Yesterday was an interesting day, at 2 p.m., Baidu held a press conference for the generative AI product "Wen Xin Yiyan", and Director Li's speech was slightly nervous.
There are no live product demonstrations, just pre-recorded videos. Maybe it's the fear of overturning like Google, or maybe it's to present the best press conference.
However, during the press conference, Baidu's stock price fell.
As I expected, Baidu was sprayed after the press conference. Many people's views are surprisingly consistent, and they are disappointed with Baidu's "Wen Xin Yiyan", which in turn turns into disappointment with Baidu.
Then, this picture appeared in large numbers in the circle of friends to express a complex mood.
The disappointment seemed very real, as if they had already experienced the product "Wen Xin Yiyan" in depth.
But the fact is that when this group mood began to ferment, 99% of people did not get the invitation code to enter the product interface of "Wen Xin Yiyan".
Therefore, to announce the verdict in advance in combination with a less outstanding press conference is inevitably unfair to "Wen Xin's words".
Immediately afterwards, some people have successfully obtained the invitation code in the evening and have the opportunity to experience it in depth, and then hold the GPT-4 ruler in their hearts and measure it repeatedly.
The conclusions are still simple and crude, rubbish.
Just like the image above, a copycat with a similar appearance and interface but completely different inside.
In my opinion, this is biased.
Although I haven't experienced "Wen Xin Yiyan" in depth, judging from the screenshots of the interface and the quality of answers that have been sent out of the experience report, it is not so unbearable.
If GPT-4 is used to benchmark, then Baidu's "Wen Xin Yiyan" obviously has a big gap. But in terms of general issues, it can at least reach the level of GPT-3.5.
Of course, there may also be a gap from this level.
For AI products, there are two stages, one is to be able to do it, and the other is to do it well.
Whether it's ChatGPT or Wenxin, they have achieved the level of "doing", which is already remarkable. To "do a good job", what is needed is continuous iterative upgrading.
Just like today's GPT-4 can pass the U.S. Bar Exam and enter the top 10%, but the previous generation of products may only be regarded as a primary school student.
Therefore, we need to give Baidu some time to iterate on its own products with technology, data, and money.
I took a closer look at the existing case, and Baidu's "Wen Xin Yiyan" is slightly superior in Chinese processing ability. If GPT-4 can do a better answer, then "Wen Xin Yiyan" has the opportunity to make a more warm answer on this basis.
As we all know, Chinese is broad and profound, and its language complexity is far more than English.
I also saw a message that "Wen Xin Yiyan" has refreshed 93 Chinese NLP (natural language processing) task benchmarks and has repeatedly topped the SuperGLUE global list.
This is a complex language understanding task evaluation jointly published by Google DeepMind, Facebook Research, New York University, University of Washington and other authoritative institutions. It has been applied in more than 60 tasks such as machine reading comprehension, text classification, semantic similarity calculation, etc.
I think what we need is an AI infrastructure that can be used locally, understands Chinese, and can help Chinese people improve work efficiency.
For now, ordinary people can't even register OpenAI accounts, so I naturally hope that Baidu can do a good job.
Everything has many facets, some people complain about Baidu's search engine, some people hate Baidu's medical advertising, and some people feel that Baidu has no lower limit in business.
But it has to be said that Baidu is still the most deeply rooted and invested enterprise in the field of AI in China.
Wen Xin Yiyan is not a product that began to be developed last month, and Wen Xin Yiyan is only a specific application in Wen Xin's large model system.
If you've seen Baidu's Wenxin panorama, you know how big and how long they've been in this space.
Baidu is the first All in AI among domestic manufacturers, and with its own technical capabilities, I think there should still be something to look forward to.
This is not to beautify Baidu, but I also hope to have an AI service that can be easily accessed, convenient to use, and more suitable for the Chinese market.
From the perspective of Internet history, foreign companies are unlikely to become bigger in our market, and Baidu is indeed the most likely to have the opportunity.
Of course, I also hope to see other big manufacturers follow suit. After all, this matter is really very expensive.
Last night, Microsoft released the GPT-4-based office family bucket, which can be called a king fried product.
After its own search engine Bing first upgraded to GPT-4, the new experience brought by office is even more eye-catching.
Perhaps, this is the first time many people have really seen the future of working.
Since then, tools have changed from mechanical to intelligent, and the way people use tools will upgrade from hands to brains.
In the GPT-4 version of the PPT tool, you only need to express your needs in natural language, and PPT will automatically help you make beautiful pages.
If you have an Excel table dataset that needs to be analyzed, tell Excel what you need, and it will automatically help you understand the rules and make corresponding conclusions.
If you have a product meeting minutes document to write about, you only need to describe your requirements to Word, and it can help you generate a document with content and layout in seconds.
Today, the GPT-4 version of Excel does it perfectly.
Since then, we no longer need to memorize complex operations and formulas to operate Excel, nor do we need to struggle with how to phrase and express a piece of text, nor do we need to compete with our brains to typeset PPT.
What we need is to use our wisdom to describe what we want to AI, and the rest to AI.
Some people worry that this means that the value of human work has been replaced?
I think not.
As mentioned above, what humans have to do is to use their wisdom to prepare what they want to describe to AI, give the hands-on things to AI, and keep the brains to themselves.
Therefore, it is a factual trend that mechanical repetitive work will be replaced by technology, but the creativity, judgment and decision-making power of the human brain are still irreplaceable advantages in the short term.
Today's Internet seems to be back to the state of 2010, some excited, some ignorant, some excited, some hesitant, some hungry, some later.
But in 2010, a small number of people already sensed in advance that a great era had begun.
As someone who has lived through that era and been deeply involved in the process, I would say that it feels a lot like it was then, but the logic will be completely different.
In 2007, Apple released the first generation of iPhone, and in 2010, the cross-era iPhone 4 was released.
However, the wave of mobile Internet entrepreneurship and product application blowout around the world only happened after 2012.
It's 2023, it's still early, we'll wait.
Author | Tang Ren uses product thinking to understand the logic behind the phenomenon, former non-famous programmer, now unknown product person.