What is an AI server? Why artificial intelligence needs specialized systems

Stay Ahead, Stay ONMINE

What is an AI server? Why artificial intelligence needs specialized systems

Always remember: Design AI infrastructure for scalability, so you can add more capability when you need it. Comparison of different AI server models and configurations All the major players — Nvidia, Supermicro, Google, Asus, Dell, Intel, HPE — as well as smaller entrants are offering purpose-built AI hardware. Here’s a look at tools powering AI […]

Always remember: Design AI infrastructure for scalability, so you can add more capability when you need it.

Comparison of different AI server models and configurations

All the major players — Nvidia, Supermicro, Google, Asus, Dell, Intel, HPE — as well as smaller entrants are offering purpose-built AI hardware. Here’s a look at tools powering AI servers:

– Graphics processing units (GPUs): These specialized electronic circuits were initially designed to support real-time graphics for gaming. But their capabilities have translated well to AI, and their strengths are in their high processing power, scalability, security, quick execution and graphics rendering.

– Data processing units (DPUs): These systems on a chip (SoC) combine a CPU with a high-performance network interface and acceleration engines that can parse, process and transfer data at the speed of the rest of the network to improve AI performance.

– Application-specific integrated circuits (ASICs): These integrated circuits (ICs) are custom-designed for particular tasks. They are offered as gate arrays (semi-custom to minimize upfront design work and cost) and full-custom (for more flexibility and to process greater workflows).

– Tensor processing units (TPUs): Designed by Google, these cloud-based ASICs are suitable for a broad range of aI workloads, from training to fine-tuning to inference.

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.

Building connected data ecosystems for AI at scale

In partnership withSAP
.cst-large,
.cst-default {
width: 100%;
}

@media (max-width: 767px) {
.cst-block {
overflow-x: hidden;
}
}

@media (min-width: 630px) {
.cst-large {
margin-left: -25%;
width: 150%;
}

@media (min-width: 960px) {

Intel details new efficient Xeon processor line

The new chips will be able to support up to 12-channel DDR5 memory with speeds of up to 8000 MT/s, a substantial increase over the 8 channels of 6400MT/s in the prior generation. In addition to that, the platform will support up to 6 UPI 2.0 links with up to

Beyond Ping and SNMP: Building an AI-ready observability framework for enterprise networks

The solution and challenges For AI to work for us and alert us in advance, it should have good quality, reliable data over time, and this data can be retrieved from our classic logs when any event is triggered. Ping and SNMP would only provide data in polling time intervals

Fields medalist: ‘As of today we have no quantum computer. It does not exist.’

What’s more, I’m sure many of the employees of these companies have studied mathematics. That’s right, many are former mathematicians or professionals with a mathematical background. And, with respect to those trained in computer science, well, you have to remember that, at the beginning, computer scientists were part of the

Midad to Invest $5.4B in Algerian Production Sharing Contract

Sonatrach signed a production sharing contract (PSC) with Midad Energy for the Illizi South area, nearly 100 kilometers (62.14 miles) south of In Amenas, Algeria’s national oil and gas company said Monday. The Saudi company, through its Netherlands-based subsidiary Midad Energy North Africa BV, agreed to shoulder the full exploration and development cost. That is expected to be $5.4 billion, including $288 million for exploration, Sonatrach said in a statement on its website. The PSC is expected to produce, according to Sonatrach, a total of 993 million barrels of oil equivalent consisting of 125 billion cubic meters (4.41 trillion cubic feet) of gas for marketing and 204 million barrels of liquids, including 103 million barrels of liquefied petroleum gas and 101 million barrels of condensate. The contract lasts 30 years, including seven years for exploration, Sonatrach said. The contract is extendable by 10 years, it said. “This program includes also the use of the latest technological and digital solutions”, Sonatrach said. “Furthermore, calling on local content and subcontracting with national supplier under the execution of this contract will be prioritized”. Sonatrach chair and chief executive Rachid Hachichi and Midad Energy North Africa counterpart Sheikh Abdulelah Bin Mohammed Bin Abdullah Al-Aiban signed the PSC in the presence of Algerian Hydrocarbons and Mines Minister Mohamed Arkab and Saudi Ambassador to Algeria Abdullah Bin Nasser Abdullah Al-Busairi, according to Sonatrach. On July 21 Sonatrach announced it has entered five hydrocarbon contracts under Algeria’s 2024 bidding round. Ahara in Illizi province was signed with QatarEnergy and France’s TotalEnergies SE. Guern El Guessa II in the provinces of Bechar, Beni Abbes, El Bayadh and Timimoun was signed with China Petroleum and Chemical Corp (Sinopec). Toual II in the provinces of Ouargla and Illizi was signed with Austria’s Zangas Hoch- und Tiefbau GmbH and Switzerland’s Filada AG.

Valeura Raises Production at Nong Yao in Gulf of Thailand

Valeura Energy Inc has completed a 10-well infill drilling campaign at the Nong Yao oilfield in the Gulf of Thailand, increasing the company’s net production before royalties to 24,800 barrels per day (bpd) in the last seven days of the third quarter. “The campaign was primarily production-oriented and resulted in the company’s working interest share oil production before royalties from the Nong Yao field increasing from approximately 7,996 bpd prior to the first new wells coming on stream, to a recent rate of 11,562 bpd, over the seven-day period ending September 30, 2025”, Canada’s Valeura said in an operations update on its website. “The company anticipates that the reservoirs encountered may add to the ultimate production potential of the Nong Yao field and can thereby further extend its economic life”. The campaign involved all three wellhead infrastructure facilities in Nong Yao, Valeura said. The Nong Yao B segment of the campaign “included some of the most technically challenging wells ever drilled in the Gulf of Thailand basin, influenced by both geological complexity and also their extended reach from the wellhead platform, in one instance measuring a total drilled length of over 9,800′”, Valeura noted. Valeura operates Nong Yao with a 90 percent stake through License G11/48, in which Palang Sophon Co Ltd owns the remaining 10 percent. Nong Yao held proven and probable gross pre-royalties reserves of 16.9 million as of yearend 2024, according to Valuera. The field produces medium sweet crude from Miocene-age reservoirs, according to the company. Elsewhere in the Gulf of Thailand, work is progressing on the Wassana oilfield redevelopment project. Wassana’s newbuild wellhead production facility is on track to start up in the second quarter of 2027, Valeura added. “The Wassana redevelopment project is intended to increase production, reduce unit costs and create a hub for

Putin Relaxes Rules for Fuel Subsidies to Refiners

Russian President Vladimir Putin relaxed the rules on fuel subsidies to the nation’s refiners, allowing them to continue receiving billions of rubles in aid amid intensified Ukrainian attacks on the industry. Refineries will remain eligible for subsidies even if market wholesale prices for diesel and gasoline significantly exceed threshold prices, according to a presidential decree published on Sunday. The changes will stay in place from Oct. 1 until May 1. The subsidies are designed to encourage Russian oil producers to keep supplying gasoline and diesel to the domestic market, even when export prices become more attractive. Under current rules, the government is legally bound to pay the subsidy if the market wholesale price of gasoline doesn’t exceed 10% of the threshold price, with the range for diesel at 20%. Refineries won’t receive state payments after that level. Russia paid 1.8 trillion rubles ($22 billion) in fuel subsidies last year, while payments over the first nine months of 2025 fell to 716 billion rubles. Since early August, Ukraine has intensified attacks on the Russian oil industry, with repeated strikes on refineries resulting in a drop of crude-processing rates, exacerbating a fuel crunch and leading to rising domestic prices. Last month Deputy Prime Minister Alexander Novak said the fuel-price hike that triggers subsidies will be increased by 10 percentage points from September. The maximum allowed market-price hike would be raised to 20% of the threshold price for gasoline and 30% for diesel. WHAT DO YOU THINK? Generated by readers, the comments included herein do not reflect the views and opinions of Rigzone. All comments are subject to editorial review. Off-topic, inappropriate or insulting comments will be removed.

Oil Rebounds Amid Tariff Jitters

Oil stayed under $60 after the White House signaled openness to reaching a deal with China to quell fresh trade tensions between the two largest oil consumers. West Texas Intermediate rose 1% to settle at $59.60 a barrel, rebounding from Friday’s 4.2% plunge — the steepest drop since May. The shift in tone toward Beijing followed a fresh round of US tariffs and export curbs announced on Friday. Equities also rebounded on the development, lending support to crude. Tariff headlines are putting renewed bearish pressure on crude, said Razan Hilal, market analyst at StoneX. “Looking beyond the short term, such tariff tensions have previously acted as short-lived demand shocks in 2025 that paved the way for more favorable trade resolutions and cyclical recoveries.” “We’ll be fine with China,” the US president told reporters on Air Force One in early Asian hours on Monday, although the tariffs on Nov. 1 remained the plan. He also said he’d consider arming Ukraine with long-range Tomahawk missiles that would allow strikes deeper into Russia, which increases the risk of further disruptions to oil supply from the OPEC+ member. “Traders remain skeptical that trade with China may remain erratic until a deal is made, which could be a pressure point for crude near-term,” said Dennis Kissler, senior vice president for trading at BOK Financial. “On the other hand, WTI crude prices below $60 a barrel should continue to drop oil rig drilling numbers which will eventually equate to a drop in US production numbers.” China’s move to levy fees on US-owned vessels arriving at its shores prompted last-minute cancellations across ship types, including oil tankers, leading to a jump in shipping rates. The taxes, which take effect from Oct. 14, mirror those implemented by Washington on Chinese vessels as the Trump administration takes aim at

OPEC Keeps Oil Outlook Unchanged

OPEC kept its outlook for the oil market unchanged as the group and its allies continue to add barrels to the global market. Global oil demand will grow by 1.3 million barrels a day this year and by 1.4 million in 2026, the same as previously forecast, the Organization of the Petroleum Exporting Countries said in its monthly market report on Monday. OPEC+ — representing the group and its allies — is adding barrels to a global market that’s expected to be in surplus next year. The group increased output by 630,000 barrels a day in September, according to the report, citing information from secondary sources. Brent oil futures have slumped 18% in the past year, with a big part of the drop coming since April, when the OPEC+ hikes were first announced, to the surprise of traders. The projected demand for OPEC+ crude was also unchanged from last month’s report, at 42.5 million barrels a day this year. Next year, demand for the group’s crude is expected to rise to 43.1 million barrels a day. This month, the OPEC+ alliance agreed to revive just 137,000 barrels a day of halted supply, a slower pace than earlier this year. Other key figures from the monthly report: September output average about 43.05 million barrels a day. Oil demand in the OECD is expected to rise by around 130,000 barrels a day, with OECD Americas leading the increase. Non-OECD demand is expected to grow by around 1.2 million barrels a day. Transport fuels are expected to be the key growth source for oil demand this year, with jet fuel/kerosene adding 380,000 barrels a day of demand and diesel 300,000 barrels a day. Non-OPEC production is expected to grow by about 800,000 barrels a day this year, and 600,000 next year, also unchanged from last

USA EIA Lowers Henry Hub Price Projections for 2025 and 2026

In its latest short term energy outlook (STEO), which was released on October 7, the U.S. Energy Information Administration (EIA) lowered its Henry Hub natural gas spot price forecast for both 2025 and 2026. According to this STEO, the EIA now sees the commodity coming in at $3.42 per million British thermal units (MMBtu) in 2025 and $3.94 per MMBtu in 2026. In its previous STEO, which was released in September, the EIA projected that the Henry Hub natural gas spot price would average $3.52 per MMBtu this year and $4.28 per MMBtu next year. The EIA’s latest STEO sees the commodity averaging $3.33 per MMBtu in the fourth quarter of 2025, $3.86 per MMBtu in the first quarter of next year, $3.31 per MMBtu in the second quarter, $3.91 per MMBtu in the third quarter, and $4.68 per MMBtu in the fourth quarter of 2026. In its September STEO, the EIA forecast that the Henry Hub natural gas spot price would average $3.04 per MMBtu in the third quarter of this year, $3.72 per MMBtu in the fourth quarter, $4.25 per MMBtu in the first quarter of 2026, $3.64 per MMBtu in the second quarter, $4.26 per MMBtu in the third quarter, and $4.99 per MMBtu in the fourth quarter. The EIA’s October STEO highlighted that the commodity came in at $3.03 per MMBtu in the third quarter of 2025. Both the October and the September STEOs showed that the Henry Hub natural gas spot price averaged $4.15 per MMBtu in the first quarter of this year, $3.19 per MMBtu in the second quarter, and $2.19 per MMBtu overall in 2024. In its October STEO, the EIA said it expects the Henry Hub spot price “to increase from around $3.00 per MMBtu in September to $4.10 per MMBtu by

OpenAI–Broadcom alliance signals a shift to open infrastructure for AI

The decision also reflects a future of AI workloads running on heterogeneous computing and networking infrastructure, said Lian Jye Su, chief analyst at Omdia. “While it makes sense for enterprises to first rely on Nvidia’s full stack solution to roll out AI, they will generally integrate alternative solutions such as AMD and self-developed chips for cost efficiency, supply chain diversity, and chip availability,” Su said. “This means data center networking vendors will need to consider interoperability and open standards as ways to address the diversification of AI chip architecture.” Hyperscalers and enterprise CIOs are increasingly focused on how to efficiently scale up or scale out AI servers as workloads expand. Nvidia’s GPUs still underpin most large-scale AI training, but companies are looking for ways to integrate them with other accelerators. Neil Shah, VP for research at Counterpoint Research, said that Nvidia’s recent decision to open its NVLink interconnect to ecosystem players earlier this year gives hyperscalers more flexibility to pair Nvidia GPUs with custom accelerators from vendors such as Broadcom or Marvell. “While this reduces the dependence on Nvidia for a complete solution, it actually increases the total addressable market for Nvidia to be the most preferred solution to be tightly paired with the hyperscaler’s custom compute,” Shah said. Most hyperscalers have moved toward custom compute architectures to diversify beyond x86-based Intel or AMD processors, Shah added. Many are exploring Arm or RISC-V designs that can be tailored to specific workloads for greater power efficiency and lower infrastructure costs. Shifting AI infrastructure strategies The collaboration also highlights how networking choices are becoming as strategic as chip design itself, suggesting a change in how AI workloads are powered and connected.

Inside Nvidia’s ‘grid-to-chip’ vision: How Vera Rubin and Spectrum-XGS push toward AI giga-factories

Vera Rubin MGX brings together Nvidia’s Vera CPUs and Rubin CPX GPUs, all using the same open MGX rack footprint as Blackwell. The system allows for numerous configurations and integrations. “MGX is a flexible, modular building block-based approach to server and rack scale design,” Delaere said. “It allows our ecosystem to create a wide range of configurations, and do so very quickly.” Vera Rubin MGX will deliver almost eight times more performance than Nvidia’s GB 300 for certain types of calculation, he said. The architecture is liquid-cooled and cable-free, allowing for faster assembly and serviceability. Operators can quickly mix and match components such as CPUs, GPUs, or storage, supporting interoperability, Nvidia said. Matt Kimball, principal data center analyst at Moor Insights and Strategy, highlighted the modularity and cleanness of the MGX tray design. “This simplifies the manufacturing process significantly,” he said. For enterprises managing tens or even hundreds of thousands of racks, “this design enables a level of operational efficiency that can deliver real savings in time and cost.” Nvidia is also showing innovation with cooling, Kimball said. “Running cooling to the midplane is a very clean design and more efficient.”

Cisco seriously amps-up Silicon One chip, router for AI data center connectivity

Some say deep buffers shouldn’t be used to handle this type of traffic; the contention is that these buffers fill and drain, creating jitter in the workloads, and that slows things down, Chopra told Network World. “But the real source of that challenge is not the buffers. It’s a poor congestion management scheme and poor load balancing with AI workloads, which are completely deterministic and predictable. You can actually proactively figure out how to place flows across the network and avoid the congestion,” he said. The 8223’s deep-buffer design provides ample memory to temporarily store packets during congestion or traffic bursts, an essential feature for AI networks where inter-GPU communication can create unpredictable, high-volume data flows, according to Gurudatt Shenoy, vice president of Cisco Provider Connectivity. “Combined with its high-radix architecture, the 8223 allows more devices to connect directly, reducing latency, saving rack space, and further lowering power consumption. The result is a flatter, more efficient network topology supporting high-bandwidth, low-latency communication that is critical for AI workloads,” Shenoy wrote in a blog post. NOS options Notably, the first operating systems that the 8223 supports are the Linux Foundation’s Software for Open Networking in the Cloud (SONiC) and Facebook open switching system (FBOSS) – not Cisco’s own IOS XR. IXR will be supported, too, but at a later date, according to Cisco. SONiC decouples network software from the underlying hardware and lets it run on hundreds of switches and ASICs from multiple vendors while supporting a full suite of network features such as Border Gateway Protocol (BGP), remote direct memory access (RDMA), QoS, and Ethernet/IP. One of the keys to SONiC is its switch-abstraction interface, which defines an API to provide a vendor-independent way of controlling forwarding elements such as a switching ASIC, an NPU, or a software switch in a uniform

Utilities Race to Meet Surging Data Center Demand With New Power Models

Over the last 18 months or so, the energy generation industry and its public utilities have been significantly impacted by the AI data center boom. It has been demonstrated across North America that the increase in demand for power, as driven by the demand for hyperscale and AI data centers, greatly exceeds the ability of the industry to actually generate and deliver power to meet the demand. We have covered many of the efforts being made to control the availability of power. In response, utilities and regulators have begun rethinking how to manage power availability through means such as: temporary moratoriums on new data center interconnections; the creation of new rate classes; cogeneration and load-sharing agreements; renewable integration; and power-driven site selection strategies. But the bottom line is that in many locations utilities will need to change the way they work and how and where they spend their CAPEX budgets. The industry has already realized that their demand forecast models are hugely out of date, and that has had a ripple effect on much of the planning done by public utilities to meet the next generation of power demand. Most utilities now acknowledge that their demand forecasting models have fallen behind reality, triggering revisions to Integrated Resource Plans (IRPs) and transmission buildouts nationwide. This mismatch between forecast and actual demand is forcing a fundamental rethink of capital expenditure priorities and long-term grid planning. Spend More, Build Faster Utilities are sharply increasing CAPEX and rebalancing their resource portfolios—not just for decarbonization, but to keep pace with multi-hundred-megawatt data center interconnects. This trend is spreading across the industry, not confined to a few isolated utilities. Notable examples include: Duke Energy raised its five-year CAPEX plan to $83 billion (a 13.7% increase) and plans to add roughly 5 GW of natural gas capacity

Duos Pairs Mobile Power and Modular Edge Data Centers for Rapid Texas Rollout

Duos Technology Group has launched the fifth of its AI edge data centers, part of a plan to deploy 15 units by the end of 2025. The projects are executed through Duos Edge AI, a subsidiary focused on modular, rapidly installed edge data centers (EDCs) in underserved markets, beginning with school districts and regional carrier hubs across Texas. The newest site is being deployed on-premises with the Dumas Independent School District in Dumas, Texas. High-Density Edge Design Duos’ EDCs emphasize very high rack densities (100 kW+ per rack), SOC 2 Type II compliance, N+1 power with dual generators, and a 90-day build/turn-up cycle. Each site is positioned approximately 12 miles from end users, cutting latency for real-time workloads. To meet the power demands of these edge deployments, Duos formed Duos Energy and partnered with Fortress/APR Energy to deliver behind-the-meter mobile gas turbines. This approach allows compute to go live in 90 days without waiting years for utility interconnection upgrades. The goal is straightforward: move power and compute close to demand, with rapid deployment. Duos’ modular pods are designed for exurban and rural locations as localized compute hubs for carriers, schools, healthcare systems, and municipal users. The rugged design pairs high-density racks with the short deployment cycle and proximity targeting, enabling a wide range of applications. With Dumas ISD now live, Duos has five sites in Texas, including Amarillo/Region 16, Victoria/Region 3, Dumas ISD, and multiple Corpus Christi locations. Mobile Power vs. Modular Compute While Duos doesn’t consistently describe its data center units as “mobile,” they are modular and containerized, engineered for rapid, site-agnostic deployment. The “mobile” label more explicitly applies to Duos’ power strategy—a turbine fleet that can be fielded or re-fielded to match demand. From an operator’s perspective, the combined proposition functions like a mobile platform: pre-integrated compute pods

Report: AMD could be Intel’s next foundry customer

[ Related: More Intel news and insights ] AMD has lagged behind Nvidia in the AI business but has done well in the federal supercomputing business, holding numerous top spots with supercomputers like El Capitan and Frontier. Manufacturing its chips in the United States would be a good way to get the Trump administration off its back given its push for domestic manufacturing of semiconductors. The Trump administration is pushing for 50% of chips sold in America to be manufactured domestically, and tariffs on chips that are not. It also faces outbound restrictions. Earlier this year, AMD faced export restrictions GPUs meant for China as part of U.S. export controls against China’s AI business. “I believe this is a smart move by AMD to secure capacity in the local market without fighting against Nvidia and Apple and their deeper pockets for the limited capacity at TSMC,” said Alvi Nguyen, senior analyst with Forrester Research.” With the US investment in Intel, followed by Nvidia, this is can be seen as diversifying their supply chain and providing cheaper, locally sourced parts.” For Intel, this will continue a streak of good news it has enjoyed recently. “Having customers take up capacity at their foundries will go a long way in legitimizing their semiconductor processes and hopefully create the snowball effect of getting even more US-based customers,” said Nguyen. In recent weeks, Intel has partnered with Nvidia to jointly make PC and data center chips. Nvidia also took a $5B stake in Intel. Earlier the Trump administration made a $11.1B, or 10%, stake in Intel.

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.

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

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

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

Stay Ahead, Stay ONMINE