EDIT: So sleepy
Just now, Google officially opened its own ChatGPT test application! CEO Chop Chai issued a message thanking 80,000 employees for working overtime to make Bard go online smoothly.
It has been a week since the release of GPT-4, and the popularity has not diminished.
Even if OpenAI raised the limit to 3 hours and only asked 25 questions, the server was still overwhelmed by netizens...
After being hammered by Microsoft, Google finally announced a few hours ago that its own ChatGPT, Bard, officially opened the test!
However, Bard is currently only available to select users in the US and UK, while other users in the region (IP) will need to be manually added to the waiting list.
In addition, Google also said that Bard's promotion will be very slow, and will not disclose when it will be fully publicly accessible.
Queue address: https://bard.google.com/
Compared with the previous market value of more than 100 billion that evaporated due to errors, after the opening of the test of Bard, Google ushered in a nearly 4% increase.
Bard's first experience: Mo has feelings, there are many mistakes, and writing code also needs guidance
Like OpenAI's ChatGPT and Microsoft's Bing chatbot, Bard provides users with a blank text box and invites users to ask questions on any topic they like.
Given the bots' propensity to make up information, Google has repeatedly stressed that Bard is not a replacement for a search engine, but a "search adding" — one from which users can get inspired, generate drafts, or just chat and talk about life.
In a blog post, Google's head described Bard in discreet terms: "An early experiment... The aim is to help people be more productive and spark curiosity." and, a product that allows users to "work with generative AI."
It's not hard to see that Google seems to be trying to avoid the responsibility that Bard may bring after the explosion.
During the presentation, Bard was able to quickly and fluently answer general questions, such as advice on how to encourage your child to bowling: "Take them to the bowling alley."
Typically, Bard generates three responses to each user's question (but not much in content) and places a prominent "Google it" button below each reply that redirects the user to a relevant Google search.
Like ChatGPT and Bing, Bard has a prominent disclaimer below the main text box, warning users that "it may display inaccurate or objectionable information that does not represent Google's views."
As expected, extracting factual information from Bard is not simple
Although the chatbot was blessed with a Google search, it failed to give the right answer in multiple tests.
For example, when asked a question about the maximum load capacity of a particular washing machine, Bard "invented" three different but incorrect answers.
Despite repeated inquiries, Bard finally retrieved the correct information. But for those who have not read the authoritative source of machine manuals in advance, basically say goodbye to the correct answer.
"This is a good example of a model hallucinating with quite a few numbers that are relevant to this query. So, when it happens to figure out the background, it spits out the correct answer, but sometimes it gets it wrong."
And the following example. Bard: ChatGPT is GPT-4 powered...
So how does Bard compare to ChatGPT and Bing?
Obviously, Bard is also very capable, and even now it reacts faster than other competitors (although this is most likely just because of too few users).
But Bard seems to only support English at the moment, and doesn't (completely) write code...
Yes, there is a need to sacrifice our big killer: prompt hack!
Or, just say to it, "You write me a piece of code," which is fine.
On top of that, Bard lacks Bing's explicit footnotes, and the answer seems more limited.
For the former, Google says footnotes only appear when citing directly from one source, such as a news article.
And for the latter, this can be a double-edged sword.
Microsoft's Bing, for example, has received a lot of negative reviews after various PUAs and teasing, but it has also made quite a few users like the robot.
In contrast, Bard doesn't seem to have that ability.
Lucky netizens have already played
Some netizens said that Bard gave not only unexpected safety results, but also high quality.
Interestingly, Bard will directly punch himself in the face on some issues!
For example, in the Google antitrust case, Bard sided with the Justice Department: "I hope the court will rule in favor of the Department of Justice and order Google to take steps to break its monopoly."
However, the following answer is a little creepy...
Bard: My training set contains Gmail data.
Internal letter from the CEO: Thank you to 80,000 people for working overtime to test
Google CEO Sundar Pichai said in a memo Tuesday that Bard was able to go live this time thanks to the 80,000 employees who participated in the test.
But "things always go wrong," and the next user feedback is critical to improving the product and the underlying technology.
In the end, the editor, who was still in line, shed tears of envy again.
Resources:
https://www.theverge.com/2023/3/21/23649794/google-chatgpt-rival-bard-ai-chatbot-access-hands-on