Project: Automate Me I

Will the Next Commercial LLM Reach AGI? Who Cares!

Posted by Jason Holzer on Sept. 8, 2024, 7:54 a.m.

Around the time I began pasting code directly from GPT into my projects, I also started diving into discussions about AGI (Artificial General Intelligence) and ASI (Artificial Superintelligence). Like me, many professionals across industries were discovering that, despite the occasional hallucination, these large language models (LLMs) were as good as—or even better than—them at their jobs. In my case, it was undeniably better.

I even reached out to my DBA to say we probably wouldn’t be speaking much anymore, but we should stay in touch. A request came in for me to add a UI to some tools I was developing. I’m not a UI person, so I initially reached out to our UI expert but stopped mid-conversation, questioning why I even bothered. The next day, I used Claude to build a UI and integrate it with the backend. It worked. It took one hour. When I demoed it to my team, they were impressed and assumed I had spent days on it. They even told me to stop working on it—because surely it had consumed too much time. But it hadn't.

I wasn’t the only one experiencing this. Professionals everywhere were realizing, “Hey, this LLM is better than me,” which led to the inevitable buzz online: had we hit AGI? Others, more conservative, speculated that AGI would definitely arrive in the next release. At first, I looked into how AGI is defined and how to test it. That was a rabbit hole. Do we use Wozniack’s robot-making-coffee idea? Turing’s test? Each approach seemed flawed, so I quickly gave up on creating my own test.

Here’s what I learned: When it comes to AGI, forget about the definitions. Be selfish. The question isn’t whether the next LLM reaches AGI—it’s whether it’s smart enough to automate you. Before any of us, as individuals in this simulation, can leverage the grand potential of AGI or ASI, we should be focusing on using these tools to remove ourselves from the grind. Can the latest LLM do everything you need to do? If so, it’s time to automate yourself.

I’m not talking about automating trivial tasks like grocery lists or filing taxes. I mean automating the very thing I’ve done for the last 25 years—writing code, building tools, and earning a living. If I can use LLMs and supporting technology to automate myself, I won’t need to interface with this keyboard anymore. That sounds fantastic to me.

Will I still work? Sure, but it’ll be more like interacting with a ‘Listen and Do’ interface powered by a commercial LLM on the backend, with some custom-made systems sprinkled in.

Over the next 12 months, I’ll be embarking on this journey. Not just to automate myself, but to create tools and services that will help others automate themselves too. The first piece I’ve chosen to offload is the vacation rental host part of me. You see, I have a bit of an addiction to backcountry snowboarding. Resorts just don’t cut it anymore—not because I’m a snow snob, but because you can’t find untracked powder in most resorts.

In 2015, I bought a cabin in the Montana wilderness to feed this addiction. During the summer, I rent it out to subsidize my winter trips. It’s only 7 miles from the Northeast entrance of Yellowstone National Park, so it’s a hot summer rental. But managing guest questions and bookings eats up a lot of time, making this the perfect candidate for my first automation project.

I’m starting with a Vacation Host chatbot that will handle pre-booking, post-booking, and in-stay questions. Eventually, it’ll handle the reservations too. I want to eliminate the need for me to interact with guests manually. Here’s the tech stack I’m using for the project: Python, Flask, JavaScript, FAISS, and OpenAI.

The process looks like this: Web page → Record → Speech-to-Text (STT) → Retrieval-Augmented Generation (RAG) → Prompt → Response → Text-to-Speech (TTS).

If you want to try it out, feel free to role-play as a guest by visiting the website and asking the Host Agent your questions.

Property: Airbnb Link Agent: Roost Host

Next up: Emails

I like to keep my inbox at zero. My inbox is where I manage all the tasks across different parts of my life, and automating this will be complex. I’m not looking for an LLM to summarize my emails—I want it to process each email, understand the necessary actions, and archive the email once the task is done. This project will be far more complicated and time-consuming than automating the vacation rental work.

Log in to leave a comment.