Microsoft releases a 166-page GPT-4V evaluation report, where the advanced usage of ChatGPT is found | Hunt Good Weekly Report

Tune in to the latest issue of Hunt Good Weekly!

In this issue you will see:

6 fresh news
3 useful tools
3 distinct points of view
1 interesting case

Hunt for News|Advanced Headlines

 AI giants such as Microsoft and OpenAI are exploring self-developed AI chips

According to Reuters, OpenAI is exploring manufacturing its own AI chips and has evaluated potential acquisition targets.

OpenAI has discussed various options to address its AI chip shortage since at least last year, people familiar with the matter said. These plans include self-developed AI chips, closer cooperation with chip manufacturers such as NVIDIA, and diversification of suppliers.

The company has not yet decided to move forward, the report said. OpenAI CEO Sam Altman has previously stated that acquiring more AI chips has been the company's top priority.

He has publicly complained about the scarcity of GPUs, a market dominated by Nvidia. In Altman's view, OpenAI faces many problems such as a shortage of advanced processors and high hardware operating costs.

Industry veterans said that OpenAI's self-developed chips will be a major strategic move and a huge investment, with annual costs that may be as high as hundreds of millions of dollars. Even if OpenAI devotes resources to the task, success is not completely guaranteed.

Compared with OpenAI's hesitation in developing its own AI chips, Microsoft is ahead of the curve and plans to launch the company's first chip specifically designed for AI at its annual developer conference next month.
Microsoft's chip is similar to Nvidia's GPU and is designed for data center servers that train and run large language models.
People familiar with the matter said that some employees at Microsoft and OpenAI have been testing Microsoft AI server chips. By developing self-developed AI chips, Microsoft will basically catch up with Amazon and Google.
Microsoft hopes the chip will be comparable to Nvidia's H100 GPU, but more performance will have to wait until its actual debut next month.

 Microsoft releases 166-page GPT-4V evaluation report, a must-read for advanced users

In a previous article, we reported that ChatGPT ushered in an epic update, which jumped from simple text "coming and going" to image "coming and going", realizing image-based analysis and dialogue. You can simply understand it as a change from GPT- 4 converted to GPT-4V(ision).

While people were still immersed in studying the gameplay of GPT-4V, OpenAI's "big owner" Microsoft directly released a 166-page report describing in detail the testing and use of this function.

Tanishq Mathew Abraham, co-founder and CEO of MedARC (Medical Artificial Intelligence Research Center), said, "This report will be a must-read for GPT-4V advanced users."

Specifically, this report is divided into 11 chapters, including paper overview, GPT-4V input mode, GPT-4V working mode and prompt technology, visual-language ability, interaction with humans: visual reference prompts, time And research on video understanding, visual reasoning and IQ tests, emotional intelligence tests, emerging application highlights, LLM-based agents, etc.

The test method is also very simple. By inputting Prompt (prompt word), the author of the report can ask questions to GPT-4V.

For example, the report author input a photo of the United Nations emblem and asked GPT-4V to write svg code to generate a photo of the United Nations emblem. Unfortunately, the final generated effect was not very good, or even a little bad.

Of course, the above examples are just a few examples. In fact, throughout the entire report, GPT-4V has performed well in logical reasoning capabilities, image recognition capabilities, and multi-modal understanding capabilities.

For example, when the report author "fed" GPT-4V a photo of Jen-Hsun Huang, GPT-4V was still able to accurately identify the identity of NVIDIA CEO and co-founder and the specific behaviors in the photo.

Due to space limitations, I will not give examples one by one. The link to the report is as follows. Welcome to read it.

 Meta pays top creators millions to build AI characters using their likenesses

According to The Information, Meta paid stars such as Snoop Dogg, Tom Brady, MrBeast and Charli D'Amelio to use their likenesses to create virtual AI characters.

For several top creators, compensation could run into the millions each over the years, according to people familiar with the matter.

Currently, these roles are mainly text-based chatbots on Instagram, WhatsApp and Messenger, but Zuckerberg said voice capabilities will also be launched early next year.

By partnering with high-profile celebrities, Meta may be looking to tap into the large audiences these stars have built up across different platforms.

Creators who have become famous on platforms like YouTube and TikTok have proven that they can get fans to watch content, buy products and use different services, and their AI avatars may have the same impact.

 Is Activision Blizzard’s acquisition coming to an end?

Microsoft plans to complete its $68.7 billion acquisition of Activision Blizzard next week, The Verge reports.

The company expects Friday, October 13, to be the deadline when Microsoft will announce to the world that the 20-month process to acquire Activision Blizzard has ended, sources said.

However, whether Microsoft can announce this news as expected by then will also depend on the British antitrust regulator "Competition and Markets Authority" (CMA). Microsoft and Activision Blizzard recently extended the transaction deadline to October 18 to await the outcome of negotiations with British regulators.

Microsoft received preliminary approval from the CMA last month after Activision Blizzard agreed to sell its streaming rights to Ubisoft Entertainment. If Microsoft succeeds in acquiring Activision Blizzard, it would mark the biggest deal in the gaming industry.

 The Forbes 400 in the United States announced that Musk ranked first

Recently, "Forbes" in the United States released the "Forbes 400" ranking of the super rich in the United States.

Overall, the wealth of the top 400 richest people in the United States has reached a total of US$4.5 trillion (approximately RMB 32.38 trillion), tying the wealth record set in 2021.

According to Forbes, thanks to the AI-driven and technology boom and the rebound in the stock market, the wealth of these rich people increased by about $500 billion compared with the previous year.

The wealth ranking of the top 8 rich people is as follows

Interestingly, the entry barrier for this "Forbes 400" is not low. The net worth must reach at least US$2.9 billion, which was only US$2.7 billion the previous year.

According to Forbes, legendary American star Michael Jordan became the first professional player to appear on the list thanks to his deal with Nike and the sale of his stake in the Charlotte Hornets. athlete.

Other billionaires who are too poor to make the list include Apple CEO Cook ($1.8 billion), Facebook COO Sandberg ($1.8 billion) and former U.S. President Trump ($2.6 billion).

 The 2023 Nobel Prizes are being announced

During the National Day holiday, in addition to the Hangzhou Asian Games that dominated the screen, the "who will win the Nobel Prize" in 2023 is also attracting attention.

The awards that have been announced so far are:

The Nobel Prize in Physics is awarded to Pierre Agostini, Ferenc Krausz and Anne L'Huillier "for their research For his contributions to the experimental methods for generating attosecond lasers due to the dynamics of electrons in matter.

The Nobel Prize in Physiology or Medicine was awarded to Hungarian scientist Katalin Karikó and American scientist Drew Weissman for their discoveries in nucleoside base modification.

The Nobel Prize in Literature is awarded to Norwegian novelist and playwright Jon Fosse for his innovative plays and essays that give voice to the ineffable.

In addition, a number of awards such as the Nobel Peace Prize are also being announced. It is worth noting that this year, the prize money for each Nobel Prize will be increased by 1 million Swedish kronor (approximately RMB 650,000), reaching 11 million Swedish kronor (approximately RMB 7.15 million), which is the highest level in history.

Hunt for Tools|Advanced Tools

 Assistant with Bard: Your AI Assistant

Recently, at the Made by Google 2023 conference, Google launched Assistant with Bard.

Assistant with Bard is a personal assistant powered by generative AI that combines the generative reasoning capabilities of the Bard chatbot with the personalized assistance capabilities of the Assistant.

The tool will integrate with Google apps like Gmail and Docs, so you can easily talk to a Google Assitant and have it help you do things like create captions for photos, create shopping lists, and find information in your email inbox. .

Google promises that Assistant with Bard will be gradually pushed to iOS and Android in the next few months. For more specific information, you can refer back to APPSO's previous article.

The Pixel 8 series is officially released! This is Google’s AI killer feature

 It only takes 26 seconds for AI to design a robot

Recently, a team led by researchers at Northwestern University developed the first AI that can design robots on its own.

It took nature billions of years to evolve the first walking species, but this AI was able to design a successful walking robot in just 26 seconds.

"We have discovered a very fast AI-driven design algorithm that bypasses evolutionary detours and does not rely on human bias," said Northwestern University researcher Sam Kriegman.

The researchers gave the AI ​​a basic prompt, "Design a robot that can walk on a flat surface." In the blink of an eye, the AI ​​could build a basic blueprint.

The AI ​​starts with a lump of material about the size of soap, which at first only shakes and is completely unable to walk. The AI ​​then evaluates the effectiveness of its design and rapidly iterates on the design.

Finally, the robot can bounce in place, then jump and move forward. After nine attempts, the AI ​​designed a robot that could walk every second, at about half the pace of a human.

 Amazon plans to launch AI chatbot

Amazon is revamping the search experience on its website and apps to incorporate generative AI capabilities that are more conversational, deeper and more personalized, BusinessInsider reported.

The project, codenamed "Project Nile," aims to add artificial intelligence capabilities to Amazon's existing search bar to enable instant product comparisons, obtain more product details and reviews, and provide recommendations based on user search content and personal shopping data.

People familiar with the matter said that the new search was originally scheduled to be launched in September this year, but has now been delayed to around January next year and will be launched first in the US market.

AI-powered interactive search could improve the Amazon shopping experience. Additionally, Amazon hopes the changes will increase sales, especially on mobile devices.

Hunt for insights|Prophet

 SoftBank CEO Masayoshi Son: General artificial intelligence will surpass human intelligence within ten years

SoftBank CEO Masayoshi Son recently spoke publicly at the SoftBank World 2023 Corporate Conference in Tokyo.

At the meeting, Masayoshi Son said: "General artificial intelligence will exceed human total intelligence 10 times within 10 years."

▲ Picture from: Reuters

He believes that general artificial intelligence will affect various industries from transportation and pharmaceuticals to finance, manufacturing, logistics, etc., and companies and individuals working on artificial intelligence will become leaders in the next 10 to 20 years.

"It would be sad to stand on the side that is left behind, like an old, closed commercial street." Therefore, Masayoshi Son called on Japanese companies to "wake up" in time to the vision of artificial intelligence leading the future. .

But he also noted the need to address concerns about disinformation and privacy and copyright risks, and not be hesitant to worry about data breaches and other possible risks.

 JPMorgan Chase CEO Jamie Dimon: AI is expected to eliminate the 5-day work week

JPMorgan Chase CEO Jamie Dimon recently accepted an interview with Bloomberg TV. In the interview, he said JPMorgan Chase & Co. is embracing artificial intelligence and that the technology has the ability to significantly improve the quality of life for its employees.

▲ Picture from: Getty Images

"Thanks to advances in technology, your kids will live to be 100 and not get cancer," Jamie Dimon told reporters. "They could literally work three and a half days a week."

According to Bloomberg, data from consulting firm Evident showed that JPMorgan Chase hired more than 3,500 artificial intelligence-related positions between February and April.

Jamie Dimon said artificial intelligence is "real" and "critical" to JPMorgan Chase's future success, helping improve productivity, product design, customer engagement and risk management.

 Artificial intelligence could deliver $10 trillion in productivity gains

IBM CEO and Chairman Arvind Krishna recently said at the World Government Summit in Dubai: "In the next 10 years, artificial intelligence will bring a contribution of US$10 trillion to the global economy."

He believes that artificial intelligence has the potential to change the world economy due to increased per capita productivity.

"If you think about GDP growth, that could be between 10%, 20% or 30%. But we have to be careful, we have to leverage skills and deploy it in the right way."

In a conversation with UAE official Omar bin Sultan Al Olama, Arvind Krishna said that increased adoption of AI in the UAE alone could lead to productivity gains of up to $200 billion by 2030.

Hunt for Fun|First

 Comic picture translation artifact

If you are a senior comic fan, you may be annoyed because you cannot read foreign language comics, and it seems not cost-effective to learn a language temporarily. The comic AI translation artifact that I will introduce next may help you solve this problem.

Manga-Image-Translator is an open source comic image translator that can automatically translate text in comic images. Of course, ordinary images can also be used. After the translation is completed, the translation will be replaced in the original position of the image.

Currently, this tool mainly supports Japanese, Chinese, English and Chinese. The translation speed is quite fast and the effect is not bad.

Tool trial link:

Easter egg time

To celebrate its 50th anniversary, 10 iconic architects used AI to redesign the Sydney Opera House. Which version do you prefer?

# Welcome to follow the official WeChat public account of aifaner: aifaner (WeChat ID: ifanr). More exciting content will be provided to you as soon as possible.

Ai Faner | Original link · View comments · Sina Weibo