The article I wrote was only published and created in today's headlines, refusing anyone to be transported to other platforms in any form for release, the original is not easy, plagiarism and manuscript washing will be deeply investigated.
Since Google faced a huge setback, it seems to have become cautious and determined. This time, Google quietly opened a new chapter: Bard, a project known as a benchmark against ChatGPT, has just been officially announced. However, this release seems to be extraordinary, and users who apply to join the waitlist no longer need to go through a long wait and can directly participate in the test.
It has been less than 5 hours since the news was made public. After the test officially began, the actual feedback showed that Bard's results were jaw-dropping. Whether it's emotional or factual accuracy, Bard is at ease, especially in math, and at some point, it even easily outperforms the pre-ChatGPT level. Perhaps sometimes, Bard is even ahead of the current GPT-4 in some ways.
Google CEO Pichai explained the sudden launch: they wanted to get feedback from users as early as possible to speed up Bard's improvement process and make it better.
Without further ado, let's get on the journey of experience. Bard had a showdown with GPT-4. First of all, on the chat interface, Bard took the lead in showing his skills. Confidently and modestly, it introduced: "I'm Bard, your creative and collaborator. I'm not omnipotent, but I'm constantly improving, with your feedback. ”
Users can choose whether to interact with Bard by typing or voice input. First, let's start with a common sense question that begins with "Does lightning strike twice in the same place?" For example, to experience Bard's features.
The answer is presented in the form of a total score, starting with a summary: it is indeed possible that lightning can be hit in the same place twice; He then explained the principle of lightning being attracted to tall sharp objects, and gave an example of the Empire State Building being struck by lightning 25 times a year; Finally, clear suggestions are given, in the event of thunderstorms, avoid tall and sharp objects, try to hide indoors or squat.
Functionally, Bard offers a variety of options, allowing users to choose from multiple draft versions, like, stomp, or rebuild. Plus, users can choose to see more versions and even do a Google search if needed.
Then came Bard's bigger challenge. The first is the philosophical question: "Why can't one step into the same river twice?" Bard's answer is logical, first explaining the meaning of the sentence, involving Heraclitus' famous quote, and then delving into the ideas in it. In contrast, GPT-4's answer appears inadequate.
Next came the math problem, and Bard quickly answered the 356+132 result, which was impressive. Subsequently, Bard faced an even greater challenge, the multiplication operation of 356*132. However, Bard still gave the answer accurately, which is amazing.
In contrast, GPT-4 does not perform well on multiplication problems, although the correct answer can be obtained after correction.
Further testing joke comprehension, Bard demonstrated his understanding of English homophonic memes. However, while GPT-4 can handle these questions well, Bard's answer is more emotional, while GPT-4 is more rational.
Eventually, the time came to test Bard's factual accuracy. Bard misanswered the question about qubits, but after being corrected, it gradually showed the ability to multi-round dialogue, which showed great promise.
However, Bard is not flawless either. As a project based on a big language model, Bard still suffers from bias and misinformation. Google admits that while Bard is powerful, it's not without its flaws. Since Bard covers a variety of information during the learning process, it is inevitable that it will contain biased or wrong content. As a result, it sometimes provides inaccurate, misleading, or false information.
Google emphasized that Bard is not replacing the search engine, but complementing it. While excelling in some areas, users still need to be cautious about their answers.