Stay Ahead, Stay ONMINE

Cheaper buildings, courtesy of mud

One costly and time-consuming step in constructing a concrete building is creating the “formwork,” the wooden mold into which the concrete is poured. Now MIT researchers have developed a way to replace the wood with lightly treated mud. “What we’ve demonstrated is that we can essentially take the ground we’re standing on, or waste soil from a construction site, and transform it into accurate, highly complex, and flexible formwork for customized concrete structures,” says Sandy Curth, a PhD student in MIT’s Department of Architecture, who has helped spearhead the project. The EarthWorks method, as it’s known, introduces some additives, such as straw, and a waxlike coating to the soil material. Then it’s 3D-printed into a custom-designed shape. “We found a way to make formwork that is infinitely recyclable,” Curth says. “It’s just dirt.” A particular advantage of the technique is that the material’s flexibility makes it easier to create unique shapes optimized so that the resulting buildings use no more concrete than structurally necessary. This can significantly reduce the carbon emissions associated with concrete construction. “What’s cool here is we’re able to make shape-optimized building elements for the same amount of time and energy it would take to make rectilinear building elements,” says Curth, who recently coauthored a paper on the work with MIT professors Lawrence Sass, SM ’94, PhD ’00; Caitlin Mueller ’07, SM ’14, PhD ’14; and others. He has also founded a firm, Forma Systems, through which he hopes to take EarthWorks into the construction industry.

One costly and time-consuming step in constructing a concrete building is creating the “formwork,” the wooden mold into which the concrete is poured. Now MIT researchers have developed a way to replace the wood with lightly treated mud.

“What we’ve demonstrated is that we can essentially take the ground we’re standing on, or waste soil from a construction site, and transform it into accurate, highly complex, and flexible formwork for customized concrete structures,” says Sandy Curth, a PhD student in MIT’s Department of Architecture, who has helped spearhead the project.

The EarthWorks method, as it’s known, introduces some additives, such as straw, and a waxlike coating to the soil material. Then it’s 3D-printed into a custom-designed shape. “We found a way to make formwork that is infinitely recyclable,” Curth says. “It’s just dirt.”

A particular advantage of the technique is that the material’s flexibility makes it easier to create unique shapes optimized so that the resulting buildings use no more concrete than structurally necessary. This can significantly reduce the carbon emissions associated with concrete construction.

“What’s cool here is we’re able to make shape-optimized building elements for the same amount of time and energy it would take to make rectilinear building elements,” says Curth, who recently coauthored a paper on the work with MIT professors Lawrence Sass, SM ’94, PhD ’00; Caitlin Mueller ’07, SM ’14, PhD ’14; and others. He has also founded a firm, Forma Systems, through which he hopes to take EarthWorks into the construction industry.

Shape
Shape
Stay Ahead

Explore More Insights

Stay ahead with more perspectives on cutting-edge power, infrastructure, energy,  bitcoin and AI solutions. Explore these articles to uncover strategies and insights shaping the future of industries.

Shape

Linkerd 2.18 advances cloud-native service mesh

The project’s focus has evolved significantly over the years. While early adoption centered on mutual TLS between pods, today’s enterprises are tackling much larger challenges. “For a long time, the most common pattern was simply, ‘I want to get mutual TLS between all my pods, which gives me encryption, and

Read More »

18 essential commands for new Linux users

[jdoe@fedora ~]$ ls -ld /home/jdoedrwx——. 1 jdoe jdoe 106 Apr 3 14:39 /home/jdoe As you may have suspected, “r” stands for read, “w” means write and “x” is for execute. Note that no permissions are available for other group members and anyone else on the system. Each user will be

Read More »

Lightshift to build 11-MW storage project for Virginia municipal utility

Danville Utilities, a municipal utility based in Danville, Virginia, is contracting for an 11-MW/44-MWh battery storage project to be built and owned by Lightshift Energy, the battery storage company said Wednesday. Danville Utilities plans to use the storage facility to reduce its peak demand, which it expects will lower its demand and transmission charges by about $30 million over the 20-year life of the project. About 40% of Danville Utilities’ power supply costs are demand-related and they are increasing by 10% to 15% a year, according to a Nov. 7 Danville City Council memo on the project, which Lightshift expects will be operating in the second quarter next year. The memo states that PJM Interconnection’s capacity rate for Danville Utilities is jumping to $8.29/kW-month starting June 1 from 96 cents/kW-month. An existing 10.6-MW storage project built by Lightshift has helped reduce the Danville Utilities’ transmission, capacity and congestion charges since it started operating in 2022, according to the memo from Lightshift, formerly Delorean Power. Lightshift expects the project will save the utility $40 million over its life. Lightshift will use the new battery system to sell ancillary services into the PJM market and arbitrage wholesale energy prices, according to the memo.  “The project will help lower transmission and capacity costs and allow the city’s electric rates to be competitive to other neighboring electric utilities when competing for economic development projects,” Jason Grey, Danville director of utilities, said in the press release. Danville Utilities has about 42,000 customers. The project in February received $1.5 million from the Virginia Tobacco Region Revitalization Commission’s Energy Ingenuity Fund. Municipal utilities and rural cooperatives are focused on cost, and energy storage can save them money while adding resilience to their systems, according to Laura Coriell, head of market development for Lightshift, which is based in

Read More »

Carbon capture technology is ready. Permitting needs to catch up.

Anna Littlefield is the program manager for low carbon energy technologies for the Payne Institute at the Colorado School of Mines. It’s no secret that carbon capture is gaining momentum — and the United States is leading the pack. U.S. capacity alone for carbon capture, utilization and storage, or CCUS, is expected to increase sevenfold by 2035. In 2023, announced carbon capture capacity by 2030 increased by 35%. In the U.S., there are already 14 operational commercial-scale facilities that can capture and store roughly 21.4 million metric tons of CO2 per year, equivalent to the annual emissions of 5 million passenger vehicles. And new projects continue to be announced. Since 2018 when Congress enhanced the 45Q tax credit, which supports CCUS, there have been more than 120 CCUS projects announced in the states. In 2023, Chevron announced it was expanding its Bayou Bend CCS project, positioning it to be one of the largest carbon storage projects in the country. Despite all this investment and decades of proven operations, critics continue to insist the technology doesn’t work — apparently because it hasn’t scaled fast enough. A Scientific American headline from 2017 asked, “Will Carbon Capture and Storage Ever Work?” One environmental group asserted the technology “won’t work,” while others call it a “pipe dream.” This narrative allows other critics to suggest any federal support for the technology is misallocated. Two U.S. lawmakers recently introduced the 45Q Repeal Act, a bill aimed at repealing the 45Q tax credit. Rep. Scott Perry, R-Pa., one of the sponsors of the bill, called CCUS an “inefficient and market-distorting technology.” Opponents of CCUS cheered the legislation. What these critics overlook is that the primary challenge facing CCUS is the regulatory uncertainty created by current federal policies. Fifteen years ago, the Task Force on Carbon Capture and Storage

Read More »

Smaller, public utilities see growth potential in data centers, but there are risks: APPA

Dive Brief: The boom in AI data center development is driving significant growth in large-load interconnection requests to not-for-profit public utilities, two experts with the American Public Power Association told Utility Dive. Public power electric utilities tend to be smaller on average than investor-owned utilities,  emphasize affordability and reliability, and are often self-governing, all of which appeal to data center operators, said Patricia Taylor, APPA’s manager for regulatory policy and business programs. But smaller utilities weighing data center proposals that could dramatically increase their system loads must also take steps to protect existing ratepayers and other stakeholders, such as carefully studying the potential grid reliability impacts and ensuring fair compensation for infrastructure upgrades, said Latif Nurani, senior regulatory counsel for APPA. Dive Insight: Smaller, not-for-profit utilities have worked productively with data center operators for decades and have in some cases helped their communities evolve into significant IT hubs, Nurani and Taylor said. In north-central Oregon, Northern Wasco County People’s Utility District nearly quadrupled its annual revenues from $32 million in 2016 to $120 million in 2024 and expects to reach $300 million “in the next few years,” CEO and General Manager Roger Kline said on an APPA podcast in July. Google data centers drove much of that increase. Drawn by abundant hydropower and associated energy infrastructure, a robust workforce and a “business-friendly environment and policies,” the tech giant developed its first data center near The Dalles, Oregon, in 2006 and has since invested more than $2.4 billion in the state, it says. Google’s presence has been transformative for Northern Wasco County PUD, enabling investments in new substations, transmission lines, hydroelectric plants, advanced metering infrastructure and other projects that may have otherwise exceeded the capacity of a “small semi-rural utility” with fewer than 25,000 residential and commercial customers, Kline said last

Read More »

USA Upstream M&A Hits $17B in 1Q

In a release sent to Rigzone by the Enverus team late Tuesday, the company’s subsidiary, Enverus Intelligence Research (EIR), revealed that U.S. upstream mergers and acquisitions (M&A) hit $17 billion in deal value in the first quarter of 2025 and highlighted that this was the second best start to a year since 2018. “However, activity was disproportionately driven by one company, Diamondback Energy, which accounted for nearly 50 percent of total value between its acquisition of Double Eagle IV and a dropdown of minerals to its affiliate Viper Energy Partners,” EIR noted in the release. “Outside of Diamondback, buyers were already feeling the pressure of limited acquisition opportunities and high asking prices for undeveloped drilling inventory,” EIR added. “On top of that, upstream companies will now have to navigate significant headwinds from falling oil and equity values,” EIR warned. In the release, EIR stated that, prior to OPEC and tariffs creating waves in oil markets, pricing for quality shale inventory was a perpetually rising tide. It added that, historically, lower crude prices have taken the wind out of the sails of upstream M&A. “Going back to the start of 2014, oil prices have fallen by more than five percent quarter over quarter 17 times,” EIR highlighted. “In 11 of the quarters with materially lower crude prices, deal activity fell compared to the prior three months with an average decline in transacted deal value of 30 percent,” it pointed out. “Asset values have also declined when crude prices moved 20 percent or more lower year over year, with the value of Permian acreage falling about one-third in 2015 compared to 2014 and losing more than half its value in 2020 over 2019, based on the average price per acre paid,” it continued. “The only exception to the trend over the last

Read More »

Naftogaz Keeps Gas Prices Unchanged for Another Year

Naftogaz Group said Wednesday it will continue delivering natural gas to households in Ukraine at the current regulated rate under its fixed tariff plan. The price will remain at UAH 7.96 ($0.19) per cubic meter until April 2026 to reflect the current moratorium on raising natural gas tariffs, the state-owned oil and gas company said in an online statement. However, the company added, “Naftogaz also encourages consumers to pay their bills on time”. Acting chief executive Roman Chumak added, “These payments help maintain the country’s energy stability in the face of ongoing attacks on critical infrastructure”. “Supplying gas to households remains a core priority. We continue to ensure stable delivery and meet our obligations to consumers, even in the most critical conditions”, said Chumak. Naftogaz said it remains a reliable gas supplier for 12.5 million households. According to the company it led Ukraine through last winter without gas blackouts. “Risks peaked in February, when large-scale Russian missile attacks on gas production facilities led to the sudden loss of nearly half of the state’s output. These events created a perfect storm that could have caused a nationwide gas blackout at any moment”, according to a company report Monday that said 34 of Naftogaz’s gas production sites had been attacked in 2024-25. The report said for next winter Naftogaz has contracted 400 million cubic meters of gas. “Since the beginning of the year, 1.5 billion cubic meters of gas have been contracted: 800 million cubic meters were urgently imported early in the year, 400 million cubic meters will be delivered to Ukraine as part of the winter readiness plan”, Chumak said. “In addition, Naftogaz purchased 300 million cubic meters of LNG from ORLEN”. Naftogaz had signed an LNG cooperation deal with Poland’s majority state-owned ORLEN SA to help Ukraine diversify its energy

Read More »

Decarbonise North Sea ‘damn hard and damn fast’

The UK needs to focus on decarbonising the North Sea energy industry “damn hard and damn fast”. A panel of industry specialists at a conference discussed what Energy Industries Council (EIC) president Campbell Keir called an “urgent necessity” to “reshape industries, create new partnerships, generate sustainable jobs and secure a future where energy and environment responsibilities built together”. In the first session of the North Sea Decarbonisation conference, RV Ahilan Ahilan, chief energy transition officer of ABL Group, admitted annoyance with UK energy policy. ABL, is headquartered in London but listed on the Oslo stock exchange which gives the global firm insight into both sectors of the North Sea. “Even looking across the pond to Norway, we have a completely inconsistent policy about oil and gas development between the two countries. “One is ready to have no oil and gas licenses. The other one is saying we should have more “We have to somehow find a way to attract industry that can work on both sides of this agenda, and that’s the bit that I’m consistently having to struggle with. “The second part is that I think it’s also very annoying that we have a decarbonisation target and yet heavy subsidies for fossil fuels and subsidies for renewables without actually redirecting those in an efficient manner across the country’s GDP. That’s a big problem. It’s putting the foot on the accelerator and on the break at the same time and expecting decarbonisation to happen. And that’s very provocative. Matt Abraham, head of operations for James Fisher and Sons, called for acceleration of North Sea decommissioning as a means of driving carbon reduction and industry innovation while also maintaining oil and gas production “We could lead the world in the decommissioning of installation. Genuinely, there’s a whole lot of new different

Read More »

Slowdown in AWS data center leasing plans poses little threat to CIOs

Oracle, according to Westfall, is committed to investing $10 billion in 2025 to build 100 new data centers and expand 66 existing ones, aiming to double its capacity this year. Likewise, Google is investing $75 billion in 2025 for data center construction, focusing on AI and cloud infrastructure, with projects such as a $600 million facility in Mesa, Arizona, and a $2 billion data center in Fort Wayne and Indiana underway, Westfall said. Meta, too, plans to spend up to $65 billion in 2025, a sizable bump up from $40 billion in 2024, primarily for data center expansion to support AI (Llama models, Meta AI) and metaverse workloads, Westfall added. However, these expansion plans will not result in the relatively smaller players catching up with AWS and Microsoft. “For smaller players like Google and Oracle, catching up with AWS and Microsoft would require historically large capital investments that likely aren’t justified by their current growth rates,” Alletto said.

Read More »

TSMC targets AI acceleration with A14 process and ‘System on Wafer-X’

Nvidia’s flagship GPUs currently integrate two chips, while its forthcoming Rubin Ultra platform will connect four. “The SoW-X delivers wafer-scale compute performance and significantly boosts speed by integrating multiple advanced compute SoC dies, stacked HBM memory, and optical interconnects into a single package,” said Neil Shah, partner and co-founder at Counterpoint Research. “This approach reduces latency, improves power efficiency, and enhances scalability compared to traditional multi-chip setups — giving enterprises and hyperscalers AI servers capable of handling future workloads faster, more efficiently, and in a smaller footprint.” This not only boosts capex savings in the long run but also opex savings in terms of energy and space. “Wafer-X technology isn’t just about bigger chips — it’s a signal that the future of AI infrastructure is being redesigned at the silicon level,” said Abhivyakti Sengar, practice director at Everest Group. “By tightly integrating compute, memory, and optical interconnects within a single wafer-scale package, TSMC targets the core constraints of AI: bandwidth and energy. For hyperscale data centers and frontier model training, this could be a game-changer.” Priorities for enterprise customers For enterprises investing in custom AI silicon, choosing the right foundry partner goes beyond performance benchmarks. It’s about finding a balance between cutting-edge capabilities, flexibility, and cost. “First, enterprise buyers need to assess manufacturing process technologies (such as TSMC’s 3nm, 2nm, or Intel’s 18A) to determine if they meet AI chip performance and power requirements, along with customization capabilities,” said Galen Zeng, senior research manager for semiconductor research at IDC Asia Pacific. “Second, buyers should evaluate advanced packaging abilities; TSMC leads in 3D packaging and customized packaging solutions, suitable for highly integrated AI chips, while Intel has advantages in x86 architecture. Finally, buyers should assess pricing structures.”

Read More »

Cloudbrink pushes SASE boundaries with 300 Gbps data center throughput

Those core components are functionally table stakes and don’t really serve to differentiate Cloudbrink against its myriad competitors in the SASE market. Where Cloudbrink looks to differentiate is at a technical level through a series of innovations including: Distributed edge architecture: The company has decoupled software from hardware, allowing their platform to run across 800 data centers by leveraging public clouds, telco networks and edge computing infrastructure. This approach reduces network latency from 300 milliseconds to between 7 and 20 milliseconds, the company says. This density dramatically improves TCP performance and responsiveness. Protocol optimization: Cloudbrink developed its own algorithms for SD-WAN optimization that bring enterprise-grade reliability to last mile links. These algorithms significantly improve efficiency on consumer broadband connections, enabling enterprise-grade performance over standard internet links. Integrated security stack: “We’ve been able to produce secure speeds at line rate on our platform by bringing security to the networking stack itself,” Mana noted. Rather than treating security as a separate overlay that degrades performance, Cloudbrink integrates security functions directly into the networking stack. The solution consists of three core components: client software for user devices, a cloud management plane, and optional data center connectors for accessing internal applications. The client intelligently connects to multiple edge nodes simultaneously, providing redundancy and application-specific routing optimization. Cloudbrink expands global reach Beyond its efforts to increase throughput, Cloudbrink is also growing its global footprint. Cloudbrink today announced a global expansion through new channel agreements and the opening of a Brazil office to serve emerging markets in Latin America, Korea and Africa. The expansion includes exclusive partnerships with WITHX in Korea, BAMM Technologies for Latin America distribution and OneTic for African markets. The company’s software-defined FAST (Flexible, Autonomous, Smart and Temporary) Edges technology enables rapid deployment of points of presence by leveraging existing infrastructure from multiple

Read More »

CIOs could improve sustainability with data center purchasing decisions — but don’t

CIOs can drive change Even though it’s difficult to calculate an organization’s carbon footprint, CIOs and IT purchasing leaders trying to reduce their environmental impact can influence data center operators, experts say. “Customers have a very large voice,” Seagate’s Feist says. “Don’t underestimate how powerful that CIO feedback loop is. The large cloud accounts are customer-obsessed organizations, so they listen, and they react.” While DataBank began using renewable energy years ago, customer demand can push more data center operators to follow suit, Gerson says. “For sure, if there is a requirement to purchase renewable power, we are going to purchase renewable power,” she adds.

Read More »

Copper-to-optics technology eyed for next-gen AI networking gear

Broadcom’s demonstration and a follow-up session explored the benefits of further developing CPC, such as reduced signal integrity penalties and extended reach, through channel modeling and simulations, Broadcom wrote in a blog about the DesignCon event. “Experimental results showed successful implementation of CPC, demonstrating its potential to address bandwidth and signal integrity challenges in data centers, which is crucial for AI applications,” Broadcom stated. In addition to the demo, Broadcom and Samtec also authored a white paper on CPC that stated: “Co-packaged connectivity (CPC) provides the opportunity to omit loss and reflection penalties from the [printed circuit board (PCB)] and the package. When high speed I/O is cabled from the top of the package advanced PCB materials are not necessary. Losses from package vertical paths and PCB routing can be transferred to the longer reach of cables,” the authors stated. “As highly complex systems are challenged to scale the number of I/O and their reach, co- packaged connectivity presents opportunity. As we approach 224G-PAM4 [which uses optical techniques to support 224 Gigabits per second data rates per optical lane] and above, system loss and dominating noise sources necessitate the need to re-consider that which has been restricted in the back of the system architect’s mind for years: What if we attached to the package?” At OFC, Samtec demonstrated its Si-FlyHD co-packaged cable assemblies and Samtec FlyoverOctal Small Form-factor Pluggable (OSFP) over the Samtec Eye Speed Hyper Low Skew twinax copper cable. Flyover is Samtec’s proprietary way of addressing signal integrity and reach limitations of routing high-speed signals through traditional printed circuit boards (PCBs). “This evaluation platform incorporates Broadcom’s industry-leading 200G SerDes technology and Samtec’s co-packaged Flyover technology. Si-Fly HD CPC offers the industry’s highest footprint density and robust interconnect which enables 102.4T (512 lanes at 200G) in a 95 x

Read More »

The Rise of AI Factories: Transforming Intelligence at Scale

AI Factories Redefine Infrastructure The architecture of AI factories reflects a paradigm shift that mirrors the evolution of the industrial age itself—from manual processes to automation, and now to autonomous intelligence. Nvidia’s framing of these systems as “factories” isn’t just branding; it’s a conceptual leap that positions AI infrastructure as the new production line. GPUs are the engines, data is the raw material, and the output isn’t a physical product, but predictive power at unprecedented scale. In this vision, compute capacity becomes a strategic asset, and the ability to iterate faster on AI models becomes a competitive differentiator, not just a technical milestone. This evolution also introduces a new calculus for data center investment. The cost-per-token of inference—how efficiently a system can produce usable AI output—emerges as a critical KPI, replacing traditional metrics like PUE or rack density as primary indicators of performance. That changes the game for developers, operators, and regulators alike. Just as cloud computing shifted the industry’s center of gravity over the past decade, the rise of AI factories is likely to redraw the map again—favoring locations with not only robust power and cooling, but with access to clean energy, proximity to data-rich ecosystems, and incentives that align with national digital strategies. The Economics of AI: Scaling Laws and Compute Demand At the heart of the AI factory model is a requirement for a deep understanding of the scaling laws that govern AI economics. Initially, the emphasis in AI revolved around pretraining large models, requiring massive amounts of compute, expert labor, and curated data. Over five years, pretraining compute needs have increased by a factor of 50 million. However, once a foundational model is trained, the downstream potential multiplies exponentially, while the compute required to utilize a fully trained model for standard inference is significantly less than

Read More »

Microsoft will invest $80B in AI data centers in fiscal 2025

And Microsoft isn’t the only one that is ramping up its investments into AI-enabled data centers. Rival cloud service providers are all investing in either upgrading or opening new data centers to capture a larger chunk of business from developers and users of large language models (LLMs).  In a report published in October 2024, Bloomberg Intelligence estimated that demand for generative AI would push Microsoft, AWS, Google, Oracle, Meta, and Apple would between them devote $200 billion to capex in 2025, up from $110 billion in 2023. Microsoft is one of the biggest spenders, followed closely by Google and AWS, Bloomberg Intelligence said. Its estimate of Microsoft’s capital spending on AI, at $62.4 billion for calendar 2025, is lower than Smith’s claim that the company will invest $80 billion in the fiscal year to June 30, 2025. Both figures, though, are way higher than Microsoft’s 2020 capital expenditure of “just” $17.6 billion. The majority of the increased spending is tied to cloud services and the expansion of AI infrastructure needed to provide compute capacity for OpenAI workloads. Separately, last October Amazon CEO Andy Jassy said his company planned total capex spend of $75 billion in 2024 and even more in 2025, with much of it going to AWS, its cloud computing division.

Read More »

John Deere unveils more autonomous farm machines to address skill labor shortage

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Self-driving tractors might be the path to self-driving cars. John Deere has revealed a new line of autonomous machines and tech across agriculture, construction and commercial landscaping. The Moline, Illinois-based John Deere has been in business for 187 years, yet it’s been a regular as a non-tech company showing off technology at the big tech trade show in Las Vegas and is back at CES 2025 with more autonomous tractors and other vehicles. This is not something we usually cover, but John Deere has a lot of data that is interesting in the big picture of tech. The message from the company is that there aren’t enough skilled farm laborers to do the work that its customers need. It’s been a challenge for most of the last two decades, said Jahmy Hindman, CTO at John Deere, in a briefing. Much of the tech will come this fall and after that. He noted that the average farmer in the U.S. is over 58 and works 12 to 18 hours a day to grow food for us. And he said the American Farm Bureau Federation estimates there are roughly 2.4 million farm jobs that need to be filled annually; and the agricultural work force continues to shrink. (This is my hint to the anti-immigration crowd). John Deere’s autonomous 9RX Tractor. Farmers can oversee it using an app. While each of these industries experiences their own set of challenges, a commonality across all is skilled labor availability. In construction, about 80% percent of contractors struggle to find skilled labor. And in commercial landscaping, 86% of landscaping business owners can’t find labor to fill open positions, he said. “They have to figure out how to do

Read More »

2025 playbook for enterprise AI success, from agents to evals

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More 2025 is poised to be a pivotal year for enterprise AI. The past year has seen rapid innovation, and this year will see the same. This has made it more critical than ever to revisit your AI strategy to stay competitive and create value for your customers. From scaling AI agents to optimizing costs, here are the five critical areas enterprises should prioritize for their AI strategy this year. 1. Agents: the next generation of automation AI agents are no longer theoretical. In 2025, they’re indispensable tools for enterprises looking to streamline operations and enhance customer interactions. Unlike traditional software, agents powered by large language models (LLMs) can make nuanced decisions, navigate complex multi-step tasks, and integrate seamlessly with tools and APIs. At the start of 2024, agents were not ready for prime time, making frustrating mistakes like hallucinating URLs. They started getting better as frontier large language models themselves improved. “Let me put it this way,” said Sam Witteveen, cofounder of Red Dragon, a company that develops agents for companies, and that recently reviewed the 48 agents it built last year. “Interestingly, the ones that we built at the start of the year, a lot of those worked way better at the end of the year just because the models got better.” Witteveen shared this in the video podcast we filmed to discuss these five big trends in detail. Models are getting better and hallucinating less, and they’re also being trained to do agentic tasks. Another feature that the model providers are researching is a way to use the LLM as a judge, and as models get cheaper (something we’ll cover below), companies can use three or more models to

Read More »

OpenAI’s red teaming innovations define new essentials for security leaders in the AI era

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More OpenAI has taken a more aggressive approach to red teaming than its AI competitors, demonstrating its security teams’ advanced capabilities in two areas: multi-step reinforcement and external red teaming. OpenAI recently released two papers that set a new competitive standard for improving the quality, reliability and safety of AI models in these two techniques and more. The first paper, “OpenAI’s Approach to External Red Teaming for AI Models and Systems,” reports that specialized teams outside the company have proven effective in uncovering vulnerabilities that might otherwise have made it into a released model because in-house testing techniques may have missed them. In the second paper, “Diverse and Effective Red Teaming with Auto-Generated Rewards and Multi-Step Reinforcement Learning,” OpenAI introduces an automated framework that relies on iterative reinforcement learning to generate a broad spectrum of novel, wide-ranging attacks. Going all-in on red teaming pays practical, competitive dividends It’s encouraging to see competitive intensity in red teaming growing among AI companies. When Anthropic released its AI red team guidelines in June of last year, it joined AI providers including Google, Microsoft, Nvidia, OpenAI, and even the U.S.’s National Institute of Standards and Technology (NIST), which all had released red teaming frameworks. Investing heavily in red teaming yields tangible benefits for security leaders in any organization. OpenAI’s paper on external red teaming provides a detailed analysis of how the company strives to create specialized external teams that include cybersecurity and subject matter experts. The goal is to see if knowledgeable external teams can defeat models’ security perimeters and find gaps in their security, biases and controls that prompt-based testing couldn’t find. What makes OpenAI’s recent papers noteworthy is how well they define using human-in-the-middle

Read More »