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

JPL, short for Jet Propulsion Laboratory, is a core research and development institution under NASA responsible for unmanned deep space exploration, but its ownership and management belong to Caltech.

This is also a unique model in the world where NASA provides the funding and others manage it. Therefore, JPL's staff includes NASA employees, Caltech professors and graduate students, as well as a large number of long-term contract workers.

This unusual pattern stems from the origins of JPL. In 1936, five graduate students at Caltech, under the guidance of their advisor Theodore von Kármán, conducted rocket propulsion research. Along with a group of classmates, they conducted rocket engine experiments on the Arroyo Sekogan Riverbed behind the campus, often resulting in explosions that sent rockets flying everywhere, leading to them being chased by school security guards. This informal rocket group was the precursor to JPL.

It's worth mentioning that many people in China believe Qian Xuesen was one of those five graduate students, but that's not the case.

Qian Xuesen joined in 1937, likely brought in by Arroyoseko, and was responsible for research on compressible flow boundary layers.

However, by the time JPL was officially established in 1944, Qian Xuesen had already become a core member of the rocket team thanks to his skills, and was therefore officially designated by von Kármán as one of the founders of JPL.

After NASA was established in the United States in 1958, JPL was transferred to NASA by the U.S. military, but management rights remained with Caltech.

Their achievements were also remarkable. In 1965, Mariner 4 took the first close-up photos of Mars.

In 1976, Viking 1 achieved the first soft landing on Mars. In 1997, Mars Pathfinder sent the first rover, Sojourner, to Mars. In 2004, the Spirit and Opportunity rovers landed on Mars.

然后就是今年2011年11月26日，好奇号火星车发射升空，预计2012年8月着陆火星。

At that time, JPL was not located on Caltech's main campus, but in Flintridge, Canada, a 20-minute drive from Pasadena.

Chen Yiran's main purpose in coming here was to meet with the new project team, which is still in the conceptual design stage. They expect to launch a new Mars rover again in six to seven years.

These engineers clearly had a better understanding of Deepflow, and they made a specific request that Chen Yiran distill a tiny AI-assisted navigation system using a better and more suitable large model so that they could put it into the small chip of the Mars rover.

The selection logic for aerospace chips is completely different from that in the civilian field. Reliability, radiation resistance, and long lifespan are prioritized far over peak performance. The Curiosity rover, which just launched, uses a custom-made RAD750 chip from BAE Systems in the UK.

It uses a 250nm process, has a TDP of only 5W, a clock speed of 200MHz, and 256MB of memory. While its performance isn't high, it can withstand radiation levels as high as 2,000-10,000 Gy.

This chip is derived from the PowerPC 750 used in the Apple iMac G3, an old relic that Apple used in 1997.

With so many restrictions, they need an extremely lightweight AI, only a few tens of KB in size, that can run for only a few frames at a time.

Beyond navigation, it would be even better if this AI could perform preliminary screening, since the most expensive component on the Mars rover isn't computing power, but rather communication bandwidth. The bandwidth between Mars and Earth is approximately 2-8 Mbps, with only a few tens of minutes of transit window per day.

It's impossible to send all the photos back, so onboard intelligence is needed to prioritize the data.

JPL actually really wanted to hand these tasks over to Chen Yiran directly, since when it comes to understanding AI, no one in the world dares to say they are ahead of Chen Yiran.

However, this year the U.S. Congress passed the Wolf Amendment, which prohibits any joint research activities between the U.S. and China that are related to NASA or coordinated by the White House Office of Science and Technology Policy.

They could only explore these tasks on their own after obtaining authorization from Chen Yiran.

Chen Yiran was speechless at this clause. Looking up, he saw that Asians made up 3% of JPL's workforce, with the majority of them being of Chinese descent. Even if this clause in the US wouldn't push these people to China, it could very well lead them to switch jobs to private companies like SpaceX.

This wasn't just idle talk, because after the tour, he saw the future trillion-dollar tycoon, SpaceX creator Elon Musk, talking to several engineers at the entrance.

Upon seeing Chen Yiran approach, Musk immediately greeted her: "Chen, it's a pleasure to meet you. Do you have time to chat?"

"Let's go to that coffee shop over there."

Chen Yiran had expected Musk to talk at length about AI, given that Musk's entanglement with OpenAI would be headline news in 2026.

Although it is legally difficult for Musk to reclaim OpenAI, various published details prove that Musk was the original promoter of OpenAI.

He first proposed the concept in 2015, collaborating with Altman and others, with the initial goal of creating open-source, non-profit general AI. This was also to counter the then-overpowered Google, which had already acquired the world's strongest deep learning teams.

However, both founders, Altman and Greg Brockman, deviated from their original goals. They even wrote frankly in their diaries that their goal was to first earn a billion dollars and then kick Musk out.

It has to be said that Musk can be quite foolish at times, getting badly beaten in office politics and ultimately losing all of his rights in ChatGPT because the lawsuit was filed too late.

Surprisingly, Musk praised Chen Yiran's work highly and did not question Chen Yiran's closed-source strategy as he did with Google.

His main hope is to get help to use deep learning to solve the problem of insufficient landing accuracy during SpaceX's current recovery process.

"We're stuck in the final 500 meters of the landing zone after the Falcon 9 reentries. The onboard radiation-hardened chip doesn't have enough computing power to handle the high-precision visual feature matching. If possible, could you help us?"

The two then began to talk about rocket recovery technology. In the last few dozen seconds of recovery and landing, the system needs to decide in real time when to shut down the outer engines, when to reignite the central Merlin engine, and how much throttle to apply.

The optimization tools for this part are good at handling continuous variables, but these switching decisions are discrete. Currently, SpaceX is still relying on manually pre-setting a fixed timeline to slowly optimize, but the fuel reserves and atmospheric conditions are different for each mission, and the landing accuracy is stuck on these discrete decisions.

【^(*￣(oo)￣)^ I think we should probably go back to university and take some classes again. The material these past few days has been so troublesome, but I think I could understand some of today's lesson, it was better than yesterday's.】

Indeed, the projects of the past two days have begun to enter Chen Yiran's blind spot, but compared to quantum computers, SpaceX is obviously more attractive to him.

He thought for a moment and said, "How about this, I'll visit your headquarters in a couple of days? But let me make it clear first, I definitely won't work for free."

Musk laughed and said, "Definitely. Although our funding issues have been somewhat resolved, I would definitely accept your investment if it were possible."