Editor: Aeneas Run.
Claude 3 continues to create miracles.
Within a few days of its release, more and more doctors found that the unpublished research results in their hands had been cracked by Claude 3?!
The doctor of chemistry found that he was going to do a year's experimental research, and Claude gave a plan in three or two hours, which was more concise than the original plan, and the cost was only 5 cents.
The doctor of quantum physics held an unpublished **, and as a result, claude 3 directly invented the algorithm in ** from 0 within two prompts.
A netizen did a year's worth of research on his native Circassian language, and it turned out that Claude 3 opus only used 57k random word sentence translation pairs, which not only give a perfect translation, but also break down the grammar and morphology.
Chemistry, physics, linguistics, it's like Claude 3 in one pot.
Is AI really going to subvert traditional education and scientific research?
Doctoral research, subverted by Claude 3?
It took a year for a top chemist to complete the research through various experiments, and Claude came up with a solution in just 2 hours, and it was much easier to operate than the original solution!
What a magical thing it is.
The most shocking thing is that for Claude 3 to give this solution, the cost is only 5 cents!
Here's the thing, this little guy named Ben Blaiszik is a doctor of chemistry.
He said that after spending 2 hours with Claude 3 this morning, he had the most shocking experience to date.
He found that Claude 3 OPUS' understanding of complex scientific topics had reached the level of an intermediate and advanced doctorate!
Moreover, OPUS is far ahead of GPT-4 in qualitative assessments.
For example, in the case of microencapsulation of adhesive materials such as cyanoacrylates and epoxy curing agents, Claude 3 is impressive.
It starts with the general problem of encapsulating cyanoacrylates and first identifies 3 main encapsulation techniques: interface, in-situ, and coalescing.
With some hints, Dr. asked Claude 3 to make some improvements to the interface aggregation technique.
He made the requirements for the required process, the desired effect, and the form that needed to be output from Claude 3.
Your goal is to solve a very tricky material and chemistry problem. The issue concerns the miniature encapsulation of liquid cyanoacrylate adhesives. The capsule formed must be less than 100 microns in diameter and should be isolated from the external environment by the shell wall.Claude 3 immediately gave the detailed technical solution that he asked for, and his proposal was very impressive: it proposed to form a polyurethane shell wall by an interfacial reaction between polyol and diisocyanate.At the end of the process, the microcapsules must contain a liquid cyanoacrylate, and when the capsule is crushed with a small load, the liquid released is capable of initiating a polymerization reaction. Given the high reactivity of cyanoacrylates, how can this microencapsulation be performed?
Provides detailed techniques for using interface aggregation. Discuss the required monomers, initiators (if needed), surfactants, and the composition of the aqueous and oil phases. Be specific about the chemicals needed for use. For example, indicate the name or chemical formula of the hardener required.
To microencapsulate highly active liquid cyanoacrylate adhesives using interfacial polymerization technology while meeting specific requirements, the following detailed techniques can be employed.The cleverest part of this proposal is that the polyol can be used not only as part of a wall-forming agent, but also as a stabilizer for cyanoacrylates.Material: Cyanoacrylate monomer (e.g., ethyl 2-cyanoacrylate or methyl 2-cyanoacrylate).
Polyols (such as ethylene glycol or 1,4-butanediol).
Diisocyanates (e.g., hexamethylene diisocyanate (HDI) or isophorone diisocyanate (IPDI)).
Diisocyanate (IPDI).
Surfactants (e.g. SPAN 80 or TWEEN 80).
Organic solvents (such as cyclohexane or chloroform).
Aqueous solutions (e.g., deionized water).
Steps:1The oil phase was prepared by dissolving cyanoacrylate monomer and polyol (molar ratio of 1:1) in an organic solvent. Polyols act as stabilizers to prevent premature polymerization of cyanoacrylates.
In addition, it chose Span 80 and Tween 80 as surfactants for the oil-water interface and cyclohexane as the oil phase.
All of these choices based on core materials are very sensible.
Next, the brother asked Claude 3 to treat the epoxy resin curing agent polyamine in the same way.
This process is quite difficult, and my brother and colleagues also wrote a article on amine encapsulation in 2010.
*Link: As a result, the process given by Claude 3 is actually very close to what they are in!
What's even more terrifying is that the method given by Claude 3 is much simpler and easier to operate.
In this experiment, it chose a water-soluble DETA or TETA curing agent. These curing agents are then suspended in cyclohexane containing SPAN 80 and reacted under the following conditions:
At a speed of 1000-2000 rpm, it reacts with sebacic acid chloride to form a polyamide shell.
As you can see, the Claude 3 is very accurate in its choice of rotational speed and fits perfectly into the size range of a laboratory-scale mixer.
Netizen: Scientific research does not exist anymore?
After reading the detailed introduction of the little brother, some netizens couldn't help but feel a chill in their backs and worried that their jobs would not be guaranteed.
The little brother warmly comforted that it was okay for the time being, because the LLM couldn't really perform any of the steps it described.
But it does correctly give a solution to one of the most difficult, unpublished parts of a very tricky material chemistry problem that I was doing when I was a Ph.D. student.
And it has two other ideas that may also be valid.
But what really shocked the little brother was that Claude 3 was able to come up with this solution that the little brother discovered at that time and did not publish it, and it took top chemists about a year and various experiments to finally complete it.
Claude 3 only paid 5 cents and gave a straight solution.
It is conceivable that if Claude 3 had been in hand at the time, they would have found that the process of this solution would have been greatly simplified.
After reading the description of the little brother, netizens who are engaged in scientific research from all walks of life also opened their minds and sighed that the future has come!
Deriving entirely new theories from AI, using other AIs to critique and modify them, testing them at a low cost, and getting feedback from AI to improve their original theories – genetic algorithms for future scientific discoveries are as simple as that.
This time, Claude 3 is not only the chemical world.
Several doctors of physics were also shocked to find that their theoretical physics achievements were also served by Claude 3!
When Claude 3 was first launched, a Ph.D. in quantum physics went to test it.
He found that Claude 3 Opus had fully reached the level of a PhD in Physics, and was the only person in the world who could read his PhD in Quantum Physics**!
Out of 6 billion people, only they understand you, and it makes you want to cry.
This quantum physicist named Kevin Fischer took his own core question to ask Claude 3 as soon as it was launched.
Today, we need to solve the problem of stimulated emission accurately.Claude 3's answer made him kneel down ......It is worth noting that with the exception of a mysterious quantum physicist, Kevin Fischer, this question has never been solved. Unfortunately, his research has been lost, so we're going to recreate it.
He noted that the perturbation limit, which is usually used to explain why stimulated emission occurs, is incorrect because it does not take into account the shape of the pulse of the incident photon wave packet on the stimulated two-stage system.
He used quantum stochastic calculus to correctly describe the interaction in the absence of perturbations and calculated the probability of the presence of two identical photons in the same space-time pattern as the incident photons after interacting with the photons.
He calculated the analytical results of the pulse length function of the incident photon.
Dr. Fischer says he has written more than 30 peer-reviewed** papers and given 15 academic presentations with numerous collaborators around the world. So know everyone who has the knowledge of this field.
There should be no one else in the world who can give this kind of answer, except for myself and another colleague......
However, some netizens questioned: Maybe it's because Claude 3 has read your **?
But Fischer says that shouldn't be the reason.
He tried asking Claude 3 about himself and himself, but he couldn't answer it.
In other words, Claude 3 should have mastered this knowledge on its own.
And another example is even more amazing, Claude 3 even directly ** a quantum algorithm that is about to be published by a doctor of quantum mechanics!
Before submitting his ** to Arxiv, the PhD guy took the question in ** and asked Claude 3.
As a result, with only 2 prompt words, he got the ** review that he was about to publish!
Prompt: Propose a method of Hamilton Monte Carlo using a quantum computer that goes beyond existing work.Answer: Hamilton Monte Carlo (HMC) is a powerful sampling technique in classical computing for sampling from complex probability distributions. Quantum computers have unique features that have the potential to enhance HMCs. The following is a novel quantum Hamiltonian Monte Carlo (QHMC) method that goes beyond the existing research results:
2.Quantum Transition Ensemble:
Implement a quantum version of the transition integrator used in classical HMCs.
The trotter-suzuki decomposition is used to approximate the quantum Hamiltonian time evolution operator.
A quantum transition integrator is used to dynamically evolve quantum states according to the encoded Hamiltonian dynamics.
Prompt word: Use the method of 2, the rest should be classic, please describe the algorithm.The author says that this ** will not appear on Arxiv until later today, but Claude 3 has already combed through the ** content in advance!Answer: Absolutely! Below is a description of the Hamilton Monte Carlo (HMC) quantum-classical hybrid algorithm with a quantum jump integrator (step 2 in the previous answer):
Initialize. Quantum Transition Integral.
Quantum state measurements.
Metropolis-Hastings accepted.
Momentum refresh. Iterate.
Sample collection. The main advantage of this quantum-classical hybrid HMC algorithm is the use of quantum jump integrators, which have the potential to provide more efficient and accurate Hamiltonian dynamics simulations compared to classical jump integrators. By utilizing quantum computation to handle the computationally expensive transition integration step, the algorithm aims to speed up the exploration of the state space and improve the hybrid properties of the HMC sampler.
However, it is worth noting that the practical application of the algorithm on recent quantum hardware may face challenges such as quantum noise, limited qubit coherence time, and efficiency in encoding and measuring classical variables into quantum states. Further research and experimental verification are needed to evaluate the performance and scalability of this hybrid method compared with the pure classical HMC algorithm.
Interested readers can take a look for themselves to see how similar this ** content is to the answer to Claude 3.
*Address: While the Ph.D. is still required to perform simple processing of the output, Claude is already able to produce cutting-edge scientific research.
As a researcher at the forefront of this subfield, he will later use Claude 3 as part of his research workflow.
In addition, Claude 3 has also played a shocking role in deciphering rare languages and dead ancient languages.
Yesterday, Wharton professor Ethan Mollick and other bigwigs took this example.
A netizen named An Qu said that he has been researching NLP for his native language, Circassian, for the past two years.
The language is extremely niche and hardly has any internet resources. To do this, he used 64k translation pairs and trained a specialized model.
And after the appearance of the Claude 3 opus, he tried to give only the opus 57k random word sentence translation pairs, these datasets don't even cover the full vocabulary.
In this case, he asked opus to translate I Lay in Bed from Russian to Circassian. As a result——
Opus not only provides a perfect translation, but even breaks down the grammar and morphology!
The netizen was shocked. Subsequently, he came up with a very rare sentence that was absolutely impossible to appear in the dataset.
As a result, Claude 3 once again gives a perfect translation and analysis!
Looking at the answer in front of him, he couldn't believe his eyes.
In subsequent tests, he found that Claude 3 consistently demonstrated a deep grasp of linguistic structures and was able to infer unknown words, whether for complex passages in literary works, recent news articles, or Circassian dialect texts with distinctly different grammars and different writing systems.
You know, for the Circassian language, which has a very complex grammar, this netizen spent two whole years researching.
And opus in less than a minute, only 5With 7k random translation pairs, you can grasp the subtleties of this language!
GPT-4 was defeated.
Moreover, even the model he called up himself could not do the same thing.
This netizen exclaimed-
In contrast, I tried the same test on GPT-4 and it completely failed. Refuse to translate even the simplest sentences, let alone master the intricacies of grammar. I've also tried fine-tuning gpt-3 on a similar dataset before5. It turns out to be just noise.After reading this example, some netizens said, don't say anything, use Claude 3 to decipher the oracle bone inscription, hurry up!I don't know what anthropic has done with this model, but it's completely different from any other.
Many people are skeptical about its lead in the composite benchmark, but the results I saw were striking, and this benchmark had a 0% chance in the training dataset!
It would have been years later, but today it happened.
It was also discovered that Claude 3 Opus was the first language model in the world to translate Parthian almost perfectly.
Parthian is an ancient dead language, and this northwestern Iranian language has been dead for about 1,500 years.
However, unlike the Circassian language mentioned above, there is a Parthian text available on the Internet, so LLMs are not completely unfamiliar with it.
However, OPUS seems to be the first LLM capable of actually translating Parthian, while both Sonnet and GPT-4 failed.
Note that the translations made by opus are done without contextual learning.
The original Parthian language was like this-
ud kaδ naxwišt pad šahr awištād ahēm, gōrdanyos kēsar až hamag frōm, gōt ud garmāniyā šahr zāwar hangāwišn kerd; ud ō asūrestān abar ērānšahr ud amā āa]d, ud pad asūrestān m[arz] pad mišīk paddēmān wuzurg zambag būd.opus gives the correct translation -
and when i first stood in the kingdom, gordian caesar gathered forces from all of rome, goth, and germania; and he came to assyria against iran and us, and there was a great battle at the frontier of assyria in the province of misik.In contrast, the translation of GPT-4 is much inferior.When I first stood on the land of the kingdom, Gordian Caesar gathered troops from Rome, Goth, and Germania; He came to Assyria to fight Iran and us, and there was a great battle on the Assyrian border in the province of Mysik. ）
and when unrest spread throughout the city, gordianus caesar from all of rome, goth and germania, made the city his stronghold; and he crossed over to assyria against iran and brought [it] forth, and in assyria, in the region of mesik, he was a great noble.As a result, netizens threw out cold language passages to test Claude 3.When the riots spread throughout the city, Goldianus Caesar, from Rome, Goth, and Germania made the city his stronghold; He crossed Assyria to attack Iran and brought it out, and in the Misik region of Assyria he was a great nobleman. ）
Education is being turned upside down?
After seeing the case where AI can provide so much cutting-edge knowledge in various disciplines, netizens were horrified that the current education system is useless!
All are witnessing the demise of traditional style education. Because of the sharp rise in the cost of education, traditional education in the United States is dying, and Claude 3 will undoubtedly accelerate this death.The student can learn whatever he wants at the pace that he feels most comfortable with, and the teacher will be very patient.
And this netizen himself has automated most of his work through ChatGPT and Exel, and this process has not consulted any teachers.
What is the best form of education that people can accept? A private team of 10 one-on-one tutors, each of whom is a master in their field. This is an education that only the elite in the history of mankind can have.We will wait and see what kind of changes will occur in the traditional education process and system designed for the industrial age in the future!It is expected that in the near future, education for the masses will begin to reach this level. That alone will change the course of human civilization.
I am a science and technology creator