🤖 1000-Agent Simulation, ⚙️ Toyota's Drifting AI, and 🔋 Bloom's Data Power
Thomas's Innovation Wrap #104
Greetings,
Here’s your weekly wrap of technology, innovation, and finance news.
💎 Artificial Intelligence
From the researchers that brought you "Generative Agents: Interactive Simulacra of Human Behavior" comes the thrilling sequel: "Generative Agent Simulations of 1,000 People". This new study creates custom AI agents for 1,000 participants based on in-depth interviews, replicating their attitudes and behaviours across various social science measures. The AI agents predicted participants' responses to surveys and experiments with high accuracy, often matching the participants' own consistency over time. This approach could enable more realistic simulations for social science research and policy testing.
The generative agents replicate participants' responses on the General Social Survey 85% as accurately as participants replicate their own answers two weeks later, and perform comparably in predicting personality traits and outcomes in experimental replications.
AI is advancing genomics research with genomic foundation models like Evo, which can create entire genomes and design new proteins from scratch. Trained on 300 billion DNA nucleotides, Evo has demonstrated the ability to design new versions of CRISPR genome editors that perform as well as commercial versions. This development could accelerate drug discovery by rapidly generating and testing new protein variants, potentially reducing the time and cost of bringing new medicines to market.
"This work is extremely significant," says computational biologist Arvind Ramanathan of Argonne National Laboratory, who wasn't connected to the study. The tests the authors put Evo through, he says, provide "a great showcase of applications" for the AI.
In chemistry labs, AI-driven robots are autonomously performing and analysing chemical reactions. Human intervention is still needed for complex cases.
‘The workflow not only automates synthesis and analysis but also uses AI to make decisions on what steps to do next based on the analytical data it collects. It’s like a grad student that doesn’t need sleep!’ comments David Leigh, a chemist at the University of Manchester, who was not involved in the work.
Microsoft is launching specialised AI models for manufacturing, agriculture, and financial services, partnering with industry giants like Siemens, Bayer, and Rockwell Automation. These purpose-built models, available through Microsoft's Azure AI catalog, address specific challenges in traditional industries. For example, Siemens' NX X copilot uses natural language processing for industrial design tasks, while Bayer's E.L.Y. Crop Protection model helps farmers make informed decisions about pesticide application.
"Microsoft is in a unique position to deliver the industry-specific solutions organizations need through the combination of the Microsoft Cloud, our industry expertise, and our global partner ecosystem," Satish Thomas, Corporate Vice President of Business & Industry Solutions at Microsoft, said in a LinkedIn post announcing the new AI models.
SF-based startup Puzzle claims its AI can automate 90% of accounting tasks.
"What we're launching now is effectively taking the general ledger, the backbone of accounting, and bringing complicated accounting logic from spreadsheets into the core accounting software," Orloff said.
Andreessen Horowitz partners argue that intelligent automation, powered by AI and large language models, is set to replace traditional Robotic Process Automation (RPA).
With LLMs, however, we believe the original vision of RPA is now possible. Instead of hard-coding each deterministic step in a process, AI agents will instead be prompted with an end goal (e.g., book an appointment for the customer, transfer data from this document into this database), and then be empowered with the right tooling and context to take those actions on behalf of the company.
In a study of 1,600 participants, readers consistently failed to distinguish between AI-generated poems and classic works by Shakespeare, Dickinson, and other famous poets. The AI-generated poems were not only indistinguishable but often preferred.
"Over 78 per cent of our participants gave higher ratings on average to AI-generated poems than to human-written poems by famous poets," says Brian Porter at the University of Pittsburgh in Pennsylvania.
While poets grapple with AI-generated competition, digital artists and illustrators are taking a different approach. They're fighting back against AI scraping with tools like Glaze and Nightshade. These innovative programs add imperceptible alterations to images, either cloaking the artist's style or potentially "poisoning" AI training data, making it harder for AI models to replicate their work without consent.
"I would love a tool that if someone wrote my name and made a prompt, like, garbage came out," responded Karla Ortiz, a prominent digital artist. "Just, like, bananas or some weird stuff."
Osmo has developed AI-powered 'smell' technology to detect counterfeit products. Their portable device uses sensors and artificial intelligence to analyse the chemical composition of items, successfully identifying fake sneakers with over 95% accuracy.
"We run these sensors 24/7, constantly collecting data about the chemical makeup of everything from plums and peaches to manufactured products," Wiltschko explains.
In another area of fraud prevention, British mobile operator O2 has introduced 'Daisy', an AI bot designed to waste scammers' time. This innovative fraud prevention tool uses a combination of AI models to engage in lifelike conversations, keeping fraudsters on the line for up to 40 minutes. Daisy's ability to ramble about family stories and knitting has successfully frustrated scam callers, potentially diverting them from targeting real victims.
"Daisy" is claimed to be indistinguishable from a real person, fooling scammers into thinking they've found perfect prey thanks to its ability to engage in "human-like" rambling chat, the biz claims.
⚙️ Mobility
Waymo has opened its robotaxi service to the public in Los Angeles, covering an 80-square-mile area. The company now transports over 50,000 weekly passengers across its operations, showcasing the growing adoption of autonomous ride services.
Waymo’s robotaxis have driven more than 20 million fully autonomous miles and provided more than 2 million rides to passengers without encountering a serious accident that resulted in its operations being sidelined.
Chinese autonomous driving startup Pony AI is seeking up to $224 million in a US IPO.
Pony operates a fleet of 190 "robotrucks" in Beijing and Guangzhou and over 250 robotaxis in Beijing, Guangzhou, Shenzhen, and Shanghai. The company told TechCrunch it can charge for robotaxi fares in all four cities and is fully driverless in Beijing, Guangzhou, and Shenzhen.
Tire manufacturers are racing to create the perfect rubber for EVs, balancing reduced rolling resistance, durability, and minimal particulate pollution. While EVs reportedly cause 20 percent more tire wear than petrol vehicles due to their extra weight and torque, the specific requirements for EV tires might not be as different from conventional tires as one might think.
"There is basically no difference between a conventional tire and a special tire for electric vehicles," says Wanka. "It is more the case that certain vehicle models or classes have specific tire requirements." Better service life, rolling resistance and rolling noise have been particularly beneficial for electric vehicles. "But all vehicles—regardless of their drive type—benefit from the technological advances."
Toyota is teaching its driverless cars to drift like pros, sliding sideways at up to 63 degrees. Apart from letting the cars audition for the next Tokyo Drift movie, this tech could help vehicles recover from unexpected skids.
"When a vehicle is driving and unexpectedly hits a patch of ice on the road, we need our AI to control the vehicle at its limits to ensure safety," says Avinash Balachandran, who researches autonomous driving at TRI, but wasn't involved in this work.
💻 Chips and Computing
Amazon is developing custom AI processors to reduce dependence on Nvidia. AWS claims its Inferentia chips are 40% cheaper for generating AI responses, potentially offering significant cost savings for large-scale AI operations.
"We want to be absolutely the best place to run Nvidia," said Dave Brown, vice-president of compute and networking services at AWS. "But at the same time we think it's healthy to have an alternative."
SoftBank is partnering with Nvidia to build Japan's most powerful AI supercomputer using Nvidia's Blackwell AI chips. The collaboration aims to create a nationwide "AI grid" by transforming SoftBank's base stations into mini data centres, bringing AI capabilities closer to end-users.
"Our cell towers used to just carry bits for telecommunication and internet surfing," Son told the audience. "Now, with this network, we'll densely connect them with each other and make them into one big neural brain and the infrastructure of intelligence for Japan."
Nvidia's highly anticipated Blackwell chips are facing overheating issues requiring server rack redesigns. Despite these challenges, CEO Jensen Huang reports "insane" demand, with the chips reportedly sold out for the next year.
In line with this trend, Foxconn expects AI servers to dominate its server revenue by 2025, with AI-related products projected to account for over 50% of its server business.
💊 Health
Promising new treatments for myopia are emerging, offering hope for those affected by near-sightedness. Researchers are exploring novel approaches including low-level red light therapy, which has shown potential to slow and even reverse myopia progression. This treatment involves shining 650-nanometre laser light at 1600 lux onto the retina for about 3 minutes, twice daily. Other interventions being studied include atropine eye drops, special glasses and contact lenses designed to blur peripheral vision, and orthokeratology - rigid contact lenses worn at night to reshape the cornea.
The real villain seems to be too much time indoors. Spending time outside has long been known to be protective against developing myopia, for reasons that are still unknown, and “it is certainly the case that kids are outdoors less and indoors more than previous generations”, says Mutti.
In the weight loss arena, Novo Nordisk is developing CagriSema, a potent combo promising up to 25% weight loss. Competitors like Eli Lilly, Viking Therapeutics, and Amgen are also vying for the projected $100 billion market with drugs offering faster results and less frequent dosing.
"Anti-obesity medications are emerging at an astonishing pace," Dr. Christopher McGowan, a gastroenterologist and obesity medicine specialist, told Business Insider. "Forthcoming medications take the blockbuster GLP-1 medications and up the ante."
University of Copenhagen scientists have identified a new weight loss drug target that burns energy and lowers appetite without common side effects. By activating the neurokinin 2 receptor (NK2R), they've increased calorie-burning and reduced appetite in mice and non-human primates, offering hope for those with obesity and type 2 diabetes.
"While GLP-1-based therapies have revolutionized patient care for obesity and type 2 diabetes, safely harnessing energy expenditure and controlling appetite without nausea remain two Holy Grails in this field. By addressing these needs, we believe our discovery will propel current approaches to make more tolerable, effective treatments accessible to millions more individuals," says Associate Professor Zach Gerhart-Hines from the NNF Foundation Center for Basic Metabolic Research (CBMR) at the University of Copenhagen.
🌞 Energy
Bloom Energy, a fuel cell technology company, is working to provide power for AI-related data centres. As data centres shift towards AI applications, they require more power, often exceeding local grid capacity. Bloom's solid oxide fuel cells offer a solution, providing on-site power generation. The company recently made a deal with American Electric Power (AEP), a major US utility, to power high-power data centres. AEP plans to purchase up to 1,000 MW of Bloom's fuel cells, with an initial firm contract for 100 megawatts.
"This is easily Bloom's largest-ever customer order, and it is one of the largest fuel cell contracts we have ever come across, anywhere in the world," wrote Raymond James Pavel Molchanov, who calculates that it could be worth as much as three times Bloom's 2024 sales. Molchanov rates Bloom at Outperform. "Details are slim at this early stage, so it would be premature to change estimates, but the headline is certainly very positive."
Speaking of power solutions, Texas is set to receive a boost with a 1 gigawatt AI-powered virtual power plant (VPP), capable of powering 200,000 homes during peak demand. This innovative system will utilise Google Cloud's AI to optimise energy use across hundreds of thousands of smart devices, effectively managing energy consumption in Texan homes to improve grid resilience and efficiency.
In the nuclear sector, researchers have developed a smart component system for small nuclear reactors, combining 3D printing with AI to detect potential hazards in just two seconds.
"We tackled the challenges associated with traditional inspection methods through our AI convergence technology, which can greatly enhance the stable and efficient operation of next-generation small nuclear power plants."
Ofgem has approved five new subsea power cable projects, connecting the UK to power grids in Europe and offshore windfarms. These underwater energy highways aim to transform Britain into a net exporter of green electricity by the 2030s.
As we shift to a clean power system more reliant on intermittent wind and solar energy, these new connections will help harness the vast potential of the North Sea and play a key role in making our energy supply cheaper and less reliant on volatile foreign gas markets and associated price spikes,
🌱 Agritech
Scientists are using CRISPR gene editing to create sweeter large tomatoes without compromising size or yield. This development is noteworthy because it overcomes a long-standing challenge in tomato breeding where increased size typically resulted in lower sugar content. By precisely disabling sugar-limiting genes, researchers have increased glucose and fructose levels by up to 30% in larger varieties, improving both taste and commercial tomato production.
"We are working with some companies to develop some commercial varieties by knocking out these genes," says Zhang. "It is still at the beginning stages."
⚡ Other Snippets
In advertising, AI is enabling unprecedented personalisation. Known, a full-service agency, uses AI to optimise ad placement and content, creating tailored experiences for individual viewers.
In another example, the company’s AI chatbot, known as “The Big Letbotski,” dug through 80,000 active subreddits to identify the most relevant conversations about chicken sandwiches for its client, Shake Shack. After finding 30 of the most content-specific subreddits, it ran targeted ads promoting its Chicken Shack Sundays giveaway in April. That helped Shake Shack beat sales estimates by 31%.
Coinbase has launched its new Coin50 index, aiming to be the crypto equivalent of the S&P 500. The index offers investors a broader view of the market beyond Bitcoin-centric metrics.
"What we're trying to do is establish a benchmark that is not specific to any asset, that's trying to give what is the broad basket of crypto assets doing, the same way that you gauge the performance of the equity markets through the S&P 500," Tusar said.
Spotify launched a new revenue-sharing program for video podcasts .
"By giving you guys, the creators, another path to monetization beyond ads, we're freeing you up to do what you like doing, which is creating," Spotify CEO Daniel Ek told a crowd of creators and industry professionals during the company's Now Playing event on Wednesday, hosted at Spotify's Los Angeles campus.
A new mathematical approach dubbed "Crystal Math" could revolutionise drug development and electronics. Developed at NYU, it predicts crystal structures in hours on a laptop, rather than months on a supercomputer. Why does this matter? Crystal structures are crucial for everything from medicines to smartphone screens. For instance, this could dramatically speed up the creation of new drugs or more efficient solar panels.
"These physics-based approaches—which are costly and time-consuming—produce predictions that are only as accurate as the physics you put into them, which is why there has been a push toward computational methods that can address this shortcoming," said Mark Tuckerman, professor of chemistry and mathematics at NYU and the study's senior author.
Bloomberg explores the rise of combat sports empire TKO and its CEO Ari Emanuel's vision for the future of entertainment in this 24-minute video.
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 X at @thomasrice_au.