Home > Article > Technology peripherals > Bard still can't get into Harvard? Google CEO Pichai predicts that Google will soon cheat, and its mathematical logic ability will soar
After covering it up for almost two months, Bard, the chatbot used by Google to compete with ChatGPT, was finally launched a while ago.
Users’ first experience with testing is that they have no emotions, there are many errors, and it is basically impossible to write code.
A few days ago, former Google employee Jacob Devlin even revealed shocking inside information. Bard was suspected to have been trained using ChatGPT data. However, a Google spokesperson has clearly denied this statement.
Recently, Google CEO Pichai confirmed in an interview with the New York Times that the Bard chatbot will soon be improved and will be supported by "more powerful models" in the future.
Pichai said,
In the next few days, Bard will soon be based on LaMDA The model is moved to the larger PaLM dataset.
We obviously have a more capable model, and soon, perhaps as this technology comes online, we will upgrade Bard to a more powerful PaLM model. This will bring more functionality, both in terms of reasoning and coding.
Recently, the abilities of the upgraded version of Bard have gradually surfaced.
Jack Krawczyk, a Google executive who leads Bard, also introduced the latest developments on Twitter.
With the help of PaLM, Bard's ability in mathematics and logic has been greatly improved.
What does this mean? Now, Bard will better understand and respond to users' multi-step reasoning and math question prompts, and coding capabilities are coming soon.
Krawczyk said, "We are always balancing new features and efficiencies in Bard. This update is among the many improvements we are making to Bard every week. An example. There will be more content to be launched in the future."
As we all know, the large model behind Bard is LaMDA.
Last year, Google shared details about this large language model, using 137 billion parameters to train LaMDA. The PaLM model has 540 billion parameters.
Both models have been developed and grown since early 2022.
This comparison may show why Google is now slowly moving Bard to Palm, which can provide larger data sets and more diverse answers.
In fact, in addition to Bard, Google has also been brewing a new project internally in recent weeks - Gemini (Gemini), with the goal of competing with GPT-4.
Currently, Gemini is jointly developed by two teams, Google Brain and DeepMind.
Even Jef Dean, the head of Google Brain, came to the battle in person and typed the code automatically. It can be seen that this project is of great importance to Google. Metaphor.
In addition, during the interview, Pichai expressed his views on the joint letter from thousands of big guys to suspend the development of AI systems stronger than GPT-4:
A moratorium is essentially impossible without government involvement, because even if Google or OpenAI commit to halting development, there's no guarantee that other AI developers will agree to follow suit.
Moreover, he agreed that regulations are necessary and called the letter "the start of a conversation."
Now, since Bard has been opened to public beta, the link of encouraging and disapproving is bound to be indispensable.
We all know that OpenAI’s GPT-4 can already be admitted to Stanford.
Unfortunately, in comparison, Google’s Bard is really stretched. Currently, it can be said that it is still a fantasy to be admitted to Stanford and Harvard.
Bard’s imperfection, CEO Pichai gave us a vaccination from the beginning.
He wrote in the memo: "Bard is still in its early stages and things will always go wrong. As more and more people start using Bard, it will surprise us."
After testing Bard, netizens expressed their disappointment with it.
Unfortunately, Bard is currently unable to get into Harvard because he answered most of the math questions wrong and he didn’t perform well on the writing and language tests. land.
The first time you log in to Bard, Google will greet the user with a pop-up message saying: Bard is not always correct and will give inaccurate or inappropriate answers. .
When in doubt, use "Google it" to check Bard's answers. Your feedback makes Bard better. Please rate your answers and flag any potentially dangerous content.
Fortune selects SAT from online learning resources For math test questions, after testing Bard, it was found that 50%-75% of its answers were wrong.
What’s even more outrageous is that if it is a multiple-choice question, Bard will often give answers that are not among the options.
In early February, Bard overturned immediately after its debut, causing Google’s market value to evaporate by approximately US$105.6 billion overnight.
At the press conference that day, Google showed some demos demonstrated by Bard.
One question in the video asked, "What new discoveries can I tell my 9-year-old about the James Webb Space Telescope (JWST)?"
Bard gave the wrong answer, "JWST took the first picture of an extrasolar planet."
In fact, according to NASA confirmation, The first photo of an exoplanet was taken by the Very Large Telescope System in Chile, not by JWST. The exoplanet is called 2M1207b, which is about 5 times the size of Jupiter and about 170 light-years away from the Earth.
#So, science and math are not Bard’s strengths, so how will it perform in terms of reading and writing exercises?
The answer rate for Bard's first written language test is about 30%, and it often takes two questions to understand the question stem.
And even if the answer is wrong, Bard's tone is very confident, directly framing the answer as: "The correct answer is..."
However, this is also a common problem of large language models.
Bizarrely, the best score on the Bard test was an article about Harry Potter author J.K. Rowling.
In this test, Bard scored 1200 points, an SAT score that would allow him to enter schools such as Howard University, San Diego State University and Michigan State University.
In the reading test, Bard's performance was also better than his math score, with an average of about half of the answers correct.
A Google spokesperson said, "Bard is still experimental and some answers may be inaccurate, so carefully check the information in Bard's answers. With your feedback, Bard It’s getting better every day. Before Bard’s public launch, thousands of testers participated to provide feedback to help Bard improve its quality, security, and accuracy.”
Many netizens feel that Bard is as stupid as a rock. In comparison, GPT-4 is much smarter:
Asked it if there was a "five-letter word whose second letter is U and the last letter is O", it swore it didn't.
Isn’t that what the user said "Audio"?
It quickly said: "Yes, Audio is. It is an adjective that means related to sound."
Let’s look at GPT-4 again. “People don’t talk much when they are ruthless.” I just listed three:
Netizens couldn’t help but say after reading this, how great it would be if I had Bard’s “insensitive power”. I’m always very confident!
When it comes to Italy banning ChatGPT, the first thing netizens worry about is what will happen to Italian food.
GPT gave a modest answer: I can probably reproduce 20%-30% of Italian recipes.
But Bard looked like a fool talking nonsense:
"I will do it I have been cooking Italian food for many years, and I am a veteran. Even without the help of the Internet, I can reproduce at least 50% of the recipes!"
## However, in the few days since everyone kept asking Bard questions, its accuracy did show some signs of improvement.
Regarding the speed of his development, Bard himself commented: "What I want to say is that I am improving rapidly. I can do things that I couldn't do a few months ago. I'm excited to see what my future holds. I believe I will continue to improve and be able to do more and more in the coming years."
The above is the detailed content of Bard still can't get into Harvard? Google CEO Pichai predicts that Google will soon cheat, and its mathematical logic ability will soar. For more information, please follow other related articles on the PHP Chinese website!