Google launches PaLM2 to overthrow ChatGPT!
- I. Introduction
- 2. PaLM2: Google launches PaLM2 to overthrow ChatGPT!
- 3. Bard upgrade: multi-modal function, support image generation and recognition
- 4. Duet AI: Hand over office affairs to AI
- 5. Almighty and comprehensive, the redefinition of Google search
- 6. The anxiety of Google AI: it can only be solved by AI
- 7. Summary
## Benefits: At the end of the article, there is a free Chinese version of chat GPT
1. Foreword
There is a surprising tidbit at the beginning of the article: the Google I/O developer conference has ushered in its 15th year this year.
This high-profile event has always been an important place to attract developers and technology enthusiasts from all over the world. Every year, many new products and technologies are expected to be released. This year, people are even more looking forward to what kind of surprises and innovations Google I/O will bring, pushing the technology industry to new heights.
For nearly seven years, Google’s I/O conference has been targeting the field of AI. It’s a pity that before ChatGPT became popular, Google had previously demonstrated the LaMDA series of language generation models at the I/O conference, but it was a pity that it did not cause heated discussions at the time.
The explosion of AI technology ushered in a prosperous age for ChatGPT, but it also made Google feel anxious.
At the conference, Google hopes to prove that it is still at the forefront of AI technology by applying AI to almost all products, including Gmail, search and Pixel.
However, if this conference had been held half a year ago, it would have been the main event: important technical applications such as AI assistants, AI search and AI programming.
Unfortunately, before that, Microsoft was already one step ahead. The response from the live audience was not very enthusiastic, only Bard’s dark mode and the Pixel tablet giveaway could cause enthusiastic cheers.
At today’s I/O conference, Google’s new products and technologies are catching up with Microsoft and OpenAI, catching up with the pioneer who has already realized the future of AI.
2. PaLM2: Google launches PaLM2 to overthrow ChatGPT!
Large-scale language models have become one of the top products in the technology world. Many companies hope to develop independent models comparable to GPT to prove their technological strength, and Google is no exception.
Now, Google has released a new generation of large-scale language model PaLM2 to fight against GPT-4. According to Google, they used a large number of mathematical and scientific data sets to train PaLM 2. Compared with last year’s PaLM model, PaLM 2 has greatly improved its multilingual processing, reasoning and coding capabilities.
According to Google’s official test, PaLM2 even surpasses GPT-4 in some aspects (such as mathematics).
Google did not disclose the specific parameter size of PaLM2, only saying that it was built using the latest JAX and TPU v4 computing tools. Google has optimized the algorithm of PaLM2, which makes it smaller, with better performance and faster calculation speed.
PaLM 2 is a super language master, supporting more than 100 languages and more than 20 programming languages, including JavaScript, Python, Prolog, Verilog and Fortran, etc.
It can not only do text understanding and generation, but also supports multi-modality, and can understand and generate audio and video content.
Google showed a fine-tuned model in the professional medical field called Med-PaLM, which can analyze patient injuries based on X-rays, and it has reached the level of human doctors in this regard.
The PaLM 2 model can also be decomposed into four versions of different sizes, which can be deployed on various devices. For example, a smaller-scale Gecko model can be run on a smartphone, so that the mobile terminal can also have a large language model.
PaLM 2 is currently open for preview, and Google’s own chat robot Bard has used the PaLM 2 model to enhance its answering ability. Google content has more than 70 product teams using PaLM 2 to build products. This update is also designed based on the PaLM 2 model. PaLM 2 can be said to be the core engine of Google’s comprehensive AI.
In addition to revealing PaLM 2, Google CEO Sundar Pichai also briefly introduced the situation of the next-generation large-scale model, Gemini. He said that Gemini is a model that is multi-modal, efficient and can be built for future innovation.
3. Bard upgrade: multi-modal function, supports image generation and recognition
It looks better, is more comprehensive, and can be used by more people.
Google’s Bard will be upgraded to the more powerful large language model PalM2 after an update of less than two months, this competitor of ChatGPT and BingChat will be greatly updated and more powerful:
1. Cancel the waiting list and open to more than 180 countries and regions;
2. Added picture question and answer, the dialogue is more intuitive;
3. Integrate Adobe Firefly to quickly generate pictures;
4. Source reference, dark mode, code export button, more details update.
It is not easy to use Bard, which is limited to the United Kingdom and the United States, only supports English, and has many restrictions. However, from now on, Bard has officially announced that it will cancel the waiting list and open the experience to more than 180 countries and regions.
In addition to English, Bard now supports Japanese and Korean, and will soon expand to support 40 languages, including Chinese.
Bard has also upgraded the picture function, which can be summed up in four words as “both pictures and texts”. When you ask a question, it will not only answer the text dryly, but also show richer and intuitive answers with pictures.
In addition, Google has also integrated the function of Google Lens into Bard, which allows Bard not only to answer pictures, but also to write content based on the information of pictures.
For example, you can send it a photo of a dog and have it write some captions based on the dog’s breed.
Perhaps because of the well-received feature of Vincent next door, Google has added a similar feature to Bard. However, this feature is a collaboration with Adobe Firefly to integrate Firefly’s image generation capabilities into Bard to generate high-quality images in a natural language manner.
In addition, with Adobe’s support, users don’t have to worry about the copyright and quality of the pictures. In addition to working with Adobe, Google will work with Kayak, OpenTable, a zip recruiting company, Khan Academy, and several other partners to bring more features to Bard.
Compared with ChatGPT, Bard’s positioning is more biased towards BingChat. It is a search engine and a productivity tool. It will add a reference function next week and display the source for the generated code to ensure accuracy. And the ability to export and run code has been added to facilitate developers to work efficiently. By the way, a dark mode was added, which aroused cheers from the scene.
Although there were some oolongs in his debut, Google still let us see a Bard that is constantly improving. It’s a pity that “we’ll” is the word we heard the most in the introduction. After all the functions released today are implemented and launched, Bard will be a sufficiently comprehensive and mature tool.
4. Duet AI: Hand over office affairs to AI
Here comes the Copilot competitor.
Google has introduced a new tool, Duet AI, for its own Workspace, allowing us to see the role AI plays in the cloud-based Google office suite:
1. Documents: summarize the full text and automatically write;
2. Slideshow: Vincent diagram, improve efficiency;
3. Forms: Analyze data and formulate strategies.
In Google Docs, you can tell Duet AI what you want to create and have it automatically generate articles. You can also let it generate different styles of writing according to your needs.
The “ChatPDF” function of “reading” articles and key questions has not been left behind.
In terms of slideshows, Duet AI acts as a picture generator, which can help you save time looking for pictures on the Internet and enrich slideshows with original content.
In the form, Duet AI can automatically organize the data according to your questions to make the data more standardized.
These capabilities can also be linked with other Google services. For example, after you create a form in Bard, you can generate a form document with one click for subsequent use.
Duet AI has also added a set of machine learning-driven Google cloud development tools, which can provide real-time code error correction, generation, etc., and support multiple development languages such as Go, Java, Javascript, Python, and SQL.
We don’t have to wait for Bard to pass, now we can wait for Duet AI. If you want to experience these features, you still need to join the waiting list and wait for Google to open the test in the corresponding region.
Generally speaking, Duet AI is less amazing than seeing Copilot for the first time, and the function demonstration seems to be weaker. But it took less than two months for Bard to catch up, so expect more possibilities from Duet AI.
5. Almighty and comprehensive, redefining Google search
Future searches may look like this:
Driven by PaLM 2, Google search has also ushered in an innovation, and a new search engine SGE (Search Generative Experience) has been launched. You no longer need to divide a big question into multiple small questions, and then put together the answers after searching , with the blessing of generative AI, you can let search do some heavy work.
At the conference, Google highlighted the improvement of generative AI in shopping. When you use Google to search for a product, you will see its product description, reviews, prices and display pictures.
Google, which has more than 35 billion product listings, is constantly updating at a rate of 1.8 billion per hour. By searching and shopping on Google, you can get fresher and more reliable results.
picture
Google believes that advertising is an important part of the Internet, and they will continue to uphold their commitment to advertising transparency and to ensure that ads and search results are distinguished.
Google is opening up sign-ups for Search Labs today and opening access in the coming weeks, with support for the Google app and the Chrome browser.
The application link is as follows: labs.google.com/search
6. The anxiety of Google AI: it can only be solved by AI
Some people jokingly say that this year’s Google I/O should be called Google AI/O. Microsoft and OpenAI’s gems are in front, and Google has lost a little amazing moment, but this is undoubtedly Google’s most resolute self-revolution in recent years.
Google, which claims to be AI-first, has been missing opportunities many times because of its overly cautious and conservative strategy. Now Google has finally realized that it must act quickly, as many Google executives have said, to be “bold and responsible”, and can no longer hesitate because of fear of making mistakes.
Of course, AI is not a panacea, and the upgraded Bard may not be able to change the trend of Google search being subverted. But there is no doubt that AI will reshape search and revolutionize this interaction.
Bing, whose share is far inferior to that of Google, has reached more than 100 million daily active users in just a few months through chatbots, and conducted more than 500 million chats, of which 70% of users who try the new Bing chat function will use it for search related tasks.
7. Summary
When the world’s largest search engine decides to abandon the burden of the past and fully embrace this trend, what will be changed may not only be Google’s future, but it may also subvert the operating mechanism of the Internet and change the interaction of billions of users.
A new, simpler world is coming.
Charger will bring you the latest and most comprehensive interpretation as soon as possible, don’t forget the triple wave.
Pay attention to the WeChat public account: Recharge your resources
Re: Chat GPT
(ps: I won’t tell the average person, the official account has all kinds of resources on the whole network, only you can’t think of it, there is no information you can’t find if you don’t charge it)
Charging Jun sent you: enjoy the Chinese version for free