HomeAI News
Google's magic tool for fishing is here: launch AI conference stand-in, one-click summary of questions and speeches

Google's magic tool for fishing is here: launch AI conference stand-in, one-click summary of questions and speeches

Hayo News
Hayo News
August 31st, 2023
View OriginalTranslated by Google

The workers who have been tortured by the meeting can finally be liberated!

At the just-held Google Cloud Next'23 conference, Google announced a series of new technological advances, the most notable of which is the "meeting AI stand-in" .

Just say "help me join" and AI can hold the meeting for you and speak and ask questions at the right time:

In this way, even if you are trying to catch something during a meeting, you will not be afraid that your boss will suddenly call your name (manual dog head)

Some netizens were ecstatic. If I really had this AI digital double to help me hold meetings every day, my productivity would be maxed out!

Some netizens joked that this has changed from "AI will replace my job" to "AI will replace my boss's job."

Let’s take a look at what black technologies were launched at this Next conference.

AI writes a summary for your meeting speech

Workers have all experienced the pain of meetings to a greater or lesser extent.

Sometimes, while eating or even on the road, I am suddenly pulled into a meeting, but the camera and sound quality, or the environment may not be suitable for meetings.

Google believes that when encountering "inconvenience", it can use AI to help summarize the meeting content and ask questions on your behalf.

Even if you catch a fish, AI can quickly summarize for you what was said in the first half of the meeting , allowing you to quickly join the topic:

Not just a "meeting double".

In the middle of the meeting, AI can also automatically take on the task of taking shorthand——

Help take notes, even record video clips, and organize them into shorthand notes of the day's meeting after the meeting, and send them to each participant in real time.

Dave Citron, Senior Director of Google Conference Products Department, believes that video conferencing has three innovation stages. The first is when video conferencing appears, the second is the period of mixed office, and the third is the period of AI outbreak.

Now, it's time to add AI to your conferencing app.

For this feature, The Verge joked that if every participant sent their AI assistant, the topic of the meeting would probably be figured out quickly and end quickly.

Some netizens lamented that the scene where everyone uses machines to help them conduct meetings and classes may soon be realized:

If you encounter a cross-border meeting or need to quickly understand the content of a foreign meeting, AI also provides automatic translation of subtitles in 18 languages.

And if the video is not convenient where you are, you can also upload your face based on AI to the Google Meeting APP in advance, and use it directly during the meeting.

This function comes from Duet AI , Google's integrated AI assistant.

At this Next conference, in addition to this conference stand-in, Duet AI was also added to other functions, realizing a wave of "AI major changes" in APP.

Duet AI turns into an "enhanced version"

Duet AI was launched at the Google I/O conference in May this year, based on the adjustment of the PaLM 2 model.

At that time, Duet AI was mainly integrated in Gmail and Google Docs. For example, like Microsoft Office, it could intelligently write content, make PPT, analyze data in tables, and summarize email content, etc.:

Now, in addition to the meeting "stand-in" mentioned above, Duet AI has also implemented a series of new functions in Google Cloud -

code generator . It can not only help you automatically complete code, generate code, but also perform code refactoring.

infrastructure management . Users only need to chat with Duet AI in Google Cloud Console (a facility that monitors cloud resources), and they can quickly grasp the infrastructure configuration and get suggestions on how to improve utilization and performance.

Database Assistant . Just "chat" with Duet AI, it can automatically understand your needs, and help you build a database, and be responsible for modifying and querying data;

Security Analyst . Duet AI is also currently introduced in Google's security products, which can assist in viewing potential threat factors in computers and assist security personnel in their work more effectively.

In addition to this series of new cloud services, Google Chat, Google's office communication software, also uses Duet AI.

Now, on Google Chat, you can directly find and extract files sent by other people, and use Duet AI to summarize the content, etc.

Google also emphasized that even if they use Duet AI as an assistant, they will never use the content in the file to train the product, believe it or not

Also integrates Llama2 and Claude2

In addition to the enhanced version of Duet AI, there are some other AI highlights at the Next conference this time.

Just like many domestic manufacturers are doing, Google has also integrated Llama 2 and Claude 2 and other popular large-scale model APIs into its own cloud service, which is convenient for users to call directly.

Google also stated that they are currently the only cloud service vendors that provide Llama 2's pre-trained model fine-tuning (adapter tuning) and RLHF.

In terms of infrastructure cooperation, Nvidia has won again.

Although Google has also launched its own fifth-generation TPU this time, it promises to increase the training performance per dollar by 2 times and the inference performance by 2.5 times ( Google also passed it before, this version of the TPU uses AI design ).

However, at the conference, Google Cloud specifically emphasized that it will further cooperate with Nvidia and use more Nvidia cards to enhance its cloud products to accelerate the training of large language models.

Next month, Google Cloud will make Nvidia's H100 GPU available to developers as part of a virtual machine on its A3 supercomputer.

Finally, Google also released an "AI image identification watermark" function.

This is an AI technology introduced by Google DeepMind earlier, that is, to add a pixel-level watermark to the image generated by AI, which can be distinguished only by identifying it.

Which black technology do you think is the most practical?

Reference link:

[1] https://www.theverge.com/2023/8/29/23849056/google-meet-ai-duet-attend-for-me

[2] https://cloud.google.com/blog/topics/google-cloud-next/welcome-to-google-cloud-next-23

[3] https://twitter.com/swiftonsecurity/status/1696541618910941530

[4] https://techcrunch.com/2023/08/29/google-cloud-announces-the-5th-generation-of-its-custom-tpus/

Reprinted from 量子位 萧箫View Original


no dataCoffee time! Feel free to comment