Shulou(Shulou.com)11/24 Report--

Recently, Tao Zhixuan, a "mathematical genius", said that he would lead the White House task force on generative artificial intelligence to evaluate and collect opinions on the current AI. In Tao Zhe Xuan's view, the ChatGPT that joins the workflow does not add much value in the field of mathematics.

Recently, the Presidential Council of Science and Technology Advisers (PCAST) set up a working group on generative artificial intelligence.

It is worth mentioning that Tao Zhe Xuan, a mathematical genius, played the role of co-leader in this working group.

'i co-chair this generative artificial intelligence working group with Laura Greene, 'Mr. Tao said in a post on his blog.

According to his blog, the group's main graduate students are artificial intelligence technologies that have a broader impact on science and society, including popular text-based large language models (such as ChatGPT), diffusion models of image generation (such as DALL-E2, Midjourney), and scientific application models (such as protein design or weather forecasting).

Tao Zhe-Xuan hosted a generative AI group and the White House said in an article released on the 13th that

The generative AI team established by PCAST helps assess key AI opportunities and risks and provides advice on how best to ensure that these technologies are developed and deployed as fairly, responsibly and securely as possible.

At the end of the article, you can see that among the members of the working group, Tao Zhe Xuan is among them.

In addition, Lisa Su, CEO of AMD, is also a member of this generative AI team.

According to Tao Zhe Xuan's blog, the generative AI group will hold a public meeting during the PCAST meeting on Friday, May 19.

LVB link: https://www.whitehouse.gov/ pcast / meetings/ 2023-meetings/

The two expert groups will introduce the current situation of generative artificial intelligence and finally conduct a wide range of question and answer sessions. These speakers include:

Artificial intelligence promotes science:

Anima Anandkumar (California Institute of Technology & Nvidia)

Demis Hassabis (Google DeepMind)

Li Feifei (Stanford)

Artificial Intelligence and Society:

Sendhil Mullainathan (Chicago)

Daron Acemoglu (MIT)

Sarah Kreps (Cornell University)

In addition, Tao also mentioned that the working group is consulting the public on how to promote the active deployment of generative artificial intelligence applications and how best to reduce risk.

The initial focus is on how to find, resist and reduce false information and DeepFake generated by artificial intelligence without sacrificing freedom of speech, which is a challenging topic.

After ChatGPT joined the workflow, ChatGPT was born, and it has also been favored by mathematicians like Tao Zhe Xuan.

In the latest post on Mathstodon, I shared my views on generative AI tools.

I'm beginning to see the comparative advantage between myself and the current generative AI tool. For the tasks I perform every day, I have developed enough technology to optimize my workflow, and artificial intelligence tools do not add much value to me. The most obvious is in the study of mathematics, but it also includes writing e-mail. I installed a plug-in that lets GPT-4 write an email reply when I click a button, but I almost never use it because after decades of practice, I can quickly write the right email reply.

For tasks that I have some expertise but rarely practice, artificial intelligence tools are helpful: usually I can use them to create first drafts of output, which I can then verify and modify, or at least as a source of inspiration. In some cases, inspiration is due to the lack of AI, which is in line with the spirit of Cunningham's law, but it is still more effective than trying to solve the problem myself. Examples of this category include data processing, translation into a foreign language, or writing text in a format that I rarely use (such as public speeches, rule files, etc.).

For tasks that I don't have much expertise and don't need extremely high quality and reliable output, I can simply ask artificial intelligence tools and follow their advice more or less. Here, the function of artificial intelligence is a slightly more convenient version of traditional search engines.

Finally, for those tasks that I do not have professional knowledge but require quality and reliability, neither artificial intelligence nor myself can solve them. I must consult human experts. Such as repairing a complex, expensive, exquisite equipment.

In short, Tao Zhe-Xuan's views on generative AI tools fall into four categories, and it is worth noting that there is not much value-added in his field of mathematics.

For the third case, Tao asked ChatGPT to summarize the previous article as a flowchart.

ChatGPT provides a description of the text. Tao said it is speculated that future multimodal GPT will be able to provide flowcharts directly rather than a text description.

For an example of the second type of task: after realizing that I could ask GPT to output the flowchart in LaTeX format, I got the first picture below, which is obviously not perfect. But since I am familiar with LaTeX, it is not difficult to manually correct it to the second picture.

Obviously, ChatPDF, Tao, which became popular some time ago, is also used in workflow.

In March, Tao said she decided to try to incorporate AI tools into my workflow in different ways. Including ChatGPT, DeepL and other tools.

Over the next period of time, he often shared some of his experiences with tools such as ChatGPT.

Many ChatGPT hiding functions have been discovered by him, such as finding formulas, parsing documents in code format, rewriting thesis sentences, and so on.

For example, ChatGPT can sometimes do a semi-finished semantic search in mathematics, that is, it can be used to generate hints.

For example, Tao Zhe-Xuan asked ChatGPT to recognize Kumer's theorem from the description, which did not give the correct answer, but according to its approximate answer (Legendre formula).

In this regard, Tao Zhe Xuan said that the role of artificial intelligence in mathematics is to provide a preliminary approximate answer, and then can be combined with traditional search engines to easily find the correct answer.

Tao Zhe Xuan also discovered the highlight of ChatGPT in dealing with mathematical problems, and was able to recognize transliterated versions of mathematical concepts in different languages.

In addition, you can ask ChatGPT to transform a bunch of references obtained from MathSciNet and format them as\ bibitems in the LaTeX bibliographic environment.

But doesn't ChatGPT ever make a mistake?

In a proof question about "whether there are infinitely many primes", Tao Zhe-Xuan found that the answer given by ChatGPT is not entirely correct.

On the other hand, he found that the idea of argument given by ChatGPT could be fixed, and he had never seen it before.

Do AI tools such as ChatGPT join your workflow?

Reference:

Https://terrytao.wordpress.com/2023/05/13/pcast-working-group-on-generative-ai-invites-public-input/

Https://www.whitehouse.gov/pcast/briefing-room/2023/05/13/pcast-working-group-on-generative-ai-invites-public-input/

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.