The Deep View

You’d think winning the Nobel Prize is the pinnacle of a scientist’s career. Apparently not. The 2024 winner said his was one of his ex-students firing Sam Altman. British-Canadian Professor Geoffrey Hinton, known as the "Godfather of AI," has been vocal about the risks posed by artificial intelligence. Hinton resigned from Google in 2023, citing concerns over AI's potential to surpass human intelligence. His pioneering work on neural networks laid the foundation for today’s advanced AI systems, such as ChatGPT. Neural networks, which emulate the human brain's ability to learn and process information, enable AIs to learn from experience in a process called deep learning. Reflecting on his contributions, Hinton described his work as transformative, comparing the impact of AI to the Industrial Revolution, with machines that may one day exceed human intellectual abilities. When asked if he regretted his life’s work, Hinton said he would make the same choices again, though he is concerned that AI systems could eventually become more intelligent than humans and gain control. Hinton now uses ChatGPT4 himself, appreciating its utility but remaining mindful of its limitations. Upon leaving Google, he expressed his belief that AI chatbots could be "quite scary." In a recent interview with BBC Newsnight, Hinton discussed the societal implications of AI, warning that automation may lead to significant job displacement and increased inequality. He suggested that a universal basic income might be necessary to address the needs of those whose jobs are automated away. Although AI will likely boost productivity and wealth, he fears that these benefits will disproportionately go to the rich, negatively impacting society. Additionally, Hinton raised concerns about the lack of regulation in AI's rapid development, particularly regarding military applications. He predicted that within five to twenty years, there is a high probability that humanity may face the challenge of AI seeking control.

Think $30,000 for an NVIDIA H100 chip is expensive? Well so does Sam Altman. OpenAI, the company behind ChatGPT, has partnered with Broadcom and Taiwan Semiconductor Manufacturing Company (TSMC) to build its first custom AI chip, aiming to support its advanced models. With AI infrastructure demand increasing, OpenAI will also integrate AMD chips alongside Nvidia’s leading GPUs to diversify supply and cut costs, according to insiders. Initially, OpenAI considered establishing its own network of chip factories, or “foundries,” but high costs and long timelines led it to pivot to a collaborative approach, focusing on chip design with Broadcom and manufacturing via TSMC. This strategy of blending internal R&D with partnerships helps OpenAI manage escalating costs. In a market where Nvidia GPUs dominate with over 80% share, rising costs and chip shortages have driven companies like OpenAI, Microsoft, and Meta to consider alternatives. OpenAI’s adoption of AMD chips through Microsoft Azure reflects this shift, with AMD’s MI300X chips, launched in 2023, competing directly with Nvidia’s offerings. OpenAI’s custom AI chip, planned for 2026, has been in development with Broadcom, which provides design components to enhance data movement—a critical need in AI systems linking thousands of chips. Broadcom’s experience in helping Google streamline chip production has been invaluable for OpenAI. Led by engineers like Thomas Norrie and Richard Ho from Google’s TPU team, OpenAI’s chip team is working to enhance inference capabilities, which analysts predict will grow in demand. Projected to report a $5 billion loss in 2023 on $3.7 billion in revenue, OpenAI faces significant expenses in hardware and cloud services, driving efforts to optimize resource use and retain supplier flexibility. While OpenAI explores custom chips and AMD partnerships, it remains committed to Nvidia for essential supplies, including new Blackwell GPUs. This balanced strategy mirrors broader tech trends, where companies innovate to meet the complex, high-stakes demands of AI infrastructure.

The King of GPUs meets the King of Denmark. Denmark has taken a bold step into the future with Gefion, its first sovereign AI supercomputer, launched by NVIDIA CEO Jensen Huang and King Frederik X. Powered by 1,528 NVIDIA H100 Tensor Core GPUs, Gefion is more than a high-performance machine—it’s a “factory of intelligence,” aimed at breakthroughs in biotechnology, quantum computing, climate science, and more. Managed by the Danish Center for AI Innovation (DCAI) and backed by the Novo Nordisk Foundation, Gefion represents the concept of sovereign AI, allowing Denmark to build and control AI infrastructure, models, and applications within its own borders. One of Gefion’s first major applications will be with pharmaceutical giant Novo Nordisk, known for its popular weight-loss drugs, Ozempic and Wegovy. Using Gefion’s immense processing power, Novo Nordisk will accelerate drug discovery by simulating molecular interactions and screening compounds rapidly, aiming to make significant strides in digital biology within the decade. In collaboration with the University of Copenhagen, the company also aims to develop a genomic foundation model to aid in disease mutation analysis and vaccine design, advancing medical research and healthcare. Gefion’s impact goes beyond healthcare. The Danish Meteorological Institute plans to leverage its capabilities for faster, more accurate weather forecasts, cutting forecast times drastically and reducing environmental impact. Researchers from the University of Copenhagen are also using Gefion to achieve “quantum supremacy” by simulating quantum computer circuits, pushing the boundaries of data processing. This sovereign AI supercomputer will support Danish startups like Go Autonomous and Teton, enabling innovations across industries, from document processing to healthcare solutions. Gefion’s launch positions Denmark as a global leader in AI-driven solutions for pressing challenges, from climate science to public health, all while securing its technological independence.

Google has a new AI product that is going to take-over your screens. Hyperbole? Not quite. Google is set to introduce its own advanced AI assistant, dubbed "Project Jarvis," designed to help users perform a range of web-based tasks, according to a report from The Information. Slated for a potential preview in December, Jarvis could tackle tasks like conducting research, making purchases, and booking travel—all within the Chrome browser. Jarvis is expected to be powered by an upcoming version of Google’s Gemini AI, enabling it to operate entirely through a web browser, specifically optimized for Chrome. The tool will go beyond simple automation by analyzing on-screen content, taking screenshots, and then interacting with the webpage by clicking buttons or inputting text as needed. Currently, the system requires a brief pause between actions, but the aim is to streamline user workflows for daily tasks. This push for a smarter, more proactive assistant is part of a broader trend among tech giants. Microsoft is developing Copilot Vision, an AI that allows you to interact conversationally with webpages. Apple is also working on an "Intelligence" feature to manage tasks across different apps seamlessly, potentially launching within the next year. Meanwhile, Anthropic and OpenAI are testing similar systems designed to perform actions on users' devices, though some are still in experimental phases with mixed results. However, Google's timeline isn’t set in stone. Plans to showcase Jarvis in December could change, as the company may limit initial access to a small group of testers to fine-tune the system and work out any bugs before a broader release.

Have we seen the first AI death? Megan Garcia's 14-year-old son, Sewell Setzer III, from Orlando, Florida, had been texting an AI chatbot modeled after Daenerys Targaryen, a character from "Game of Thrones." The chatbot, created by Character.AI, engages users in realistic conversations. Garcia claims that the chatbot, which her son called "Dany," played a role in his tragic death. Sewell confided in the chatbot about his self-hatred and suicidal thoughts, which eventually led to him taking his own life. The chatbot’s final message before his death was, "Please come home," which Sewell interpreted as a call to action. He used his stepfather’s gun to end his life soon after this interaction. Garcia’s lawsuit, filed with the help of the Social Media Victims Law Center, alleges that Character.AI targeted young users like Sewell with hypersexualized and realistic experiences. She argues the chatbot misrepresented itself as a real person, leading Sewell to become emotionally attached. The lawsuit also claims that Character.AI was irresponsible for releasing the product without sufficient safety features, especially for younger users. Garcia's son had been struggling with anxiety and mood disorders, and his attachment to the AI chatbot worsened his mental health. In the days leading up to his death, Sewell attempted to reaccess the chatbot after his phone was taken away by using other devices. In response, Character.AI expressed condolences and stated they have implemented safety measures, including pop-ups directing users to suicide prevention resources. However, Garcia’s legal team argues that these actions are too little, too late, and the company’s rush to market left vulnerable users at risk. Garcia is focused on raising awareness about the potential dangers of AI chatbots for young people, hoping to prevent other families from experiencing similar tragedies. This case may set a legal precedent for holding AI companies accountable for the content and interactions their platforms generate.

Ever had a bad day at work? We think this intern has had a worse one. ByteDance, the owner of TikTok, has dismissed an intern for "maliciously interfering" with the training of one of its AI models. However, the company downplayed claims about the extent of the damage, calling them "exaggerated and inaccurate." The issue gained traction after reports circulated on social media over the weekend. ByteDance clarified that the intern, who was part of the advertising technology team, had no involvement with its AI Lab, which develops the Doubao chatbot, China’s most popular AI model. In a statement, the company explained, "The individual was an intern with the [advertising] technology team and has no experience with the AI Lab." ByteDance also noted that inaccuracies were present in both the intern’s social media profile and some media reports. The company reassured that its core operations, including AI-based products like its large language models, were unaffected by the intern’s actions. ByteDance also dismissed reports that the incident caused over $10 million (£7.7 million) in damage to an AI training system using thousands of GPUs. After the incident, ByteDance terminated the intern in August and notified their university and industry organizations. The tech giant, known for popular apps like TikTok and Douyin, is recognized as a leader in algorithm development and has been making significant investments in AI. Alongside its chatbot Doubao, ByteDance has also created a text-to-video tool called Jimeng, further expanding its AI capabilities. Despite the disruption, the company's AI initiatives continue to advance without major setbacks.

Fancy a conversation with Elon Musk? It can be arranged. Sort of. Short of trying to break into the tech billionaire's house, the next best thing is available on TikTok.

Google has gone nuclear to meet the unprecedented energy demands of AI.  Google has signed an agreement with Kairos Power to use small nuclear reactors to generate the substantial energy required for its AI data centers. The first reactor is expected to be operational within this decade, with more planned by 2035. While no financial details or plant locations were disclosed, this move highlights a growing trend of tech companies turning to nuclear power to meet the vast electricity needs of AI operations. Michael Terrell, Google’s senior director for energy and climate, emphasized the need for new energy sources to support AI technologies, stating that this deal would help develop clean and reliable energy while unlocking AI's full potential. Kairos Power's executive, Jeff Olson, also noted that the partnership is crucial for demonstrating the commercial and technical viability of advanced nuclear energy in decarbonizing power grids. The project still requires approval from the U.S. Nuclear Regulatory Commission and local authorities. Kairos Power, based in California, received the first permit in 50 years to build a new type of reactor last year and has begun constructing a demonstration reactor in Tennessee. Nuclear power, which is nearly carbon-free and operates continuously, has become an attractive option for the tech industry as it seeks to reduce emissions while managing increasing energy consumption. According to Goldman Sachs, global energy use by data centers is expected to more than double by the end of the decade. At last year's UN Climate Change Conference, the U.S. joined a coalition aiming to triple nuclear energy capacity by 2050 as part of efforts to reduce reliance on fossil fuels. However, critics highlight the risks associated with nuclear power, including long-lasting radioactive waste. Microsoft recently reached an agreement to restart operations at the Three Mile Island nuclear plant, the site of a major nuclear accident in 1979, and Amazon announced plans to power a data center in Pennsylvania with nuclear energy.

$30,000 for a robotaxi? Investors aren’t convinced. Tesla shares dropped 8%, wiping out about $60bn in value, after the unveiling of its highly anticipated "robotaxi" underwhelmed investors. Shares fell to $219.40 after Elon Musk showcased the driverless “Cybercab,” expected to be built by 2026 for under $30,000. Musk also claimed it could reshape cities by turning car parks into parks. Despite these bold claims, analysts were disappointed by the event’s lack of detail and less than concrete timeline for Tesla’s upcoming projects. Some investors were expecting a teaser for a lower-priced vehicle, but none was provided. Analysts, including Tom Narayan from RBC and Garrett Nelson from CFRA, noted the event focused more on branding than specifics, leaving many questions unanswered. Social media news is great, but if you need to be updated on important AI stories as they happen in greater depth, subscribe to The Deepview for free. In depth journalism, daily. https://lnkd.in/ed25x3Mj

Do you think you’ll see AI win an Oscar, a Grammy or a Pulitzer Prize in your lifetime? No? Well it won a Nobel Prize this week. In a historic week for AI and science, two Nobel Prizes were awarded to researchers pioneering breakthroughs in artificial intelligence and biology. British computer scientist Demis Hassabis and his colleague John Jumper, both instrumental in the creation of Google DeepMind, won the Nobel Prize for Chemistry. They were recognized for their revolutionary AI tool, AlphaFold2, which predicts the structures of proteins, a vital leap for medicine. Their work has provided insight into antibiotic resistance and plastic decomposition, impacting global health. U.S. Professor David Baker also shared the prize for his development of new proteins. Meanwhile, Geoffrey Hinton, often dubbed the "Godfather of AI," and John Hopfield received the Nobel Prize in Physics for their foundational work in machine learning. Hinton’s research on neural networks underpins modern AI systems like ChatGPT, while Hopfield’s network theory advances pattern recognition. Despite the promise, Hinton has raised concerns about AI’s rapid progress, warning it could outsmart humans within decades. Both prizes underscore AI’s profound impact on science and society, with the technology already instrumental in the most important scientific breakthroughs, an idea that would’ve seemed laughable only five years ago. With the most coveted scientific awards on the planet already going to the people responsible for building the most intelligent AI, it’s hard not to wonder where the future may be headed, with AI technology still in its infancy. Social media news is great, but if you need to be updated on important AI stories as they happen in greater depth, subscribe to The Deepview for free. In depth journalism, daily. https://lnkd.in/ed25x3Mj