OpenAI’s new AI model switches to Chinese when reasoning
See Full Size
However, some users and experts using the o1 model encounter an interesting situation. OpenAI’s new AI model starts reasoning in English, even though it starts reasoning in conversations conducted in English later switched to Chinese. o1 from time to time Persian It is said that it can also be transferred to other languages, such as, but mostly to Chinese. This situation, witnessed by a significant number of users, naturally confused them.
OpenAI May Have Trained Its New Model with Data Packets It Received from China
Currently, one of the most important obstacles for artificial intelligence companies like OpenAI is obtaining the data to train their artificial intelligence models. The unauthorized use of all kinds of data on the internet by companies such as OpenAI to develop artificial intelligence models has become the subject of various lawsuits today. That’s why companies are now much more cautious about this issue and are trying to obtain this data through various agreements. The situation encountered in GPT-4o1, OpenAI Chinese data to train new artificial intelligence model It makes us think that he may have made such agreements with his companies. At least Hugging Face CEO’su Clément Delangue’a theory in this direction.
Ted Xiao, a researcher at Google DeepMind He says that OpenAI may have used Chinese data labeling services to improve the reasoning feature in this model, and the current situation may be the result of this.
Hugging Face software developer Tiezhen Wang Starting from a similar point, he says that o1 may now find it easier to use Chinese when reasoning due to the associations it formed while being trained. Wang, who is of Chinese origin, gives the example of how he normally speaks English but switches to Chinese when doing math calculations.
Another theory put forward is that OpenAI o1 may be making such a switch because it sees Chinese as a more suitable language for reasoning. Academician working on artificial intelligence Matthew Guzdialemphasizes that o1 does not know what language is, and for him it is all just text. In other words, he may not even perceive that he is mixing Chinese while reasoning about a question asked in English.
Whatever the reason, this behavior exhibited by OpenAI’s new artificial intelligence model has raised many questions about the training process of such models and reignited objections that there should be more transparency on this issue.
This news our mobile application Download using
You can read it whenever you want (even offline):
