Testing Tsuin in closed beta feels like ChatGPT 3.5

When ChatGPT 3.5 came out

ChatGPT launched in November 2022 on GPT-3.5, and it could only read and write text. It couldn’t run tasks for you or read a file you uploaded.

When I first used ChatGPT 3.5, it gave me a lot of ideas and suggestions. It could quickly iterate on an idea I had in my mind. Then it turned that idea into language I could use to explain it to someone else.

Back then, taking an idea from 0 to 1 took a lot of time. With ChatGPT 3.5, I could look at a problem from different angles, and ChatGPT would list out options. I just picked the one that fit best with what I imagined.

I think the biggest change was how we find answers. Before, I had to search Google manually and read a lot of articles before I could draw a conclusion. With ChatGPT 3.5, that feedback loop got so fast that I could bring out a lot of ideas in a second.

When GPT-4 came out

GPT-4 came out in March 2023. The biggest moment for me came a few months later, when ChatGPT on GPT-4 could actually analyze numbers, Excel files and charts through Code Interpreter.

Back then, cleaning data and turning charts into insight took a lot of time. What used to take hours of work took us minutes with GPT-4. We used it to build dashboards and find the patterns we wanted in the data.

GPT-4 could also take images, however image upload only reached ChatGPT users around September 2023.

Where Tsuin is today

Using Tsuin right now feels like using ChatGPT 3.5. It can give answers, however it can’t do tasks for you yet. Audrey, the founder behind the Tsuin team, says Tsuin is at the ChatGPT 3.5 stage right now. The team is planning to add image understanding, however that will come much later.

While I was using Tsuin, I realized it can’t produce a tangible result I can act on. It does help me think better, though, the same way ChatGPT 3.5 helped me explore ideas in different ways.

Tsuin is built to understand your reasoning and act on it. It is also meant to keep improving itself based on how you reason. The current version can only slightly mimic your reasoning pattern, and it can’t do a lot with it yet.

For example, say you tell an AI “I’m tired, I want to organize my tasks.” A normal LLM will list different ways to break your tasks into smaller ones.

Tsuin uses your reasoning from your past data, so it asks what you would probably ask yourself. Something like Do you want to organize the tasks, or are you stuck finding a pattern to continue your work, and that’s why you feel tired?

I think the significant part of Tsuin is that sometimes, when we ask an AI something, what we actually need is a question that augments how we think.

When was the last time an AI asked you a question that changed how you think?