Deploying AI Applications

The Beginning

I had seen that many cloud platforms provide services for deploying AI applications, such as Cloudflare’s Workers AI, so I tried it out these past few days.

It should be said that deploying AI applications with the help of AI is indeed very convenient 👍.
By the way, I also deployed some API reverse proxies 😎.
However, serverless platforms still have many limitations. It is still not possible to deploy powerful applications (such as: open-webui, lobechat), which require a platform like Vercel.
In addition, claude 3.5 sonnet seems to have become a little dumber recently. Sometimes adding a small feature will crash the entire system 🤣. I blindly guess that Anthropic is using its main computing power to train claude 3.5 Opus 🤔.
DeepSeek 3 is out, and both domestic and foreign reviews are calling it amazing, a light of domestic products. I need to give it a try.

I also deployed a very simple chat in the blog.
There is no historical conversation cache, no user management, no file upload, no knowledge base, no tool application, nothing is recorded, protecting privacy 😋.
It supports streaming output, Markdown rendering, and mathematical formula display. For example, you can ask: What is the formula for the roots of a cubic equation?
The underlying model is ChatGPT-4o-mini, which is okay for general use.
It is estimated that it will not work if the concurrency is too high.