Let's continue with the topic of language models...

A Chinese model from DeepSeek has suddenly become very popular. At first glance, it’s an exact replica of ChatGPT—no more, no less. According to comparison charts, the accuracy of its responses is almost on par with ChatGPT-o1. It feels like the Chinese have completely copied OpenAI’s work and made it open-source. The API is fully backward-compatible with ChatGPT. This means that wherever OpenAI’s model is currently in use, it can be replaced with DeepSeek’s model in a single step. I can say that many models these days can provide polished and plausible responses, but to figure out which ones actually work, you have to dig into the subject matter of the query yourself.

I’ll try not to post anything on this topic for the time being.