Hello and welcome to Sync #488!

It’s been a big week for AI community as AI researchers were awarded with not one but two Nobel Prizes—one in physics and one in chemistry.

In other news, Tesla revealed Cybercab—the long-awaited robotaxi, and Robovan—their new vehicle, both to mixed reception. Meanwhile, Meta has unveiled Meta Movie Gen, their AI-powered video generator, and Amazon announced new warehouses that will employ 10 times as many robots.

Elsewhere in AI, Microsoft doubles down on ever-present AI assistants, and AMD launches an AI chip to rival Nvidia. We’ll also explore the damage caused by a racist AI deepfake that fooled a community.

We’ll finish this week’s issue with the Nobel Prize in Medicine for the discovery of microRNA and a conversation on how the semiconductor industry actually works.

Enjoy!

Machine learning wins two Nobel Prizes

Over the past 10 years, machine learning and AI have been making a greater and greater impact on our lives. During this time, researchers have published breakthrough results one after another, from enabling computers to recognise images, sometimes better than humans, to modern large language models transforming how we work and interact with computers. Over these 10 years, virtually every field of human activity, from healthcare and medical research to education, engineering, business, and entertainment, has in one way or another been transformed by machine learning and AI.

The Royal Swedish Academy of Sciences acknowledged the impact machine learning and AI have had in recent years by awarding not one but two Nobel Prizes to researchers whose work has left a massive impact not only on AI research but also on the lives of millions of people.

John J. Hopfield and Geoffrey E. Hinton win Nobel Prize in Physics 2024

The Royal Swedish Academy of Sciences awarded the 2024 Nobel Prize in Physics 2024 to John J. Hopfield and Geoffrey E. Hinton for their “foundational discoveries and inventions that enable machine learning with artificial neural networks.”

Hilton said at the press conference with the University of Toronto, where he is professor emeritus, that this award should be seen as a recognition for a large community of people who have been working for years on machine learning and neural networks.

After seeing this announcement, some might ask why these two researchers, best known for their work in artificial intelligence, won the Nobel Prize in Physics. However, if you look closer at computing, you'll notice many parallels between computing and physics. Information theory, the backbone of computer science, is essentially based on thermodynamics. Even though both laureates work on neural networks, they borrowed many concepts from physics, math, and neuroscience to achieve their breakthrough discoveries.

The Royal Swedish Academy of Sciences awarded John Hopfield for the invention of a neural network known as Hopfield network. Hopfield network a type of recurrent neural network designed to save and recreate patterns, such as images. Inspired by the physics of atomic spin systems, where each atom acts like a tiny magnet, the network adjusts the connections between its nodes (representing pixels) so that saved patterns correspond to these low-energy configurations.

Geoffrey Hinton received the Nobel Prize for his work on using the Hopfield network as the foundation for developing the Boltzmann machine, a type of neural network that can learn to recognize and generate patterns in data. His work incorporates tools from statistical physics, and the Boltzmann machine has applications in image classification and pattern generation.

However, Hinton’s impact on machine learning goes well beyond his work on Boltzman machines.

Together with his colleagues, David E. Rumelhart and Ronald J. Williams, Geoffrey Hinton invented the backpropagation algorithm in 1986. Backpropagation is the cornerstone of modern artificial neural networks. It is the algorithm that allows neural networks to adjust their weights and learn from the mistakes made during the training process. Backpropagation is one of the most important algorithms that every practitioner of machine learning and AI must master.

Hinton's other major contribution to the field was the 2012 paper ImageNet Classification with Deep Convolutional Neural Networks, co-written with Alex Krizhevsky and Ilya Sutskever, the future co-founder and chief scientist at OpenAI. That paper proved that deep neural networks can be highly effective in image recognition and demonstrated how using GPUs could significantly improve their performance. It is one of the most important papers in computer science and essentially kickstarted the deep learning revolution of the 2010s.

Hinton then joined Google Brain, where he continued his research, exploring and developing cutting-edge neural network architectures and algorithms, contributing to advancements in areas such as image recognition, natural language processing, and other AI applications.

For his work on neural networks and deep learning, Geoffrey Hinton gained widespread respect in the community and earned the title "Godfather of AI." In 2018, alongside Yann LeCun and Yoshua Bengio, two other "Godfathers of AI," Hinton received the Turing Award, often referred to as the "Nobel Prize of Computing."

In 2023, Hinton left Google and began advocating for taking AI safety seriously, becoming one of the most prominent voices calling for increased research into AI safety. In a press conference hosted by the University of Toronto, Hinton even mentioned that he is proud of the fact that one of his students fired Sam Altman. That student was Ilya Sutskever, who led an unsuccessful coup against Altman nearly a year ago.

David Baker, Demis Hassabis and John M. Jumper win Nobel Prize in Chemistry 2024

The second Nobel Prize for AI was awarded to Demis Hassabis and John Jumper for the creation of AlphaFold2, the AI model that solved the protein folding problem. They shared the prize with David Baker, who used computational protein design to deepen our understanding of protein prediction and design, enabling the creation of novel proteins to address some of the greatest challenges in medicine, technology, and sustainability. Baker received half of the prize, while Hassabis and Jumper jointly received the other half.

For 50 years, researchers tried to solve the protein folding problem—predicting a protein's three-dimensional structure based solely on its DNA sequence. In 2020, researchers at DeepMind announced that this grand challenge in biology had been solved with AlphaFold. Two years later, DeepMind used AlphaFold2 to predict the structures of over 200 million proteins known to humanity and made them freely available for everyone to use through the AlphaFold Protein Structure Database.

AlphaFold is a massive breakthrough in computational biology. It is already being used by many companies and research labs to discover new drugs more cheaply and quickly than before. More advanced tools, built on top of AlphaFold and similar models, could be applied not only in healthcare and pharmaceuticals but also in other fields and industries, helping to solve challenges from waste pollution to the creation of new materials. What we are witnessing now could be just the beginning of a massive transformation in biomedical research and bioengineering.

Although the he Royal Swedish Academy of Sciences awarded the Nobel Prize to Demis Hassabis and John Jumper, it is better to see it as a acknowledgment for the work done by the entire team behind AlphaFold and for DeepMind as a whole. The Nobel Prize can be only awarded to individuals, not organisations, hence why it was awarded to Hassabis, the co-founder and CEO of DeepMind, and John Jumper, the director at DeepMind and lead author of the paper that introduced AlphaFold. However, next to Hassabis’ and Jumper’s names, there are 32 other names as authors of that paper and very likely more people at DeepMind contributed to the project.

What do this year’s Nobel Prizes mean for AI and science?

This year’s Nobel Prizes for AI raise two questions.

Firstly, these awards open the possibility for more breakthrough achievements in AI to receive a Nobel Prize. Could the people behind the invention of the transformer model, which is the basis of modern large language models? How about the diffusion models, which power text-to-image generators, or any future breakthrough machine learning methods, be considered as potential Nobel Prize laureates?

The second question concerns the impact of AI in science. There is no doubt AI will play an increasingly significant role in future discoveries. The scale of problems scientists are facing today means they can do very little without powerful machine learning algorithms to help find patterns and answers in the enormous amounts of data points they must deal with. Because of this, I’m confident we will see more researchers receiving Nobel Prizes, either for creating a breakthrough AI model or for heavily relying on one in their research.

Now, with more powerful AI models becoming better at reasoning, there are discussions about creating AI scientists—AI models that can independently research a problem and propose solutions to it. Proponents of AI scientists envision a future in which such models will first augment the work of scientists, acting like research assistants and helping to find solutions to significant problems and challenges. However, there may come a point when these AI scientists advance from research assistants to becoming better than humans at solving the major issues facing humanity. If (or when) such an AI scientist is created and causes a breakthrough in science, it would raise questions about the role of human scientists in advancing our understanding of the world. Would the Royal Swedish Academy of Sciences award a Nobel Prize to an AI, rather than to the people who used or built that AI?

One thing is for sure—AI is transforming how science is done, and we can expect more breakthrough results achieved with the help of artificial minds that will transform our lives.

If you enjoy this post, please click the ❤️ button or share it.

Share

Do you like my work? Consider becoming a paying subscriber to support it

Become a paid subscriber

For those who prefer to make a one-off donation, you can 'buy me a coffee' via Ko-fi. Every coffee bought is a generous support towards the work put into this newsletter.

Your support, in any form, is deeply appreciated and goes a long way in keeping this newsletter alive and thriving.

🔮 Future visions

▶️ The Day AI No Longer Needs Us (11:45)

John Michael Godier feels spooky this time of year and paints a picture of a not-so-distant future where machines are capable enough to take over human tasks. Godier envisions fully automated factories and entire industries, autonomous flying drones that can recharge themselves using power lines, beyond human control and unable to be "unplugged," and self-improving AIs that is beyond human comprehension. He also addresses the economic and societal risks of widespread automation, such as unemployment and the potential collapse of industries when too few consumers remain. Ultimately, the video touches on existential risks related to AI, including the fear that AI may surpass human intelligence and independence, making humans obsolete or irrelevant. And as Godier suggests, this could happen within this century.

🧠 Artificial Intelligence

Meta Movie Gen
Researchers from Meta AI have introduced Meta Movie Gen, their latest AI video generator, similar to OpenAI’s Sora, DeepMind’s Veo, and other models. The new model is capable of generating high-quality, 1080p HD videos with different aspect ratios and synchronized audio from a text prompt. The model is not publicly available yet, but Meta has published a playlist of videos generated by Movie Gen. There is also a paper describing the model in more detail. As Meta writes in this blog post, the company is currently working closely with filmmakers and creators to integrate their feedback before making it widely available on Facebook, WhatsApp, or Instagram.

Microsoft: 'ever present' AI assistants are coming
Speaking with the BBC, Mustafa Suleyman, head of AI at Microsoft, predicts that AI assistants with robust long-term memory, able to recall conversations and projects, are about a year away, and sees a future in which these tools work as persistent co-pilot companions in everyday life. While this deep integration raises concerns about privacy, security, and potential biases, Suleyman argues that privacy expectations are evolving, citing the widespread use of continuously recording devices, such as TVs, laptops, phones, in-car cameras, and earbuds. He also dismisses the idea that AI is a bubble, saying that AI is the fastest-growing technology, although it may not achieve smartphone-level popularity.

The racist AI deepfake that fooled and divided a community
Last December, a clip of Pikesville High School principal Eric Eiswert making racist and antisemitic remarks emerged and went viral, sparking outrage, death threats, and significant turmoil in the community near Baltimore, Maryland. The clip, however, was faked, and this whole story is a reminder of how much damage an AI deepfake can cause, leaving Principal Eiswert to move to another school and causing long-lasting emotional scars in the community.

AMD launches AI chip to rival Nvidia’s Blackwell
AMD has launched a new AI chip, the Instinct MI325X, aimed at competing with Nvidia's Blackwell GPUs in data centres. Expected to start production by the end of 2024, AMD’s new chip could challenge Nvidia’s market dominance and put pricing pressure on Nvidia's highly profitable GPUs. AMD also introduced new CPUs to better support AI workloads, as it competes with both Nvidia and Intel in the data centre market.

China Urges Local Companies to Stay Away From Nvidia’s Chips
In the latest episode of the chip war, Beijing is encouraging Chinese companies to purchase domestically produced AI chips instead of Nvidia’s H20 chips. The policy is not an outright ban but a form of guidance to help domestic AI chipmakers like Cambricon Technologies and Huawei gain market share, according to Bloomberg. Some Chinese companies are buying Huawei chips to comply with government guidance but are also rushing to acquire Nvidia chips in anticipation of further US sanctions. Nvidia shares declined following the report of China’s guidance, though the company’s sales have surged globally, driven by high demand for AI processors.

The Open Source AI Definition RC1 is available for comments
The Open Source Initiative has released the first release candidate of the definition of Open Source AI. The definition is available here. In short, for an AI model to be called open source, it needs to fulfil four essential freedoms. For these freedoms to be meaningful, the AI system must provide access to essential elements like training data, source code, and model parameters, enabling others to modify and replicate the system. The goal is to promote transparency, reuse, and collaborative improvement, following open source principles.

We’ve entered the AI grift era
The hype around AI and the billions of dollars poured into AI companies can attract talented people building the next big thing, but it also attracts grifters who want to ride the wave and get rich quickly. This article highlights PearAI as an example of a young AI company that cloned two preexisting tools, VS Code and Continue, to build their own AI-powered code editor with little to no regard for the open-source licences they broke along the way. The article also cites Demis Hassabis, who said that cash flooding the AI space “brings with it a whole attendant bunch of hype and maybe some grifting,” and draws comparisons between the present AI space and crypto.

Can Agents Spontaneously Form a Society? Introducing a Novel Architecture for Generative Multi-Agents to Elicit Social Emergence
In this paper, researchers placed AI agents without predefined personalities or roles into a virtual town and discovered that they began to form social relationships. They observed the spontaneous emergence of behaviours such as forming cliques, establishing leadership hierarchies, and organising collective activities. These results are fascinating as they demonstrate that large language models have social interactions encoded within them.

If you're enjoying the insights and perspectives shared in the Humanity Redefined newsletter, why not spread the word?

Refer a friend

🤖 Robotics

Tesla's Cybercab Is Here
At the "We, Robot" event in Las Vegas, Tesla revealed two new vehicles—the Cybercab, a self-driving taxi prototype, and a Robovan capable of transporting up to 20 people. Tesla CEO Elon Musk announced that the Cybercab would enter production by 2026, aiming for a price under $30,000, with full self-driving technology becoming available in California and Texas by next year. However, Musk did not provide specifics about production plans or timelines for the Robovan during the event. Investors were unimpressed with Tesla's presentation, leading to a 9% drop in Tesla’s stock, wiping $67 billion from the carmaker’s market valuation. The entire "We, Robot" presentation is available on X (just skip the first 52 minutes).

Inside 1x’s plan to test humanoids in homes
The Robot Report sits down with Bernt Børnich, CEO and founder of 1X, a robotics company building humanoid robots. In this conversation, Børnich shares his vision of first bringing humanoid robots to homes, then introducing them into industry. He argues that transformative technologies, such as computers and cars, began to accelerate and find broader applications once they proved valuable for consumers. Additionally, starting with humanoid robots for home use forces their design to be affordable and safe, which will help build the scale necessary for humanoid robots to succeed. The discussion also touches on public perception of robots and the future of humanoid robotics in society. The interview starts at 36:28.

Will humanoid robots take off? Conrad Gray · Jun 11 Read full story

Sidewalk robots are teaming with drones for Dallas food deliveries
Serve Robotics and Wing, Alphabet’s drone delivery company, have announced a partnership to improve delivery efficiency in Dallas. Serve's four-wheeled sidewalk robots will transport orders from restaurants to a designated area, where Wing's drones will then pick up the packages and deliver them to customers up to six miles (about 10km) away. The partnership aims to leverage both companies' strengths to provide a more dependable food delivery solution.

Amazon’s new warehouses will employ 10x as many robots
At its "Delivering the Future" event, Amazon announced its first "next-generation fulfillment center" in Shreveport, Louisiana, a massive 3-million-square-foot facility (roughly the equivalent of 55 football fields) spanning five floors. This new warehouse will deploy 10 times more robots than a standard fulfillment center, including advanced systems like Sequoia, which can store over 30 million items and enhance employee safety and efficiency. Despite the significant increase in automation, the center will employ 2,500 people.

How robots are building Amazon's future Conrad Gray · December 4, 2023 Read full story

▶️ Agile perching maneuvers in birds and morphing-wing drones (2:19)

In this video, robotics researchers share the results of their work to better understand how avian perching manoeuvres work by first simulating and then building a bird-like drone that can morph its wings like a real bird. Their study not only helped us better understand the mechanics of birds' flight but also showed that drones can be used to study birds' flight and perform experiments that would be considered unethical if a real animal were used.

🧬 Biotechnology

Nobel Prize in Medicine for the discovery of microRNA
The 2024 Nobel Prize in Physiology or Medicine has been awarded to Victor Ambros and Gary Ruvkun for their discovery of microRNA and its role in regulating gene activity. MicroRNA is a small, non-coding RNA molecule that plays a crucial role in gene regulation by binding to complementary sequences in messenger RNA (mRNA) to inhibit protein production. It plays a vital role in the development and function of multicellular organisms, helping coordinate networks of genes. Dysfunction in microRNA regulation has been linked to diseases such as cancer and genetic disorders.

World-first therapy using donor cells sends autoimmune diseases into remission
Three patients in China with severe autoimmune diseases went into remission after receiving bioengineered and CRISPR-modified CAR T cell therapy, which uses immune cells from donors rather than their own bodies. None of the treated patients experienced severe inflammatory reactions or signs of graft-versus-host disease. More studies are needed to assess the long-term efficacy of the therapy, but if it continues to show success in more patients and over a longer period, it could revolutionise the treatment of autoimmune diseases.

Global effort to map the human brain releases first data
The BRAIN Initiative Cell Atlas Network, part of the NIH’s effort to create a comprehensive atlas of the human brain, has released its first major dataset. This includes single-cell and single-nucleus transcriptomic and epigenomic profiles from humans, mice, and 10 other mammalian species, aimed at defining brain cell types based on molecular profiles. The data, available online for free, is intended to accelerate research by encouraging early data sharing and collaboration. This initiative aims to provide a deeper understanding of the human brain, ultimately aiding the development of more precise treatments and cures for brain disorders.

Human Stem Cells Repair Retinas in Monkey Models of Macular Holes
Researchers from Japan demonstrated that human embryonic stem cell-derived retinal sheets successfully repaired macular holes in monkeys. The new therapy closed macular holes, improved eye fixation, and increased responsiveness to light. However, as is always the case in these kinds of research projects, more research is needed before this treatment can be considered safe and effective for humans.

💡Tangents

▶️ How the Semiconductor Industry Actually Works (2:10:53)

In this podcast, Dwarkesh Patel speaks with Dylan Patel from SemiAnalysis and Jon from Asianometry about the mindblowingly complex nature of semiconductors, different ways China could catch up to and even outcompete the US in AI, the datacenter boom, and behind-the-scenes stories of the semiconductor industry. The conversation spans over two hours and is equally insightful, intense, and entertaining, and I highly recommend listening to it.

Thanks for reading. If you enjoyed this post, please click the ❤️ button or share it.

Share

Humanity Redefined sheds light on the bleeding edge of technology and how advancements in AI, robotics, and biotech can usher in abundance, expand humanity's horizons, and redefine what it means to be human.

A big thank you to my paid subscribers, to my Patrons: whmr, Florian, dux, Eric, Preppikoma and Andrew, and to everyone who supports my work on Ko-Fi. Thank you for the support!

My DMs are open to all subscribers. Feel free to drop me a message, share feedback, or just say "hi!"

Message Conrad Gray