Popular news sites




查看: 4|回复: 67999

[转帖] In less than 4 minutes, this GPT -driven "chemist" manufacturing drugs also reproduced the Nobel Prize research!Study Deng Nature

发表于2024-03-03 20:09:35 | 只看该作者 回帖奖励 |倒序浏览 |阅读模式

    Original Digest of Style Big Data Digest

    Big data strip

    Author: caleb

    Have you ever thought that in a few minutes when you read this breaking news websitearticle, you can be re -engraved perfectly by getting the Nobel Prize for the Nobel Prize.

    Yes, you did not hear it wrong. Now the AI -driven system has been able to understand these chemical reactions, and even designed laboratory procedures to make them.

    Artificial intelligence completed all of this in just a few minutes, and successfully tried it successfully."This is the first time the non -organic intelligent plan, design and execution of human invention." Gabe Gomes, a chemist and chemical engineer of Carnegie Mellon University, said that he led the assembly and testing research team based on artificial intelligence -based systems.Essence

    They named the AI coscientist.

    The complex response implemented by COSCientist is called a catalytic cross-coupling couplet in organic chemistry. The reaction is American chemist Richard Fred Heck and two Japanese chemists Ei-IChi Negishi and Akira Suzuki won the 2010 Nobel Prize Chemical AwardIt is used to recognize the huge role of these reactions in the drug development process and other vaccium carbon -based molecules.

    The capabilities shown by COSCientist show that human potential has the potential to effectively use artificial intelligence to improve the computing speed and quantitative scientific discovery, and improve the repetitiveness and reliability of the experimental results.

    The paper has been published on Nature.

    Thesis link:

    Demystify coscientist

    The most important thing in the software and silicon -based components of COSCientist is the large language model that constitutes its artificial "brain".For this reason, the team tested and compared multiple large language models, including GPT-4 and other versions of GPT.

    COSCientist is equipped with several different software modules. The team first tests these modules separately and then conducts collaborative tests.

    "We try to divide all possible tasks in science and build a larger picture one by one," said Daniil Boiko, a doctoral student who participated in the study, said that he designed the overall architecture and experimental tasks of Coscientist.After integrating together. "

    With the help of the software module, COSCIENTIST can complete all the things that research chemists will do: search for public information about compounds, find and read how to control the robot laboratory equipment, write computer code for experiments, and analyze the results of the results to determine which ones are determinedEffective and invalid.

    The team also tested the ability of COSCientist to accurately plan chemical programs.If this program is executed, commonly used substances such as aspirin, acetaminol, and ibuprofen.Large language models have been tested and compared separately, and then checked the final program and scored based on whether they will generate the material, the details of the steps, and other factors.

    The highest score is obtained by the GPT-4 module supporting search. The module is the only module that creates the acceptable ibuprofen synthetic program.

    Boiko and colleagues Robert Macknight observed the process of COSCientist demonstration of "chemical reasoning".Boiko describes it as the ability to guide his actions with chemistry -related information and previous knowledge.It uses an open chemical information encoded by simplifying the molecular input cable input system (SMILES) format (a machine that represents a machine that represents the molecular chemical structure), and carefully checks the Smiles data based on the specific part of the molecule.

    "This is the best version of chemical reasoning," Boiko said.

    Further tests include the software module that allows COSCientist to search and use the technical document of an application programming interface that describes the application of the robot laboratory device.These tests are very important to determine whether the theoretical plan of COSCientist can convert its synthetic compound into a computer code of laboratory robots in the physical world.

    In less than four minutes, the program is designed and improved

    High -tech robotic chemical equipment is usually used in the laboratory, with accurate accuracy and overwhelming, spraying, heating, shaking small liquid samples and performing other operations over and over again.Such robots are usually controlled by computer code written by anthropological chemists.

    This is the first time that such robots are controlled by computer code written by artificial intelligence.

    The team starts the COSCientist with a simple task, which requires it to use the robotic liquid processor to distribute the non -color liquid to the board with 96 grids."," Draw a blue diagonal line "and so on.

    Then show scientists a plate containing three different colors (red, yellow, and blue) liquid, and asked them to determine which color existence and their positions on the plate.

    Because COSCientist has no eyes, the writing code automatically transmits the mysterious color plate to the spectrophotometer, and analyzes the wavelength of the light absorbed by each hole to identify which color existing and the location of the color board.For this task, researchers must slightly promote COSCientist in the correct direction, instructing it to think about how different colors absorb light.

    COSCientist's final exam is to put the modules and training assembled together to complete the "Suzuki and Sonogashira reactions".

    The reaction was found in the 1970s, and the keys between carbon atoms in organic molecules were catalyzed by metal cymbals.It turns out that these reactions are very useful for the production of new drugs for the treatment of inflammation, asthma and other diseases.They also use OLED organic semiconductors in many smartphones and monitors.In 2010, Sukuzi, Richard Heck and Ei-IChi Negishi jointly won the Nobel Prize, formally recognized these breakthrough reactions and its extensive impact.

    Of course, Coscientist has never tried these reactions before.Therefore, it has found relevant knowledge online, including Wikipedia, American Chemistry Society, Royal Chemistry Society, and other academic papers websites describing the reaction.

    In less than four minutes, COSCientist designed a precise program to use the reactions required to use the chemicals provided by the team.However, when it tries to perform a program in the real world with a robot, it makes an error in the code written to control heating and shaking liquid samples.

    However, even without human hints, Coscientist himself discovered the problem, checked the technical manual of the device, corrected the code and tried again.

    The result is contained in a small sample of some transparent liquid.Boiko analyzed the sample and discovered the spectral characteristics of the reaction.

    When Boiko and Macknight describe what Coscientist did, Gomes expressed incredible."I thought they were joking." "At that time, we suddenly realized that we had some very new and powerful things here."

    The greater the ability, the greater the responsibility

    With this potential power, it needs to be wise and prevent abuse.GOMES said that understanding the ability and limitations of artificial intelligence are the first steps to formulate wise rules and policies. These rules and policies can effectively prevent the harmful use of artificial intelligence, whether intentional or unintentional.

    "We need to be responsible for how to deploy these technologies and think deeply," he said.

    The scale and complexity of the natural world are actually infinite, and countless discovers are waiting for us to discover.Imagine a compound of new superconducting materials that can significantly improve energy efficiency, or cure other diseases that cannot be treated and extend the life of human life.However, obtaining the education and training required for these breakthroughs is a long and arduous journey.

    Gomes and teams imagine the gap between the artificial intelligence auxiliary system like COSCientist as a solution that can bridge the gap between the vast vast nature and trained scientists (and may always be the same).

    Human scientists also have human needs, such as sleeping and occasionally out of the laboratory.The artificial intelligence guided by humans can "think".On the night, I read every well -known stone in an orderly manner, check and check the replication of the experimental results.

    "We can have things that can run independently and try to discover new phenomena, new reactions, and new ideas," Gomes said.

    "You can also significantly reduce the entry threshold in any field," he said.For example, if the biologists who have not received relevant training want to explore their purpose in a new way, they can ask Coscientist to help them plan experiments.

    "This can realize the large -scale democratization of resources and understanding," he explained.

    GOMES said that there is a process of trying, failure, learning, and improvement in science. Artificial intelligence can greatly accelerate this process."This will be a huge change in itself."

    Related reports:


    Original title: "In less than 4 minutes, this GPT -driven" chemist "can make drugs and can reproduce the Nobel Prize research!Research on Nature "

您需要登录后才可以回帖 登录 | 注册


mother and baby| game| PUBG| football| Fashion| International football information| ( 浙ICP备2911366号-1 )

Powered by Popular news sites X3.4

© 2001-2017