Thoughts on ChatGPT

This blog post was entirely written by me, David Bieber, a human being. I did not consult with ChatGPT or Bieber Bot or any other assistants or conversational agents in the writing of this post. It is likely this will now be a rare occurrence, and that subsequent posts will generally be written with an assistant at my side.

Reminder: This is a snippet

If this were an ordinary blog post, I might set aside some space to bring you up to speed with what ChatGPT is (it’s a conversational assistant – a chatbot – but more sophisticated than SmarterChild ever dreamed she could be). But this is a snippet, a mini blog post written for no audience and with no quality bar to publication, so I will press on without doing so.

Reactions to ChatGPT

Futility. Existential dread. Unsettled. Excitement! There are lots of possible reasons one might be concerned about the release of ChatGPT by OpenAI. For me, the ones that hit me hardest were about my own ability to stay relevant a few years down the line. Bear in mind, when you hear this concern, that it’s not just about ChatGPT’s current capabilities. It’s about the expected advancements that are coming in the next several months or years.

I’m not concerned about being “replaced” wholesale. I am concerned about my own ability to stay relevant and continue to provide (as significant) value professionally and personally a few year’s down the line, heightening competition and expectations, a future need to be always on / always adapting to a world changing faster each year. An economic future full of contradictions I haven’t been able to wrap my head around yet. And a large number of people who are going to have sudden shifts in their lives, jobs, etc.

I previously thought self-driving trucks would be the one Big Shock, and everything else would get automated more gradually (in terms of people/year). I’m in the process of reconsidering.

Expected Advancements

Future capabilities will include web browsing, taking actions on your computer, and accessing your personal files (documents, emails, chats, etc) (with your permission, of course). We’re also going to see major improvements in quality from human feedback, and from a larger model trained on more data. I expect this is coming quite soon. A bit further down the line, integration of multiple modalities (e.g. image generation, audio, video) will come too.

Using ChatGPT with my Roam Research Notes

I use Go Note Go to jot down miscellaneous thoughts. They are often scattered and always ill-formatted. ChatGPT is able to pull out a clean list of TODOs from my notes, unperturbed by the messy format they start in. Integrating conversational assistants into more of my personal documents is going to be immensely valuable. Excited to see what progress on retrieval augmented models brings next year.

Building Products with ChatGPT

Voice assistant: Tonight I prototyped a voice conversational assistant with Twilio using GPT-3.5 (davinci-text-003). I’m excited to try it out with ChatGPT as well. Being able to place a phone call to use the power of a conversational assistant is convenient, especially when driving. There, a phone call is easy but a text based app is not safe.

Medical robo-intake: At the doctor’s office, you typical fill out an intake form which is impersonal. Then, if the doctor is busy, you may not get to tell them all the details that could be relevant to your case. Using a conversational assistant to perform the medical intake could be quite valuable. We wouldn’t be relying on the AI to have robust medical knowledge or make any diagnosis; it would simply ask relevant questions to try to elicit any details that might be pertinent for the doctor to know. The doctor could still see the full conversation, as well as a robo-summary with the highlights.

Personal tutor: One drawback of the current ChatGPT system is it isn’t always factual, making it challenging to use for learning unfamiliar subject matter. I envision a system in which a human tutor can view their student’s conversations with an AI-tutor, and can jump in with corrections and suggestions when needed. Another important aspect of an AI personal tutor will be curriculum setting (involving push messages, not just reactive messages), and goal setting, but the most important aspect is just that you can go down whatever rabbit-hole tangents catch your eye while progressing through the guided lessons.

Teddy Bear Programmer: ChatGPT already provides a great developer experience, where you can ask it technical questions or to perform code-writing tasks. To improve this experience, we can build a system where it always has visibility into the context of your development efforts. E.g. it can see your recent terminal usage or your current editor window. So when you ask it questions, it knows why you’re asking.

The “Clink” Feature for Go Note Go: The idea here is that when you take notes on Go Note Go, and someone else also has taken notes, you should be able to “clink” your two note-bases in order to find points of commonality between you. Then you can hit it off and discuss your common interests! The ChatGPT tech makes implementing this easier.

My Own Uses of ChatGPT so Far

Writing a legal document for a car sharing agreement for my partner.

Writing a blog post on the salt, fat, acid, heat metaphor. Revising the post was done with the help of ChatGPT as well.

Learning about variational autoencoders, about music theory, violin and guitar, and about Rust and building Chrome Extensions, and about topics in medicine, history, and science.

Programming in languages I’m less familiar with and using APIs I’m unfamiliar with.

Reading papers and having ChatGPT summarize sections as I read them. Similarly, having ChatGPT summarize websites and blog posts as I browse the internet.

Extracting clean TODOs from my messy Go Note Go notes.

Generating descriptions for Stable Diffusion, which in turn generates stunning imagery.

Brainstorming uses of ChatGPT, questions to ask people, thoughts on education and psychology, topics in medicine, etc.

Just for fun generating simple stories, bad jokes, poor stand up comedy routines, and silly parody songs.

Excitement for the future

Despite my visceral reactions of futility and existential dread noted above… I’m excited and energized for the future. There’s no doubt this technology is going to provide a tremendous amount of value and it’s going to come quick. Yes, the space of possible futures seems narrower today than it did before the ChatGPT launch. Now we must aim for the best possible future among all those that remain open. And ultimately, does not every passing moment narrow the space of possible futures? The space of proximal possible futures, on the other hand, has now been pried wide open. Machine-learning driven conversational assistants present so many possibilities. My brainstormings so far only scratch the surface. I’m excited to get building, and to unearth more about what this tech is doing to our future.

(Want to build together? Get in touch!)

Discussion 💬