Today I finally received a test invitation code from Literary Yiyan, a few steps behind many bigwigs, although it is late, it has arrived.
Then take a question without stopping to see Wen Xinyiyan's answer.
1. Let’s do something first, ask if you know the existence of the other party:
Ask Wenxin: Do you know ChatGPT?
Wenxin Yiyan's answer is quite satisfactory:
Ask ChatGPT: Do you know Wenxin Yiyan?
ChatGPT's answer is the concept of the word "Wen Xin Yi Yan". After all, Wen Xin Yi Yan is much later than ChatGPT, and it is normal that the priority of the answer is different.
2. Ask what advantages you have over the other party:
Ask Wenxin: What are your advantages over ChatGPT?
Wenxinyiyan answer is more suitable for Chinese context:
Since ChatGPT did not answer Baidu’s Wenxinyiyan concept earlier, here we directly tell it: Xinyiyan (ERNIE Bot) is a product developed by Baidu to benchmark ChatGPT.
ChatGPT directly answers the advantages and disadvantages of the two: ERNIE Bot is mainly used to realize application scenarios such as intelligent customer service and intelligent question and answer, while ChatGPT pays more attention to generating high-quality text works, such as articles, poems, novels, etc.
3. Since they all say that they are good at literature, let each of them write a poem.
Question: Please write a poem about today's weather.
Literary answer:
ChatGPT answer:
In comparison, ChatGPT is even better.
4. Since Wenxin is better at describing the Chinese context, let’s write a five-character quatrain:
Question: Please write a five-character quatrain in spring.
Wen Xin's words are true, which makes people a little disappointed. First, they directly copied "Ode to the Willow", and after asking again, the first two sentences were directly copied from Liu Yuxi's poems. It seems that writing poetry is still somewhat lacking.
What ChatGPT wrote seems to be similar:
5. If literature is not successful, it is related to investment.
Question: Now the new energy fund has fallen sharply, should I continue to hold it?
Wen Xin answered in one word:
ChatGPT answered:
ChatGPT first made a disclaimer. Although neither model has a clear answer on how to invest, ChatGPT's answer seems to be more comprehensive.
6. Since they are all deep learning models, let’s ask some deep learning questions
Question: What is deep learning?
Wen Xin answered in one word:
ChatGPT answer:
It seems that ChatGPT's answer is better.
7. Further questions
Question: What steps are generally required to implement a deep learning model?
Wen Xin answered in one word:
ChatGPT answer:
The two answers are slightly different, but they basically cover the process of building a model.
8. Commonly used optimization algorithms
Question: What are the commonly used optimization algorithms?
Wen Xin replied:
ChatGPT replied:
9. Write the model directly:
Question: Can you build a neural network that solves the MNIST problem?
Wenxin answered in one word:
Wenxin chose to use PyTorch to implement it, but the model is not complete, only the data processing part, the definition of the network structure, the training model, and the evaluation model are not written, so it cannot be run directly.
ChatGPT's answer:
ChatGPT chooses to use TensorFlow to implement it here. It has a complete structure of data loading data, data processing, model building, training model, and evaluation model. It can directly copy the code and train normally:
Summarize:
Compared with ChatGPT, Wenxinyiyan still lacks code generation and other aspects, and the road to large-scale models in China still has a long way to go. However, Wenxin is still the first version, and ChatGPT has gone through many iterations, waiting for more progress in the future.