. It also helps with my current struggle: the nervousness to fully engage with my output in the language after spending a lot of time on vocabulary and comprehension. This feels like a very safe space to practice, unlike real conversation partner sessions I often do online, which feel higher stakes (though still important). And of course, you also receive clear corrections and explanations each time.
You'll probably hear more from me about my French use cases or language learning experiences moving forward, but this was the first one I wanted to share.
Something AI Didn’t Do Well. I've been reading a book called Co-Intelligence by Ethan Mollick. One of the analogies he uses is the “jagged frontier” of AI being like a fortress wall, with walls and battlements extending unpredictably. His point is that it is equally important to determine by trial and error where that wall is at any given moment and what's inside and outside the wall (e.g., what things do LLMs excel at, and what do they really struggle with). Some things that you would expect AI to be very good at, that seem basic, it’s not very good at all. And some things that are very difficult, it’s actually pretty good at.
As an example, this week, I experimented with whether it could help me come up with the best fingerings for a piece I was practicing on the cello. I took a picture of the music, uploaded it, and asked for cello fingerings, and it came back with pretty much nonsense (even worse, confident nonsense). I am not sure if it would struggle with this regardless, or if it had to do that it was interpreting from a visual scan of musical notation. Perhaps the formulaic, almost mathematical logic of fingerings on a musical instrument may play against it.
Thanks for reading!
Fascinating post. As always, I learned something.