State of AI Report 2024 - Sync #489
Plus: The New York Times warns Perplexity; have we reached peak human lifespan; tech giants tap nuclear power for AI; OpenAI projects billions in losses while Nvidia's stock reaches a new high
Hello and welcome to Sync #489!
State of AI Report 2024, one of the best resources capturing the current state of AI across various fields—from research to industry, politics, and safety—has been released. In this issue of Sync, I’ll highlight three developments that caught my attention.
Beyond that, plenty has happened in the world of technology over the past week.
In AI, The New York Times warns Perplexity not to use their content. Meanwhile, Google and Amazon are turning to nuclear power for their AI datacentres, OpenAI’s internal financial report projects a loss of $14 billion by 2026, and Nvidia’s stock reaches a new high.
In robotics, Boston Dynamics has partnered with the Toyota Research Institute to develop AI for humanoid robots, and a three-armed robot conductor has made its debut in Dresden. We’ll also take a closer look at Tesla’s confusing "We, Robot" event.
Additionally, a recent study suggests we are approaching the limits of human lifespan, and lab-grown meat could be sold in the UK within the next few years.
Enjoy!
State of AI Report 2024
One of the best resources to learn where the AI research and industry currently stand is the State of AI Report, produced by Nathan Benaich and Air Street Capital. The seventh edition of the report has recently been published and is filled with insightful information about developments in AI research, industry, politics, and safety.
The report itself consists of over 200 slides packed with detailed information about the state of AI in 2024, and is therefore too extensive to summarise in a short article. Instead, I will highlight three developments that particularly caught my attention—the narrowing gap between top models, the move from models to products, and the vibe shift from safety to acceleration.
I still recommend at least skimming the report. It will provide you with a good overview of the AI industry. Perhaps other developments highlighted in the report will catch your attention. If so, please let me know in the comments what stood out to you.
The gap on top is narrowing
In 2023, OpenAI and GPT-4 were the unchallenged leaders in the new AI boom, maintaining the top position throughout the year. However, in 2024, the gap (which some liked to call a chasm) between OpenAI’s models and others began to close. Models like Anthropic’s Claude 3.5 Sonnet and Google Gemini 1.5 have caught up with OpenAI’s models, sometimes even surpassing them.
At the same time, open models began to put pressure on proprietary models. In April, Meta released the Llama 3 family, a major milestone for open models. The largest model in the Llama family, Llama 3.1 405B, came very close to top models such as GPT-4 and Claude 3.5 Sonnet across multiple benchmarks.
With the differences in performance between models becoming increasingly small, developers had to look elsewhere to gain an edge over competitors. The main frontier now is research on planning and reasoning, which combines large language models with reinforcement learning, evolutionary algorithms, and self-improvement methods. The first model to incorporate these new developments is OpenAI’s o1, which achieved some impressive results. However, as the authors of the report ask, how long will o1 top the charts? The competition is fierce, and OpenAI no longer has the first-mover advantage. It’s fair to assume that Anthropic and Google will, sooner rather than later, release their own models capable of planning and reasoning.
One interesting observation from the report worth highlighting is how much the cost of large language models has dropped over the last year. These models are now cheaper and more accessible, which opens up new possibilities and applications across various apps and industries, as well as in fields such as mathematics, biology, genomics, the physical sciences, and neuroscience.
A new frontier for large language models worth mentioning is robotics, where these models are helping to drive new capabilities.
One factor improving accessibility to this technology is the rise of small language models. Models like Google Gemma, Microsoft Phi, models from Mistral AI, and smaller Llama models open up the possibility of running language models directly on smaller devices, without needing to rely on cloud-based models.
The move from models to products
If 2023 was the year generative AI exploded onto the scene, enabling new possibilities and spawning a new cohort of AI startups that attracted billions of dollars in investments, then 2024 is the year these new companies and ideas have been put to the test.
According to the report, close to $100 billion has been invested in AI companies. The graphs included in the report show that the majority of that money was invested in US companies. In total, generative AI companies received $35 billion in 2024.
However, it is worth noting that these numbers are inflated by a handful of mega investment rounds, such as recent $6 billion fundraises by xAI and OpenAI, or Anthropic’s $4 billion mega round at the beginning of the year. It would be interesting to see how that $100 billion is distributed in detail.
The report states that AI companies have reached a combined value of $9 trillion.
All that money invested in AI companies is starting to generate revenue, at least for some of them. More established companies are beginning to report billions of dollars in revenue, while startups, particularly those in audio and video generation, are starting to gain attention.
A major takeaway from the State of AI Report 2024 is that the industry is transitioning from just creating models to building actual products around them. In 2023, simply having a large language model could secure substantial funding (as seen with Mistral or Inflection AI), but in 2024, that’s no longer sufficient. Models are now much cheaper and more accessible than a year ago, and the focus has shifted to what useful products and services can be developed with them. At the same time, as the report notes, questions around pricing and long-term sustainability remain unanswered.
A new trend that emerged over the past year is the rise of pseudo-acquisitions in the AI space. Companies unable to remain at the cutting edge are not being absorbed in the traditional way, where the larger company acquires the smaller competitor. Instead, larger players are hiring top talent and leadership from struggling companies and licensing their technology, leaving behind only the shell of the original firms. Examples of this pseudo-acquisition can be seen with Inflection AI, Adept, and Character.AI.
The report also gives a special mention to Nvidia. If OpenAI is the poster child of the current AI boom on the software side, then Nvidia is the poster child on the hardware side. The chip manufacturer has benefited enormously from the high demand for AI hardware, propelling the company into the elite club of those valued at more than $1 trillion. At the time of writing, Nvidia is worth $3.4 trillion.
The vibe-shift from safety to acceleration
The last point I’d like to highlight from the State of AI Report 2024 is what the authors referred to as “the vibe shift from safety to acceleration.” Alongside the rise of OpenAI and large language models in 2023, there was a parallel conversation about potential dangers and questions surrounding AI safety. Panels and summits dedicated to these topics were held worldwide, with notable figures in the industry urging that AI safety be taken seriously. However, the existential risk discourse has cooled, especially after the failed coup at OpenAI. Some of those who once warned about the dangers of AI are now fully on board with accelerating its development.
This does not mean AI safety is no longer being studied—it’s just no longer a priority. Researchers continue to explore potential model vulnerabilities and misuse, proposing fixes and safeguards. Meanwhile, governments worldwide are following the UK's lead by building their own state capacity around AI safety, launching institutes, and examining critical national infrastructure for potential vulnerabilities.
If you enjoy this post, please click the ❤️ button or share it.
Do you like my work? Consider becoming a paying subscriber to support it
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.
🦾 More than a human
Have We Reached Peak Human Life Span?
According to a study published in Nature Aging, human life expectancy may be nearing its upper limit, with an estimated cap of around 87 years (approximately 90 for women and 84 for men). The study examined life expectancy data from 1990 to 2019 in regions known for longevity and found that while life expectancy has generally increased, the rate of this growth has slowed almost everywhere except Hong Kong. Researchers suggest that further significant extensions in lifespan will be challenging without breakthroughs that specifically target the ageing process itself, as current medical advances may no longer be sufficient. Additionally, the ages at death have become compressed into a narrower range, indicating a possible natural limit to the human lifespan. Despite this, proponents of anti-ageing research remain optimistic that future innovations could significantly extend human life beyond current expectations.
What is wearable neurotech and why might we need it?
This article explores the growing field of wearable neurotechnology, which aims to treat mental health conditions such as depression, anxiety, and insomnia through non-invasive brain stimulation, offering an alternative to traditional medications. The article highlights the story of Alex, who is using the Flow device, which employs transcranial direct current stimulation (tDCS) to treat depression, as an example of a successful application of such devices. Despite its promising potential, wearable neurotech faces significant regulatory and commercialisation challenges, including complex approval processes and securing healthcare reimbursement. Meanwhile, a parallel consumer neurotech market is emerging, focusing on wellness rather than medical applications. These products target athletes, professionals, and consumers interested in enhancing mental performance.
🧠 Artificial Intelligence
Google to buy nuclear power for AI datacentres in ‘world first’ deal
Google has signed a deal to buy energy from a fleet of small modular nuclear reactors (SMRs) built by California’s Kairos Power. The first SMR is expected by 2030, with the rest by 2035. Nuclear energy offers Google a "clean, round-the-clock power source" to reliably meet its growing electricity needs, driven by the rise of AI and cloud storage. Google’s deal involves purchasing 500 megawatts of power from Kairos Power, whose demonstration reactor is expected to be completed by 2027. The agreement between Google and Kairos Power is still subject to regulatory approval.
Amazon.com joins push for nuclear power to meet data center demand
After Microsoft and Google, Amazon is the next company tapping nuclear power to meet growing electricity demand for its data centres. The retail giant has signed three agreements with X-Energy to develop small modular reactors (SMRs) near an Energy Northwest site in Washington. Amazon will have the right to purchase power from four SMR modules, with a potential total capacity of up to 960 MW if Energy Northwest adds eight more modules. Amazon is leading a $500 million funding round for X-Energy’s SMR development, aiming to bring over 5 GW of SMR capacity online in the US by 2039. Additionally, Amazon has signed an agreement with Dominion Energy for a 300 MW SMR project in Virginia.
The New York Times warns AI search engine Perplexity to stop using its content
The New York Times sent a cease and desist letter to Perplexity, demanding that the AI search engine startup stop using its content in its search results. Perplexity responded with a statement saying they are not scraping data to train their models but are instead indexing web pages, and added that “the law recognises that no one organisation owns the copyright over facts.” Perplexity is not the first AI company to anger The New York Times. In December 2023, the newspaper sued OpenAI and Microsoft over claims its copyright was infringed to train ChatGPT, which could cost both tech companies billions of dollars in statutory and actual damages for allegedly copying its works if the lawsuit proceeds and the judge sides with the newspaper.
OpenAI Projections Imply Losses Tripling To $14 Billion In 2026
The Information reveals internal details about OpenAI’s financial projections, which are interesting, to say the least. According to the report, OpenAI projects its losses to reach $14 billion—three times more than the current figure—by 2026, with total losses between 2023 and 2028 expected to be $44 billion. The company is projected to remain unprofitable until at least 2029, even after reaching the $100 billion revenue mark by that time.
Nvidia, the AI chipmaker, just hit another record high
Nvidia shares hit a record high on Monday, closing at $138.07 as Wall Street anticipates earnings updates from Microsoft, Meta, Google, and Amazon on their AI infrastructure spending. This made Nvidia the second most valuable company in the US, worth $3.4 trillion, with only Apple being worth more.
Silicon Valley is debating if AI weapons should be allowed to decide to kill
There is a debate over whether AI should make the final decision to kill. Shield AI’s co-founder Brandon Tseng opposes the idea, while Anduril’s co-founder Palmer Luckey and investor Joe Lonsdale are open to the use of autonomous robots on the battlefield. Proponents of AI weapons argue they are inevitable and that the US must develop them to avoid falling behind adversaries such as China and Russia. Opponents call for a ban on autonomous weapons and raise concerns about the consequences of machines independently deciding to use lethal force. Lonsdale emphasised that the debate over fully autonomous weapons should not be viewed as a simple "yes or no" question, but rather as a nuanced issue where degrees of autonomy and human control must be carefully considered based on the situation and potential threats.
We’re Entering Uncharted Territory for Math
In this interview, Terence Tao, the world’s greatest living mathematician, shares his views on AI’s potential to transform how mathematics is done. He described a kind of AI-enabled, “industrial-scale mathematics” that has never been possible before—where AI, at least in the near future, acts less as a creative collaborator and more as a tool that enhances mathematicians’ hypotheses and approaches. “I think at the frontier, we will always need humans and AI. They have complementary strengths. AI is very good at converting billions of pieces of data into one good answer. Humans are good at taking 10 observations and making really inspired guesses,” said Terence Tao.
Meta at OCP Summit 2024: The open future of networking hardware for AI
Meta is fully embracing an open approach to AI by not only making its Llama models available for free use but also open-sourcing its networking hardware for AI. At this year’s Open Compute Project Summit (OCP), Meta unveiled its next-generation network fabric for AI, along with a new portfolio of network hardware that the company uses in its AI training clusters and in its AI datacenters, which were contributed to OCP to benefit companies of all sizes across the industry.
If you're enjoying the insights and perspectives shared in the Humanity Redefined newsletter, why not spread the word?
🤖 Robotics
Tesla Optimus Bots Were Remotely Operated at Cybercab Event
During Tesla’s “We, Robot” event last week, the attendees got a chance to be served by Tesla’s humanoid robot, Optimus. However, it was later revealed that these robots were remotely controlled by humans. Some critics are calling out Tesla and Elon Musk for not disclosing this upfront, arguing it could create a mismatch between public expectations and the actual capabilities of Tesla's robots. Others have shared videos where the robotic bartender acknowledged that it was being “assisted by a human” when asked.
▶️ The Tesla Robotaxi is Confusing... (19:13)
Marques Brownlee shared his experience from attending Tesla’s “We, Robot” event, where he had the chance to ride Tesla’s self-driving taxi, Cybercab, see the Robovan, and interact with Tesla’s humanoid robots, Optimus. The word Marques used to describe the event was “confusing.” In his view, Tesla’s event showed a concept of a certain type of high-tech future but the company failed to mention any numbers or specs of the presented products. Moreover, the staged nature of the event and the uncertainty about what was real versus scripted became the main talking point, overshadowing the products and the future they were meant to represent.
Agility Robotics, Maker of Humanoid Bots, Shows Off Its ‘RoboFab’
Bloomberg got a chance to visit Agility Robotics’ RoboFab—the world’s first humanoid robot factory. The company moved into the facility earlier this year and plans to produce its first humanoid robots sometime this month. Ultimately, the factory will have the capacity to produce 10,000 robots a year. Agility Robotics’ robots are already in use at a Spanx warehouse operated by GXO Logistics and have been tested by Amazon since October last year.
Boston Dynamics teams with TRI to bring AI smarts to Atlas humanoid robot
Boston Dynamics is teaming up with Toyota Research Institute (TRI) to collaborate on building AI models for Atlas, Boston Dynamics' humanoid robot. According to a statement issued by TRI, the partnership will leverage the strengths of both companies—Boston Dynamics' world-class hardware platform and TRI’s expertise in AI and developing Large Behavior Models for robotics. “This partnership is an example of two companies with a strong research-and-development foundation coming together to work on many complex challenges and build useful robots that solve real-world problems,” says Boston Dynamics CEO Robert Playter.
This three-person robotics startup is working with designer Yves Béhar to bring humanoids home
Many well-known tech companies had their beginnings in a garage somewhere in Silicon Valley. Christoph Kohstall, a former Google Brain robotics developer, continues this tradition, but instead of building computers, he’s building humanoid robots. Kohstall’s passion for robotics led to the creation of Kind Humanoid, a three-person startup aiming to create humanoid robots for home caretaking, with a focus on mobility and functionality in cluttered home environments. The team recently partnered with renowned designer Yves Béhar, who is helping to refine the design of Mona, their humanoid robot. The startup plans to build an initial batch of a dozen Mona robots for field tests in early 2025.
Three-armed robot conductor makes debut in Dresden
The Dresden Sinfoniker, known for its innovative and politically charged performances, celebrated its 25th anniversary with a "Robotersinfonie" concert. The first half featured human conductors, while the second half was led by a robot named MAiRA Pro S. The robot, equipped with three arms, each holding a different-coloured baton, was designed to conduct complex, multi-rhythmic music that would be too intricate for human conductors. MAiRA is intended not to replace human conductors but to handle particularly demanding pieces. Although one of the most advanced, MAiRA is not the first robot conductor; previous examples include a robot conducting the Detroit Symphony Orchestra in 2008 and Andrea Bocelli performing with the robot YuMi in 2017. The world premiere of MAiRA’s performance was well-received, with a livestreamed concert following the debut.
▶️ This Robot is Building Supertall Elevators (10:00)
The B1M visits a skyscraper construction site in Vienna, Austria, to meet Ella, a construction robot that assists with installing elevators. This innovative robot autonomously climbs the elevator shaft, installing thousands of anchor bolts to which human workers then attach rails and the rest of the elevator system. By taking over the noisy and repetitive task of installing anchor bolts, Ella makes the construction process safer and more efficient.
🧬 Biotechnology
Lab-grown meat could be sold in UK in next few years, says food regulator
UK consumers may be able to taste lab-grown steak, beef, chicken, and foie gras within a few years. The UK Food Standards Agency has received £1.6 million in government funding to develop an efficient safety assessment process for cultivated meat to ensure it is safe for human consumption. The UK recently approved cultivated meat for use in pet food, becoming the first European country to do so. Cultivated meat is already approved for sale in Singapore (2020), the US (2023), and Israel (2024). Meanwhile, Italy and two US states have banned the sale of cultivated meat.
💡Tangents
Hell Freezes Over as AMD and Intel Come Together for x86
Intel and AMD, two fierce competitors, have announced a collaboration at the OCP Summit 2024 to create a unified x86 advisory board aimed at standardising the x86 instruction set architecture. The goal of this cooperation is to improve compatibility and consistency within the x86 architecture, which is facing significant challenges from alternative architectures like Arm and RISC-V. Key beneficiaries will include sectors such as data centres, AI, cloud, and embedded devices, which will benefit from streamlined development processes and simplified software compatibility. Despite this collaboration, Intel and AMD will continue to compete in the broader market. Dr Ian Cutress of TechTechPotato also discussed this news, giving a more technical explanation of what it could mean for both companies as well as for developers.
Thanks for reading. If you enjoyed this post, please click the ❤️ button or share it.
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!"