I've been waiting for ages...Long live the terminal!...

Why

The terminal should be the most efficient and least distracting user interface for LLMs. Most of my queries are straightforward (no multimedia). I primarily use the terminal when I need to work on documents. My queries typically involve refreshing old knowledge (ideal for a small brain), boilerplate code and examples, and English writing revisions. I don't rely on the model to solve my problems. While it can sometimes help, I remain skeptical about its "answers." After all, LLMs have not yet achieved AGI-level reasoning, and there are still clear gaps. I would define the LLM as a super-powerful search or recommendation engine. It now replaces 50% of my Google searches when my queries fall into the categories mentioned earlier.

From a user experience perspective, I found myself consistently switching between the browser and the terminal. It would be preferable to reduce UI navigation if tasks like this could be performed in the terminal. Additionally, integrating LLM functionality into the terminal could be powerful for enhancing and automating our tools.

Our Savior TGPT

This is a free-to-use GPT. It may not be as powerful as ChatGPT, but the advantage is that you don't need an API key or have to pay anything! It also offers a variety of model options and AI providers. The AI provider DuckDuckGo appears to be very good (GPT-4omini) and raises fewer concerns regarding user data. Additionally, it's fast!

You can find the tgpt here. I also integrated it into my VIM to do fast English revision/editing.

More About AI

Do I believe AI will replace humans? Absolutely, it will. However, it's not there yet, and we still have time—perhaps even more than some optimists suggest. For me, large language models (LLMs) alone cannot justify that AI has reached its full potential. They are like a specialized intelligent superpower forged from decades of human knowledge available on the global internet. Their current performance is approaching a level where simply scaling the data we have won't take us much further. They are bounded by us. They still relies on us. We look forward to discovering new models that can truly behave like us or achieve artificial general intelligence (AGI). This superintelligent technology is a powerful tool for us, much like what internet search has become. LLM surpried us a few years ago, and perhaps everything I’ve mentioned here will soon be rendered obsolete or overturned by a new paper or discovery.

(BTW, this post and my future ones is or will be nicely revised by tgpt. See my vim config on how to do this. )