Reborn in 2010, the AI ​​in my mind is too powerful.

Although DeepFlow was released by Chen Yiran a year and a half ago, it has not actually been promoted on a large scale in academic circles.

They simply used it quietly in their various technologies, and its use in various papers relied entirely on word-of-mouth from universities, using its actual integration with two GPUs to quietly promote it.

There's nothing wrong with letting things run wild like this when there are no troublemakers.

However, now that Google's Transformer has begun large-scale sponsored promotion, Chen Yiran can no longer ignore it.

After all, Google's strategy is quite crazy. In addition to various research grants, they have even sent coffee machines with the Google logo to many schools.

They adopted a broad-based strategy, trying to recruit any reputable school with a legitimate CS program.

No matter how shabby or flawed the project is, they will give it some money, trying to create a semblance of prosperity.

Some state universities with considerable resources are willing to invest heavily, readily providing millions of dollars worth of resources just to switch their entire experimental area to Transformer. Google has already announced that it will allocate five billion US dollars this year for the initial development of the Transformer ecosystem.

Chen Yiran's strategy differs from Google's. He has a technological advantage and doesn't need to engage in a war of attrition with Google. He just needs to ensure that his technology is not buried.

Therefore, he approached some of the top schools in the US, with Silicon Valley being the primary focus. Chen Yiran provided financial support to Stanford and Berkeley in exchange for their backing for DeepFlow.

Although Google offered better terms, given the choice, the project teams at both universities firmly chose Deepflow, which offered better performance, greater potential, and a higher chance of publishing papers.

Then other well-known universities across the United States, such as MIT, Carnegie Mellon, UC Champaign, and UC Wisconsin, also received small donations from Chen Yiran.

Chen Yiran is currently visiting Caltech, which is also one of the most important ones.

Caltech was founded in 1891 by Amos G. Slupp and is one of the world’s top science and technology research universities, although its reputation is far less than that of MIT.

However, in the actual hierarchy of STEM fields, Caltech is actually above MIT. Whether it's the density of Nobel Prizes or the depth of theoretical research, Caltech is the strongest.

Of course, if we compare the breadth of engineering, the transformation of computer science research into practical applications, and the entrepreneurial activities of alumni, MIT and Stanford would not be too convinced, while Carnegie Mellon could also compete to some extent.

Caltech has just over two thousand students, less than a tenth of Stanford's, but its influence in physics, astronomy, aerospace, and quantum computing far exceeds its size.

More than 30 alumni and faculty members have won Nobel Prizes, and half of the framework of the entire field of quantum information theory was built by Caltech professors.

The campus is located in downtown Pasadena; it takes about fifteen minutes to walk through the entire campus from east to west.

The architecture is nothing special; red tiles and white walls, with arcades and courtyards are typical of the Spanish colonial revival style common in California.

There are only two places that are somewhat unique. One of them is a turtle pond in the center of the campus, where several painted turtles and red-eared sliders have been kept for a long time. It is unknown which class of students put them in, but now it has become part of Caltech's tradition.

Another interesting feature is the old-fashioned naval cannon on the central lawn of the campus. This Fleming cannon is a traditional symbol of student pranks and inter-school rivalries.

What's rather amusing is the close, love-hate relationship between the well-known USC and UCLA in Los Angeles.

Caltech's main rival is MIT, located far on the East Coast. The 4200-kilometer distance cannot stop the mischievous spirit of Caltech's students.

In 1987, Caltech students broke into the MIT campus, cut up a massive 1.7-ton concrete school name sign (the letter I for MIT), and transported it back to California.

In 1989, after two years of careful preparation, MIT students finally stole a 1.3-ton marble sundial from Caltech's campus.

In 2005, MIT learned in advance of Caltech students' hacking plan to disrupt the Rose Bowl and acted preemptively by changing all names in the Caltech system to MIT.

In 2010, after years of hard work, Caltech students successfully breached MIT's defenses, taking over the admissions office website and changing the welcome banner to their own admissions office URL, along with a sarcastic welcome message.

Compared to the struggles in other schools, the conflict between these two schools illustrates how terrifying it is for bad people to be educated; even their pranks are on a completely different level.

These amusing stories were told to Chen Yiran by the principal with great pride during lunch.

Caltech showed Chen Yiran considerable respect, with President David Baltimore personally receiving him during both his morning visit and lunch in the cafeteria.

This is no unknown figure; he independently discovered reverse transcriptase in 1970 and was awarded the Nobel Prize in Physiology or Medicine in 1975 for it.

The communication between the two parties went smoothly, and Chen Yiran immediately agreed to the subsequent sponsorship contract. David Baltimore's wife is of Chinese descent, which gave him fluency in Mandarin, and he even showed it off to Chen Yiran.

After the most important matter was settled, Chen Yiran met with John Preskill, a theoretical physicist, in the afternoon. Preskill is also the current director of the Quantum Information and Matter Institute at Caltech.

Upon meeting Chen Yi, John Preskill said directly and urgently, "Mr. Chen, my current work focuses on the fault tolerance threshold of topological quantum error-correcting codes. Specifically, we are studying how to implement a set of transverse Clifford gates on surface codes by weaving non-Abelian anyons."

Simultaneously, the fidelity of the logic qubits is maintained at or above the decoherence threshold during the magic-state distillation process. The key issue is that the synthesis measurement of the stabilizer generator introduces Pauli noise channels in each error correction cycle. If the synthesized graph structure does not commensurate with the boundary conditions of the underlying lattice, the entire fault-tolerant computation will degenerate into a classical probabilistic mixed state.

Chen Yiran was completely bewildered and could only ask Jasmine in his mind: What? What is he saying?!

【(￣△￣；) You studied chemical engineering in your past life, while we've been studying computer science in this life. I don't understand what he's talking about either!】

Chen Yiran could only ask somewhat awkwardly, "Dr. Preskill, how should I put it about my physics level? I've never actually been to university, not even for a single day. All my current achievements are entirely self-taught."

Unfortunately, I haven't studied physics, so if you need my help, you'll at least need to translate your problem into a purely mathematical one for me.

Preskill thought for a moment and said, "First, consider a 4096 by 4096 matrix, where less than seven percent of the elements are not zero. The distribution of the non-zero elements follows a fixed, repeating pattern, and I will send you a list of the coordinates of that pattern."

Then, we need to use an initial vector to repeatedly multiply this matrix. Each time we multiply, we get a new vector, and then we use the same matrix to multiply that new vector, and so on for thousands of rounds.

After each round of multiplication, a global norm check needs to be performed by summing the squares of all elements of the entire vector and checking if the result equals one. If there is a discrepancy, the entire round's result is discarded.

Because this process takes thousands of rounds, the tiny floating-point error from the previous round will accumulate in the next round, so the floating-point precision cannot be less than one part per hundred thousand of the relative deviation.

I don't have such tools right now, so I hope you can help me update a dedicated version for DeepFlow computation.

He was probably using NVIDIA CUDA before, and the default floating-point precision would drift in extreme accumulation scenarios. We created a new version for him with optimized matrix pattern matching and global norm checking, which wasn't difficult and could be done in two or three days.

But his gaze seemed really offensive! Let's go back to studying physics! (＠￣ー￣＠)

Chen Yiran was certain of Jasmine's idea. Preskill's eyes were undoubtedly filled with the arrogance unique to top physicists. He was prepared to tell the young man not to be underestimated when he got back, but for now he needed to stall for time and let the old professor know that one must have the right attitude when asking for help.

"Sure, I'm not good at physics, but you're terrible at this. Give me two days, and I can figure it out. But I might not be free right now. I'm going to visit the JPL Jet Propulsion Laboratory tomorrow. I'll contact you when I have time."