iFlytek Spark, will AI start a prairie fire or go out?

The author of this article will compare the differences between iFlytek Spark and ChatGPT in terms of text, Q&A, translation, logic, code writing and computing capabilities. I hope this article can be helpful to you.

The iteration of iFlytek Spark on June 9th is coming soon. Let’s see if it can become a real Chinese ChatGPT? This article will comprehensively compare and evaluate the differences between iFlytek's Spark cognitive model and ChatGPT from multiple command perspectives.

iFlytek Spark: Following the release of Baidu Wenxinyiyan and many domestic AI platforms, iFlytek Spark released a large cognitive model. Its main capabilities include text generation, language understanding, knowledge question and answer, logical reasoning, and mathematical abilities. , coding capabilities, multi-modal capabilities.

The author is fortunate enough to be an invitee to the iFlytek Spark experience. As an ordinary worker, I will start from the aspects of "text generation, question and answer ability, language translation, logical reasoning, code writing, and mathematical calculation ability". Conduct evaluation and comparison.

(Note: ChatGPT test model is version 3.5)

1. Text generation

Iflytek Spark’s Chinese language understanding ability is superior to ChatGPT to a certain extent.

You can see that the processing results of "Seven Character Quatrains" below are obviously better than the other party; but for text generation in daily work, ChatGPT text is more natural.

Compared with ChatGPT's rigid accumulation of text associations, iFlytek has a deeper understanding of the generation of literary novels and stories. Overall, the Chinese understanding ability is slightly better, but in terms of ordinary text generation, ChatGPT has the advantage in most cases.

About the comparison between generating poems and emails:

2. Question and answer ability

• To answer common questions at work, iFlytek Spark is more general, and ChatGPT is more detailed. In terms of matching, ChatGPT has a higher reference value.
• can understand ordinary context records. ChatGPT has slightly stronger memory capabilities.
• For Chinese common sense questions, especially when it comes to history, culture and literature, Spark’s information retrieval and analysis is significantly better than ChatGPT.

Work question 1:

The following is iFlytek Spark:

Work question 2:

Although table and data retrieval may appear to be processed faster, the data obtained is not authentic and trustworthy. Using this data to answer questions may result in negative growth. Although ChatGPT cannot obtain accurate data, it at least does not provide false data to users, which is an area that Spark urgently needs to improve.

Common sense questions:

3. Language translation

Both of them have literal translation capabilities, but when it comes to Chinese understanding, Xinghuo's translation is more contagious.

For example, in the example below, ChatGPT uses "sprouting up" - sprouting; while Spark uses "emerging in drives" - "emergence", which is obviously a more appropriate expression.

4. Logical reasoning

1. Logic trap. For example, when asking similar questions like "Why didn't your parents call me when they got married?" most domestic language models cannot handle it correctly. ChatGPT's answer is more comprehensive, and Spark directly chooses not to answer.
2. Basic logical abilities are both available. For complex logic questions, the error rate of both answers is very high
3. ChatGPT is generally better than Spark in terms of breadth and depth of logical analysis. However, language models such as ChatGPT will basically make mistakes as long as they leave the common "common areas". However, among the mistakes, ChatGPT's error probability or error degree is smaller than Spark.

Logical question 1:

correct answer:

Logical question 2:

You can see that ChatGPT has already made errors at this time, but it still has a certain accuracy.

Looking at Spark again, starting from the second question, the "cpu" has been dry burned.

5. Code writing

Both have certain code writing capabilities, including code writing, code annotation, and code debugging capabilities.

Since the author is not a professional developer, I cannot actually verify whether it can actually be compiled. However, judging from external data and output results, iFlytek Spark's code writing capabilities have improved to a certain extent compared to when it was first released, and the accuracy of finding errors is also better than before. .

After consulting with many development students, ChatGPT is still better than Spark in terms of overall strength.

The above is an example of Spark error correction.

6. Mathematical calculation ability

For mathematical calculations, all language models have shortcomings. Difficult mathematical problems are very rigorous. As long as one of the steps is wrong, the answer will be wrong.

Computer language cannot understand some meanings as humanly as humans. For example, it may interpret "10" as the two numbers "1" and "0". Solving most mathematical problems requires complex reasoning logic, which makes it very difficult for computers to handle these "quantitative reasoning" problems.

You can take a look at Spark’s mathematical calculation examples:

It can be seen that Xinghuo didn't even understand the meaning of the question clearly, but ChatGPT at least gave an example of one situation.

7. Summary

After fully experiencing iFlytek Spark and comparing it with other products used at work, I found that in the case of problem processing and multiple prompts, the answers provided by ChatGPT are of reference value and breadth of answers. Still better.

Since the launch of Baidu Wenxinyiyan, iFlytek Spark is the best and smoothest domestic AI model product I have ever used. It can be said that it fully deserves the title of "Spark".

However, compared with OpenAI’s technology, there is still a certain gap. At the same time, it's a little less useful as a productivity tool. But it has certain advantages in terms of Chinese language understanding and specific common sense questions.

Objectively speaking, iFlytek has made great progress, and we hope that domestic Internet technology companies can catch up quickly and not let "the West become the trend" and let "Hualiu be the most popular."

This article was originally published by @wangzai product notes on Everyone is a Product Manager. Reprinting without the permission of the author is prohibited.

The title picture comes from Unsplash, based on the CC0 agreement.

The above is the detailed content of iFlytek Spark, will AI start a prairie fire or go out?. For more information, please follow other related articles on the PHP Chinese website!