🌞 Australia's Green Hydrogen Boom, 💼 Point72's AI Fund, and 💎 Claude 3.5 Sonnet's Debut
Thomas's Innovation Wrap #83
Greetings,
Here’s your weekly wrap of technology, innovation, and finance news.
🌞 Energy
Australia's green hydrogen ambitions are gaining momentum, with over 100 projects worth US$127 billion announced since 2019. A recent South China Morning Post article provides an insightful overview of this burgeoning industry, highlighting projects like the US$10 billion Green Springs initiative. These developments showcase Australia's potential to leverage its vast land and renewable resources for green hydrogen production.
If the project, spread over 200 square kilometres, gets the go ahead and is successfully financed, the solar farms will power 2,150 modules of “electrolysers”, each with a capacity of 5 megawatts (MW). They will use the abundant power generated to split water molecules into hydrogen and oxygen.
This aligns with a broader global trend in solar power growth. The Economist writes that on current trends, solar power promises to become the largest source of energy on the planet by the 2040s, with electricity costs less than half as expensive as the cheapest available today.
To call solar power’s rise exponential is not hyperbole, but a statement of fact. Installed solar capacity doubles roughly every three years, and so grows ten-fold each decade. Such sustained growth is seldom seen in anything that matters. That makes it hard for people to get their heads round what is going on.
Meanwhile, the rapid growth of AI and data centres is creating new energy challenges. Bloomberg reports that their growing energy demands are outpacing available power supplies and threatening to disrupt energy transition plans.
Globally, there are more than 7,000 data centers built or in various stages of development, up from 3,600 in 2015. These data centers have the capacity to consume a combined 508 terawatt hours of electricity per year if they were to run constantly. That’s greater than the total annual electricity production for Italy or Australia.
💎 Artificial Intelligence
Anthropic released their latest LLM, Claude 3.5 Sonnet, claiming it outperforms competitor models while being twice as fast and 80% cheaper than their previous version. After using it extensively over the weekend, I've found it to be a significant improvement over past models, particularly for coding tasks.
Interacting with LLMs is a bit of an art. Renowned data scientist and co-founder of Fast.ai, Jeremy Howard, suggests that adding a specific final sentence to prompts can greatly improve the quality of responses.
A Slack survey finds that 81% of AI tool users say it boosts their productivity and work quality. What are the other 19% doing?
Meanwhile, researchers have developed a clever "truth police" method using a second AI to detect hallucinations in language models. The system measures "semantic entropy" to catch AI's BS.
In the world of sports, AI is creating new viewing experiences. Formula 1 and Amazon are debuting an AI "Statbot" that offers viewers predictions and context during live races.
The statbot will trawl race archives and parse torrents of real-time racing data to feed context and trivia to broadcast presenters live during the Barcelona race, using technology from the Seattle-based company’s Amazon Web Services cloud computing division, said Neil Ralph, the tech company’s lead on technical collaboration with F1.
Google's DeepMind has developed "video-to-audio" technology that can generate soundtracks and dialogue to match AI-generated or existing silent videos (see the DeepMind site for demos). ElevenLabs has also launched a new Text to Sound Effects API.
💊 Health
Insilico Medicine used AI to develop a drug for a rare lung disease in just 2.5 years at one-tenth the usual cost. This breakthrough leverages AI's ability to decode complex biological and chemical "languages," potentially revolutionizing drug discovery.
Traditionally, this process would have taken six years and cost over $400 million. With generative AI, Insilico reduced the cost to one-tenth and the time to two and a half years.
MIT researchers have developed SPARROW, an AI framework that helps scientists identify promising molecules for new drugs.
"The selection of compounds is very much an art at the moment—and at times it is a very successful art. But because we have all these other models and predictive tools that give us information on how molecules might perform and how they might be synthesized, we can and should be using that information to guide the decisions we make," says Connor Coley, MIT assistant professor and senior author of the SPARROW paper.
AI is also accelerating drug candidate screening. Australian researchers developed PSICHIC, an AI tool that predicts protein-molecule interactions using only sequence data. Meanwhile, Terray Therapeutics' automated lab generates 50 terabytes of data daily, feeding AI models that rapidly identify promising candidates.
Beyond drug discovery, an international team used machine learning to identify common molecular biomarkers for transplant rejection across all major organs. This could enable doctors to predict and prevent organ rejection more effectively.
An interesting study has provided insights into why some people seem immune to Covid-19, even after direct exposure. The results suggest high activity in a particular immune system gene helped prevent infection in certain individuals.
Scientists have developed nanocircle-based CRISPR tests that promise to combine the speed of rapid COVID tests with the accuracy of PCR technology. These tests can detect various pathogens within 15 minutes, potentially superseding PCR in many applications.
Organs-on-chips are emerging as a promising alternative to animal testing for drug development. These microfluidic devices combine human cells with a dynamic microenvironment that mimics specific tissues and organs, potentially offering more relevant data for human treatments.
⚙️ Mobility
Recent studies suggest self-driving cars may be safer than human drivers in most situations, but still struggle with tricky conditions like dim lighting or turning. Meanwhile, Canadian startup Waabi Innovation has raised $200 million to deploy driverless trucks by 2025, using a novel digital simulator approach to train its AI system. This method could accelerate the development of autonomous vehicles while reducing costs.
Ferrari is investing €210 million in a new solar-powered factory in Maranello, Italy to produce its first all-electric supercar alongside hybrid and combustion engine vehicles. This marks a significant step in the luxury automaker's transition towards electrification, with the e-building featuring over 3,000 solar panels generating 1.3 MW of renewable energy.
Interest in Range Rover's first electric SUV is surging, with the waitlist doubling to over 38,000 in just four months. The company claims its EV can drive through 850mm of water, outperforming competitors like the GMC Hummer EV and Toyota Land Cruiser.
High-speed rail projects are gaining momentum globally. Spain's Ineco has been selected to analyse the feasibility of Australia's first high-speed rail line between Sydney and Newcastle, while India's ambitious bullet train project aims to connect Mumbai and Ahmedabad in just two hours, covering a 508-kilometer route.
Archer Aviation plans to launch an air taxi network in the San Francisco Bay Area, using eVTOL aircraft to connect five key locations. These air taxis could replace 1-2 hour drives with 10-20 minute flights, potentially revolutionizing urban transportation. The company has also signed a deal with Signature Aviation to identify opportunities for launching similar services across the US.
💻 Chips and Computing
The energy demands of AI computing are skyrocketing. Elon Musk claims Tesla's new AI supercluster will consume over 500 MW of power, while Nvidia's Blackwell GPUs consume around 1,500 watts per chip. These developments underscore the growing challenge of balancing computational power with energy efficiency in AI systems.
Cerebras is partnering with Dell to challenge Nvidia's dominance in the AI market. Their massive single-chip AI accelerator claims to offer 880 times the memory capacity of GPUs, potentially reducing the code required to build large language models by up to 97%.
🛰️ Space
Innospace, a South Korean rocket startup dubbed 'Korea's SpaceX', is preparing for its first commercial launch in March next year. The company is confident of success, having already completed a low Earth orbit entry test launch in Brazil last year. With plans for seven launches annually after that, Innospace aims to go public on the Kosdaq in July.
Innospace has the country's only hybrid rocket technology and proved its unrivaled technological competitiveness by succeeding in the launch of the self-developed Hanbit-TLV in March last year, a first for a domestic private firm."
SpaceX has launched Starlink Mini, a compact version of its satellite internet antenna designed for portability. The Mini is more expensive than the standard dish for now, but Musk says it will eventually cost about half as much to buy and for the monthly subscription. It offers download speeds over 100 Mbps with 23ms latency, fits in a backpack, has a built-in Wi-Fi router, and can be set up in less than five minutes. This development could significantly expand Starlink's reach and usability in various scenarios.
🤖 Robotics and Drones
Engineers at The University of Manchester have designed a small, spring-driven robot that could theoretically jump more than 120 meters in the air – twice the height of Big Ben's tower. This breakthrough could lead to robots capable of incredible feats, especially in low-gravity environments.
A new open-source robot model called OpenVLA, trained on vast real-world robot data, outperforms larger models and adapts easily to new tasks. This could accelerate the development of more versatile robots across various industries.
"Notably, most prior works achieve strong performance only in either narrow single-instruction or diverse multi-instruction tasks, resulting in widely varying success rates," the researchers write. "OpenVLA is the only approach that achieves at least 50% success rate across all tested tasks, suggesting that it can be a strong default option for imitation learning tasks, particularly if they involve a diverse set of language instructions."
AutoStore has opened a new robot assembly factory in Thailand, potentially accelerating the adoption of automated storage and retrieval systems in Australian warehouses.
The US Navy is deploying AI-equipped underwater drones to scan the ocean floor for mines and other threats, cutting mission times in half. They're now expanding this technology to detect enemy ships and aircraft.
The Australian Army has tested the world's most powerful portable laser weapon, capable of taking down drones with incredible precision at long range. The Fractl laser, built by AIM Defence, is smaller and cheaper than similar systems.
"You push a button to track the drone and the computer takes over, then you push another button to 'pull the trigger' just like a video game," said Corporal Patrick Flanagan. "With your index finger you can quickly change your aim between the drone's video camera, center mass or one of the propellers. It only takes seconds to knock out the camera and two or three seconds to disable the rotor."
💲 Finance
Steve Cohen's Point72 Asset Management is seeking to raise around $1 billion for a new hedge fund that will focus on investing in companies involved with artificial intelligence. The fund, which is expected to launch in late 2023 or early 2024, will be Point72's first new hedge fund in decades and reflects Cohen's belief in the transformative potential of AI.
"There's going to be big winners and big losers," he said in an April CNBC interview, adding that while it may take years to play out, "when you have technological change like this, it sort of reminds me of the '90s where the best new companies came out of that period."
Investment managers are excited about the potential for GenAI to generate alpha returns, but data availability and cultural change remain key challenges to adoption according to a recent Citigroup report.
For example, many of the executives surveyed highlighted that changing the company culture is a critical challenge in increasing the adoption of AI tools and solutions across their organisations. When a firm already has a successful portfolio management team, incorporating an AI tool takes a lot of convincing. Employees must also be “upskilled” or trained to use the technology.
📚 Regulation
Apple is facing EU scrutiny under the new Digital Markets Act, with concerns about app store practices and payment mechanisms.
"[Apple] are very important because a lot of good business happens through the App Store, happens through payment mechanisms, so of course, even though you know I can say this is not what was expected of such a company, of course we will enforce exactly with the same top priority as with any other business."
Adobe is in trouble with the US government for making it difficult to cancel. I hate it when companies do that.
"Adobe trapped customers into year-long subscriptions through hidden early termination fees and numerous cancellation hurdles," Samuel Levine, director of the FTC's Bureau of Consumer Protection, said in a statement. "Americans are tired of companies hiding the ball during subscription signup and then putting up roadblocks when they try to cancel."
⚡ Other Snippets
India's electronics manufacturing industry is surging, becoming the world's second-largest mobile phone maker. The government is luring foreign manufacturers with incentives to compete with China.
In the 12 months to the end of March, its electronic exports reached $29bn, up by 24%, year on year. Still, that is a far cry from the almost $900bn of electronics China exported last year.
McDonald's has concluded its two-year trial of AI-powered drive-thru ordering, opting not to expand the IBM system beyond the initial 100+ test restaurants. The company remains optimistic about AI voice ordering and aims to find a new partner by year-end.
"While there have been successes to date, we feel there is an opportunity to explore voice ordering solutions more broadly," Mason Smoot, chief restaurant officer for McDonald's USA, said in the system message.
Dell's new policy requiring 3 days in-office or forfeit promotions has backfired, with nearly half the workforce choosing to stay remote and many reportedly job hunting elsewhere.
Meta is developing a high-end VR headset, codenamed La Jolla, featuring photorealistic avatars and higher-resolution screens. This push aims to enhance VR's potential for remote work beyond current capabilities.
Photographer Miles Astray submitted a real flamingo photo to an AI photography competition and… won! Before being disqualified.
I was hoping that if it won, it would show that nature still outdoes the machine in terms of creativity and beauty and fascination. That was one part of the message. And then the other part was more related to the fact that, right now, we can hardly distinguish between real photos – and even audio and videos – and AI-generated content. And that has its dangers and pitfalls.
China plans to build a $5.2 billion particle collider by 2027, aiming to study the Higgs boson in unprecedented detail. This ambitious project could beat Europe's proposed mega-collider to the punch, potentially advancing our understanding of fundamental physics years earlier.
Scientific American reports the discovery of the oldest deep-sea shipwreck, a 3,200-year-old "time capsule" from the Bronze Age, found 60 miles off Israel's coast. This unique find, containing dozens of amphorae, offers rare insight into Mediterranean trade during the mythical era of the Minotaur. The wreck provides a tangible link to a pivotal period when the ancient world was becoming increasingly interconnected through commerce and diplomacy.
Know someone who'd enjoy the Innovation Wrap? Feel free to share it with them!
Have a great week,
Thomas
About Thomas Rice
Thomas Rice co-founded Minotaur Capital, a technology-driven, AI-led global equities fund, and is based in Sydney, Australia. He can be found on the platform-formerly-known-as-Twitter at @thomasrice_au.