Get ONMINE

Stay Ahead, Stay ONMINE

How to build a better AI benchmark

It’s not easy being one of Silicon Valley’s favorite benchmarks.  SWE-Bench (pronounced “swee bench”) launched in November 2024 to evaluate an AI model’s coding skill, using more than 2,000 real-world programming problems pulled from the public GitHub repositories of 12 different Python-based projects.  In the months since then, it’s quickly become one of the most popular tests in AI. A SWE-Bench score has become a mainstay of major model releases from OpenAI, Anthropic, and Google—and outside of foundation models, the fine-tuners at AI firms are in constant competition to see who can rise above the pack. The top of the leaderboard is a pileup between three different fine tunings of Anthropic’s Claude Sonnet model and Amazon’s Q developer agent. Auto Code Rover—one of the Claude modifications—nabbed the number two spot in November, and was acquired just three months later. Despite all the fervor, this isn’t exactly a truthful assessment of which model is “better.” As the benchmark has gained prominence, “you start to see that people really want that top spot,” says John Yang, a researcher on the team that developed SWE-Bench at Princeton University. As a result, entrants have begun to game the system—which is pushing many others to wonder whether there’s a better way to actually measure AI achievement. Developers of these coding agents aren’t necessarily doing anything as straightforward cheating, but they’re crafting approaches that are too neatly tailored to the specifics of the benchmark. The initial SWE-Bench test set was limited to programs written in Python, which meant developers could gain an advantage by training their models exclusively on Python code. Soon, Yang noticed that high-scoring models would fail completely when tested on different programming languages—revealing an approach to the test that he describes as “gilded.” “It looks nice and shiny at first glance, but then you try to run it on a different language and the whole thing just kind of falls apart,” Yang says. “At that point, you’re not designing a software engineering agent. You’re designing to make a SWE-Bench agent, which is much less interesting.” The SWE-Bench issue is a symptom of a more sweeping—and complicated—problem in AI evaluation, and one that’s increasingly sparking heated debate: The benchmarks the industry uses to guide development are drifting further and further away from evaluating actual capabilities, calling their basic value into question. Making the situation worse, several benchmarks, most notably FrontierMath and Chatbot Arena, have recently come under heat for an alleged lack of transparency. Nevertheless, benchmarks still play a central role in model development, even if few experts are willing to take their results at face value. OpenAI cofounder Andrej Karpathy recently described the situation as “an evaluation crisis”: the industry has fewer trusted methods for measuring capabilities and no clear path to better ones.  “Historically, benchmarks were the way we evaluated AI systems,” says Vanessa Parli, director of research at Stanford University’s Institute for Human-Centered AI. “Is that the way we want to evaluate systems going forward? And if it’s not, what is the way?” A growing group of academics and AI researchers are making the case that the answer is to go smaller, trading sweeping ambition for an approach inspired by the social sciences. Specifically, they want to focus more on testing validity, which for quantitative social scientists refers to how well a given questionnaire measures what it’s claiming to measure—and, more fundamentally, whether what it is measuring has a coherent definition. That could cause trouble for benchmarks assessing hazily defined concepts like “reasoning” or “scientific knowledge”—and for developers aiming to reach the much-hyped goal of artificial general intelligence—but it would put the industry on firmer ground as it looks to prove the worth of individual models. “Taking validity seriously means asking folks in academia, industry, or wherever to show that their system does what they say it does,” says Abigail Jacobs, a University of Michigan professor who is a central figure in the new push for validity. “I think it points to a weakness in the AI world if they want to back off from showing that they can support their claim.” The limits of traditional testing If AI companies have been slow to respond to the growing failure of benchmarks, it’s partially because the test-scoring approach has been so effective for so long.  One of the biggest early successes of contemporary AI was the ImageNet challenge, a kind of antecedent to contemporary benchmarks. Released in 2010 as an open challenge to researchers, the database held more than 3 million images for AI systems to categorize into 1,000 different classes. Crucially, the test was completely agnostic to methods, and any successful algorithm quickly gained credibility regardless of how it worked. When an algorithm called AlexNet broke through in 2012, with a then unconventional form of GPU training, it became one of the foundational results of modern AI. Few would have guessed in advance that AlexNet’s convolutional neural nets would be the secret to unlocking image recognition—but after it scored well, no one dared dispute it. (One of AlexNet’s developers, Ilya Sutskever, would go on to cofound OpenAI.) A large part of what made this challenge so effective was that there was little practical difference between ImageNet’s object classification challenge and the actual process of asking a computer to recognize an image. Even if there were disputes about methods, no one doubted that the highest-scoring model would have an advantage when deployed in an actual image recognition system. But in the 12 years since, AI researchers have applied that same method-agnostic approach to increasingly general tasks. SWE-Bench is commonly used as a proxy for broader coding ability, while other exam-style benchmarks often stand in for reasoning ability. That broad scope makes it difficult to be rigorous about what a specific benchmark measures—which, in turn, makes it hard to use the findings responsibly.  Where things break down Anka Reuel, a PhD student who has been focusing on the benchmark problem as part of her research at Stanford, has become convinced the evaluation problem is the result of this push toward generality. “We’ve moved from task-specific models to general-purpose models,” Reuel says. “It’s not about a single task anymore but a whole bunch of tasks, so evaluation becomes harder.” Like the University of Michigan’s Jacobs, Reuel thinks “the main issue with benchmarks is validity, even more than the practical implementation,” noting: “That’s where a lot of things break down.” For a task as complicated as coding, for instance, it’s nearly impossible to incorporate every possible scenario into your problem set. As a result, it’s hard to gauge whether a model is scoring better because it’s more skilled at coding or because it has more effectively manipulated the problem set. And with so much pressure on developers to achieve record scores, shortcuts are hard to resist. For developers, the hope is that success on lots of specific benchmarks will add up to a generally capable model. But the techniques of agentic AI mean a single AI system can encompass a complex array of different models, making it hard to evaluate whether improvement on a specific task will lead to generalization. “There’s just many more knobs you can turn,” says Sayash Kapoor, a computer scientist at Princeton and a prominent critic of sloppy practices in the AI industry. “When it comes to agents, they have sort of given up on the best practices for evaluation.” In a paper from last July, Kapoor called out specific issues in how AI models were approaching the WebArena benchmark, designed by Carnegie Mellon University researchers in 2024 as a test of an AI agent’s ability to traverse the web. The benchmark consists of more than 800 tasks to be performed on a set of cloned websites mimicking Reddit, Wikipedia, and others. Kapoor and his team identified an apparent hack in the winning model, called STeP. STeP included specific instructions about how Reddit structures URLs, allowing STeP models to jump directly to a given user’s profile page (a frequent element of WebArena tasks). This shortcut wasn’t exactly cheating, but Kapoor sees it as “a serious misrepresentation of how well the agent would work had it seen the tasks in WebArena for the first time.” Because the technique was successful, though, a similar policy has since been adopted by OpenAI’s web agent Operator. (“Our evaluation setting is designed to assess how well an agent can solve tasks given some instruction about website structures and task execution,” an OpenAI representative said when reached for comment. “This approach is consistent with how others have used and reported results with WebArena.” STeP did not respond to a request for comment.) Further highlighting the problem with AI benchmarks, late last month Kapoor and a team of researchers wrote a paper that revealed significant problems in Chatbot Arena, the popular crowdsourced evaluation system. According to the paper, the leaderboard was being manipulated; many top foundation models were conducting undisclosed private testing and releasing their scores selectively. Today, even ImageNet itself, the mother of all benchmarks, has started to fall victim to validity problems. A 2023 study from researchers at the University of Washington and Google Research found that when ImageNet-winning algorithms were pitted against six real-world data sets, the architecture improvement “resulted in little to no progress,” suggesting that the external validity of the test had reached its limit. Going smaller For those who believe the main problem is validity, the best fix is reconnecting benchmarks to specific tasks. As Reuel puts it, AI developers “have to resort to these high-level benchmarks that are almost meaningless for downstream consumers, because the benchmark developers can’t anticipate the downstream task anymore.” So what if there was a way to help the downstream consumers identify this gap? In November 2024, Reuel launched a public ranking project called BetterBench, which rates benchmarks on dozens of different criteria, such as whether the code has been publicly documented. But validity is a central theme, with particular criteria challenging designers to spell out what capability their benchmark is testing and how it relates to the tasks that make up the benchmark. “You need to have a structural breakdown of the capabilities,” Reuel says. “What are the actual skills you care about, and how do you operationalize them into something we can measure?” The results are surprising. One of the highest-scoring benchmarks is also the oldest: the Arcade Learning Environment (ALE), established in 2013 as a way to test models’ ability to learn how to play a library of Atari 2600 games. One of the lowest-scoring is the Massive Multitask Language Understanding (MMLU) benchmark, a widely used test for general language skills; by the standards of BetterBench, the connection between the questions and the underlying skill was too poorly defined. BetterBench hasn’t meant much for the reputations of specific benchmarks, at least not yet; MMLU is still widely used, and ALE is still marginal. But the project has succeeded in pushing validity into the broader conversation about how to fix benchmarks. In April, Reuel quietly joined a new research group hosted by Hugging Face, the University of Edinburgh, and EleutherAI, where she’ll develop her ideas on validity and AI model evaluation with other figures in the field. (An official announcement is expected later this month.)  Irene Solaiman, Hugging Face’s head of global policy, says the group will focus on building valid benchmarks that go beyond measuring straightforward capabilities. “There’s just so much hunger for a good benchmark off the shelf that already works,” Solaiman says. “A lot of evaluations are trying to do too much.” Increasingly, the rest of the industry seems to agree. In a paper in March, researchers from Google, Microsoft, Anthropic, and others laid out a new framework for improving evaluations—with validity as the first step.  “AI evaluation science must,” the researchers argue, “move beyond coarse grained claims of ‘general intelligence’ towards more task-specific and real-world relevant measures of progress.”  Measuring the “squishy” things To help make this shift, some researchers are looking to the tools of social science. A February position paper argued that “evaluating GenAI systems is a social science measurement challenge,” specifically unpacking how the validity systems used in social measurements can be applied to AI benchmarking.  The authors, largely employed by Microsoft’s research branch but joined by academics from Stanford and the University of Michigan, point to the standards that social scientists use to measure contested concepts like ideology, democracy, and media bias. Applied to AI benchmarks, those same procedures could offer a way to measure concepts like “reasoning” and “math proficiency” without slipping into hazy generalizations. In the social science literature, it’s particularly important that metrics begin with a rigorous definition of the concept measured by the test. For instance, if the test is to measure how democratic a society is, it first needs to establish a definition for a “democratic society” and then establish questions that are relevant to that definition.  To apply this to a benchmark like SWE-Bench, designers would need to set aside the classic machine learning approach, which is to collect programming problems from GitHub and create a scheme to validate answers as true or false. Instead, they’d first need to define what the benchmark aims to measure (“ability to resolve flagged issues in software,” for instance), break that into subskills (different types of problems or types of program that the AI model can successfully process), and then finally assemble questions that accurately cover the different subskills. It’s a profound change from how AI researchers typically approach benchmarking—but for researchers like Jacobs, a coauthor on the February paper, that’s the whole point. “There’s a mismatch between what’s happening in the tech industry and these tools from social science,” she says. “We have decades and decades of thinking about how we want to measure these squishy things about humans.” Even though the idea has made a real impact in the research world, it’s been slow to influence the way AI companies are actually using benchmarks.  The last two months have seen new model releases from OpenAI, Anthropic, Google, and Meta, and all of them lean heavily on multiple-choice knowledge benchmarks like MMLU—the exact approach that validity researchers are trying to move past. After all, model releases are, for the most part, still about showing increases in general intelligence, and broad benchmarks continue to be used to back up those claims.  For some observers, that’s good enough. Benchmarks, Wharton professor Ethan Mollick says, are “bad measures of things, but also they’re what we’ve got.” He adds: “At the same time, the models are getting better. A lot of sins are forgiven by fast progress.” For now, the industry’s long-standing focus on artificial general intelligence seems to be crowding out a more focused validity-based approach. As long as AI models can keep growing in general intelligence, then specific applications don’t seem as compelling—even if that leaves practitioners relying on tools they no longer fully trust.  “This is the tightrope we’re walking,” says Hugging Face’s Solaiman. “It’s too easy to throw the system out, but evaluations are really helpful in understanding our models, even with these limitations.” Russell Brandom is a freelance writer covering artificial intelligence. He lives in Brooklyn with his wife and two cats. This story was supported by a grant from the Tarbell Center for AI Journalism.

It’s not easy being one of Silicon Valley’s favorite benchmarks. 

SWE-Bench (pronounced “swee bench”) launched in November 2024 to evaluate an AI model’s coding skill, using more than 2,000 real-world programming problems pulled from the public GitHub repositories of 12 different Python-based projects. 

In the months since then, it’s quickly become one of the most popular tests in AI. A SWE-Bench score has become a mainstay of major model releases from OpenAI, Anthropic, and Google—and outside of foundation models, the fine-tuners at AI firms are in constant competition to see who can rise above the pack. The top of the leaderboard is a pileup between three different fine tunings of Anthropic’s Claude Sonnet model and Amazon’s Q developer agent. Auto Code Rover—one of the Claude modifications—nabbed the number two spot in November, and was acquired just three months later.

Despite all the fervor, this isn’t exactly a truthful assessment of which model is “better.” As the benchmark has gained prominence, “you start to see that people really want that top spot,” says John Yang, a researcher on the team that developed SWE-Bench at Princeton University. As a result, entrants have begun to game the system—which is pushing many others to wonder whether there’s a better way to actually measure AI achievement.

Developers of these coding agents aren’t necessarily doing anything as straightforward cheating, but they’re crafting approaches that are too neatly tailored to the specifics of the benchmark. The initial SWE-Bench test set was limited to programs written in Python, which meant developers could gain an advantage by training their models exclusively on Python code. Soon, Yang noticed that high-scoring models would fail completely when tested on different programming languages—revealing an approach to the test that he describes as “gilded.”

“It looks nice and shiny at first glance, but then you try to run it on a different language and the whole thing just kind of falls apart,” Yang says. “At that point, you’re not designing a software engineering agent. You’re designing to make a SWE-Bench agent, which is much less interesting.”

The SWE-Bench issue is a symptom of a more sweeping—and complicated—problem in AI evaluation, and one that’s increasingly sparking heated debate: The benchmarks the industry uses to guide development are drifting further and further away from evaluating actual capabilities, calling their basic value into question. Making the situation worse, several benchmarks, most notably FrontierMath and Chatbot Arena, have recently come under heat for an alleged lack of transparency. Nevertheless, benchmarks still play a central role in model development, even if few experts are willing to take their results at face value. OpenAI cofounder Andrej Karpathy recently described the situation as “an evaluation crisis”: the industry has fewer trusted methods for measuring capabilities and no clear path to better ones. 

“Historically, benchmarks were the way we evaluated AI systems,” says Vanessa Parli, director of research at Stanford University’s Institute for Human-Centered AI. “Is that the way we want to evaluate systems going forward? And if it’s not, what is the way?”

A growing group of academics and AI researchers are making the case that the answer is to go smaller, trading sweeping ambition for an approach inspired by the social sciences. Specifically, they want to focus more on testing validity, which for quantitative social scientists refers to how well a given questionnaire measures what it’s claiming to measure—and, more fundamentally, whether what it is measuring has a coherent definition. That could cause trouble for benchmarks assessing hazily defined concepts like “reasoning” or “scientific knowledge”—and for developers aiming to reach the much–hyped goal of artificial general intelligence—but it would put the industry on firmer ground as it looks to prove the worth of individual models.

“Taking validity seriously means asking folks in academia, industry, or wherever to show that their system does what they say it does,” says Abigail Jacobs, a University of Michigan professor who is a central figure in the new push for validity. “I think it points to a weakness in the AI world if they want to back off from showing that they can support their claim.”

The limits of traditional testing

If AI companies have been slow to respond to the growing failure of benchmarks, it’s partially because the test-scoring approach has been so effective for so long. 

One of the biggest early successes of contemporary AI was the ImageNet challenge, a kind of antecedent to contemporary benchmarks. Released in 2010 as an open challenge to researchers, the database held more than 3 million images for AI systems to categorize into 1,000 different classes.

Crucially, the test was completely agnostic to methods, and any successful algorithm quickly gained credibility regardless of how it worked. When an algorithm called AlexNet broke through in 2012, with a then unconventional form of GPU training, it became one of the foundational results of modern AI. Few would have guessed in advance that AlexNet’s convolutional neural nets would be the secret to unlocking image recognition—but after it scored well, no one dared dispute it. (One of AlexNet’s developers, Ilya Sutskever, would go on to cofound OpenAI.)

A large part of what made this challenge so effective was that there was little practical difference between ImageNet’s object classification challenge and the actual process of asking a computer to recognize an image. Even if there were disputes about methods, no one doubted that the highest-scoring model would have an advantage when deployed in an actual image recognition system.

But in the 12 years since, AI researchers have applied that same method-agnostic approach to increasingly general tasks. SWE-Bench is commonly used as a proxy for broader coding ability, while other exam-style benchmarks often stand in for reasoning ability. That broad scope makes it difficult to be rigorous about what a specific benchmark measures—which, in turn, makes it hard to use the findings responsibly. 

Where things break down

Anka Reuel, a PhD student who has been focusing on the benchmark problem as part of her research at Stanford, has become convinced the evaluation problem is the result of this push toward generality. “We’ve moved from task-specific models to general-purpose models,” Reuel says. “It’s not about a single task anymore but a whole bunch of tasks, so evaluation becomes harder.”

Like the University of Michigan’s Jacobs, Reuel thinks “the main issue with benchmarks is validity, even more than the practical implementation,” noting: “That’s where a lot of things break down.” For a task as complicated as coding, for instance, it’s nearly impossible to incorporate every possible scenario into your problem set. As a result, it’s hard to gauge whether a model is scoring better because it’s more skilled at coding or because it has more effectively manipulated the problem set. And with so much pressure on developers to achieve record scores, shortcuts are hard to resist.

For developers, the hope is that success on lots of specific benchmarks will add up to a generally capable model. But the techniques of agentic AI mean a single AI system can encompass a complex array of different models, making it hard to evaluate whether improvement on a specific task will lead to generalization. “There’s just many more knobs you can turn,” says Sayash Kapoor, a computer scientist at Princeton and a prominent critic of sloppy practices in the AI industry. “When it comes to agents, they have sort of given up on the best practices for evaluation.”

In a paper from last July, Kapoor called out specific issues in how AI models were approaching the WebArena benchmark, designed by Carnegie Mellon University researchers in 2024 as a test of an AI agent’s ability to traverse the web. The benchmark consists of more than 800 tasks to be performed on a set of cloned websites mimicking Reddit, Wikipedia, and others. Kapoor and his team identified an apparent hack in the winning model, called STeP. STeP included specific instructions about how Reddit structures URLs, allowing STeP models to jump directly to a given user’s profile page (a frequent element of WebArena tasks).

This shortcut wasn’t exactly cheating, but Kapoor sees it as “a serious misrepresentation of how well the agent would work had it seen the tasks in WebArena for the first time.” Because the technique was successful, though, a similar policy has since been adopted by OpenAI’s web agent Operator. (“Our evaluation setting is designed to assess how well an agent can solve tasks given some instruction about website structures and task execution,” an OpenAI representative said when reached for comment. “This approach is consistent with how others have used and reported results with WebArena.” STeP did not respond to a request for comment.)

Further highlighting the problem with AI benchmarks, late last month Kapoor and a team of researchers wrote a paper that revealed significant problems in Chatbot Arena, the popular crowdsourced evaluation system. According to the paper, the leaderboard was being manipulated; many top foundation models were conducting undisclosed private testing and releasing their scores selectively.

Today, even ImageNet itself, the mother of all benchmarks, has started to fall victim to validity problems. A 2023 study from researchers at the University of Washington and Google Research found that when ImageNet-winning algorithms were pitted against six real-world data sets, the architecture improvement “resulted in little to no progress,” suggesting that the external validity of the test had reached its limit.

Going smaller

For those who believe the main problem is validity, the best fix is reconnecting benchmarks to specific tasks. As Reuel puts it, AI developers “have to resort to these high-level benchmarks that are almost meaningless for downstream consumers, because the benchmark developers can’t anticipate the downstream task anymore.” So what if there was a way to help the downstream consumers identify this gap?

In November 2024, Reuel launched a public ranking project called BetterBench, which rates benchmarks on dozens of different criteria, such as whether the code has been publicly documented. But validity is a central theme, with particular criteria challenging designers to spell out what capability their benchmark is testing and how it relates to the tasks that make up the benchmark.

“You need to have a structural breakdown of the capabilities,” Reuel says. “What are the actual skills you care about, and how do you operationalize them into something we can measure?”

The results are surprising. One of the highest-scoring benchmarks is also the oldest: the Arcade Learning Environment (ALE), established in 2013 as a way to test models’ ability to learn how to play a library of Atari 2600 games. One of the lowest-scoring is the Massive Multitask Language Understanding (MMLU) benchmark, a widely used test for general language skills; by the standards of BetterBench, the connection between the questions and the underlying skill was too poorly defined.

BetterBench hasn’t meant much for the reputations of specific benchmarks, at least not yet; MMLU is still widely used, and ALE is still marginal. But the project has succeeded in pushing validity into the broader conversation about how to fix benchmarks. In April, Reuel quietly joined a new research group hosted by Hugging Face, the University of Edinburgh, and EleutherAI, where she’ll develop her ideas on validity and AI model evaluation with other figures in the field. (An official announcement is expected later this month.) 

Irene Solaiman, Hugging Face’s head of global policy, says the group will focus on building valid benchmarks that go beyond measuring straightforward capabilities. “There’s just so much hunger for a good benchmark off the shelf that already works,” Solaiman says. “A lot of evaluations are trying to do too much.”

Increasingly, the rest of the industry seems to agree. In a paper in March, researchers from Google, Microsoft, Anthropic, and others laid out a new framework for improving evaluations—with validity as the first step. 

“AI evaluation science must,” the researchers argue, “move beyond coarse grained claims of ‘general intelligence’ towards more task-specific and real-world relevant measures of progress.” 

Measuring the “squishy” things

To help make this shift, some researchers are looking to the tools of social science. A February position paper argued that “evaluating GenAI systems is a social science measurement challenge,” specifically unpacking how the validity systems used in social measurements can be applied to AI benchmarking. 

The authors, largely employed by Microsoft’s research branch but joined by academics from Stanford and the University of Michigan, point to the standards that social scientists use to measure contested concepts like ideology, democracy, and media bias. Applied to AI benchmarks, those same procedures could offer a way to measure concepts like “reasoning” and “math proficiency” without slipping into hazy generalizations.

In the social science literature, it’s particularly important that metrics begin with a rigorous definition of the concept measured by the test. For instance, if the test is to measure how democratic a society is, it first needs to establish a definition for a “democratic society” and then establish questions that are relevant to that definition. 

To apply this to a benchmark like SWE-Bench, designers would need to set aside the classic machine learning approach, which is to collect programming problems from GitHub and create a scheme to validate answers as true or false. Instead, they’d first need to define what the benchmark aims to measure (“ability to resolve flagged issues in software,” for instance), break that into subskills (different types of problems or types of program that the AI model can successfully process), and then finally assemble questions that accurately cover the different subskills.

It’s a profound change from how AI researchers typically approach benchmarking—but for researchers like Jacobs, a coauthor on the February paper, that’s the whole point. “There’s a mismatch between what’s happening in the tech industry and these tools from social science,” she says. “We have decades and decades of thinking about how we want to measure these squishy things about humans.”

Even though the idea has made a real impact in the research world, it’s been slow to influence the way AI companies are actually using benchmarks. 

The last two months have seen new model releases from OpenAI, Anthropic, Google, and Meta, and all of them lean heavily on multiple-choice knowledge benchmarks like MMLU—the exact approach that validity researchers are trying to move past. After all, model releases are, for the most part, still about showing increases in general intelligence, and broad benchmarks continue to be used to back up those claims. 

For some observers, that’s good enough. Benchmarks, Wharton professor Ethan Mollick says, are “bad measures of things, but also they’re what we’ve got.” He adds: “At the same time, the models are getting better. A lot of sins are forgiven by fast progress.”

For now, the industry’s long-standing focus on artificial general intelligence seems to be crowding out a more focused validity-based approach. As long as AI models can keep growing in general intelligence, then specific applications don’t seem as compelling—even if that leaves practitioners relying on tools they no longer fully trust. 

“This is the tightrope we’re walking,” says Hugging Face’s Solaiman. “It’s too easy to throw the system out, but evaluations are really helpful in understanding our models, even with these limitations.”

Russell Brandom is a freelance writer covering artificial intelligence. He lives in Brooklyn with his wife and two cats.

This story was supported by a grant from the Tarbell Center for AI Journalism.

Shape
Shape
Stay Ahead

Explore More Insights

More Articles

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

Nutanix expands beyond HCI

The Pure Storage integration will also be supported within Cisco’s FlashStack offering, creating a “FlashStack with Nutanix” solution with storage provided by Pure, networking capabilities as well as UCS servers from Cisco, and then the common Nutanix Cloud Platform. Cloud Native AOS: Breaking free from hypervisors Another sharp departure from

Read More »

IBM introduces new generation of LinuxOne AI mainframe

In addition to generative AI applications, new multiple model AI approaches are engineered to enhance prediction and accuracy in many industry use cases like advanced fraud detection, image processing and retail automation, according to IBM. LinuxONE Emperor 5 also comes with advanced security features specifically designed for the AI threat

Read More »

PJM, utilities urge FERC to dismiss call for colocation settlement talks

The Federal Energy Regulatory Commission should reject a call for a 90-day pause in its deliberations over the PJM Interconnection’s rules for colocating data centers at power plants, according to PJM, major utilities and other organizations. “The national interest will be best served by a quick dismissal of this proceeding, and a ruling that the existing PJM Tariff remains just and reasonable,” PJM transmission owners said in a Wednesday filing urging FERC to dismiss a call for stakeholder settlement talks. “Rather than fighting about a wish list of new rules, the parties will then instead begin to focus on obtaining service under the rules in place today.” The transmission owners include utility companies such as American Electric Power, Dominion Energy, Duke Energy, Exelon, FirstEnergy and PPL Electric. “The record is clear — no matter how connected to the PJM transmission system, large loads pose both a safety and a reliability concern,” the utilities said. “It is unrealistic to ask the [transmission owners] to accede to these demands in the context of settlement procedures while those questions remain unresolved.” PJM also wants FERC to ignore the call for settlement discussions that was made in late April by the Electric Power Supply Association, the PJM Power Providers Group, Calpine, Cogentrix Energy Power Management, Constellation Energy Generation and LS Power Development. “The Commission should not pause its work on offering the industry guidance on a path forward for co-location arrangements,” PJM said in a Monday filing. The call for settlement talks lacks broad stakeholder support, PJM said, noting it is holding a workshop on “large load” issues on Friday. American Municipal Power, a wholesale power provider for public power utilities, and Northern Virginia Electric Cooperative and Northeastern Rural Electric Membership Corp. also oppose holding settlement talks. Beside the power generators and trade organizations,

Read More »

IRA’s fate unclear as Republicans look to finance megabill

Dive Brief: The Inflation Reduction Act – which passed in 2022 without any Republican support and is anticipated to cost taxpayers between $780 billion and $2 trillion over its first ten years – is likely to be targeted for cuts as the Republican-controlled Congress aims to cut spending by $2 trillion in order to cut taxes by $4.5 trillion. However, certain provisions of the IRA have won support with Republican lawmakers, setting up likely disagreements over cuts in the budget reconciliation process. That process is already expected to be “very contentious,” said Harry Godfrey, who leads Advanced Energy United’s federal investment and manufacturing working group. “[House Ways and Means Committee Chairman Jason Smith] and the Ways and Means Republicans will need revenue and will be seeking it, and have been saying all along that the IRA is an area they’re going to look at,” said Ryan Abraham, a principal with Ernst & Young’s Washington Council advisory practice. “But obviously there are some concerns among some members.” Dive Insight: Abraham noted the May 1 letter sent by 26 House Republicans to Chairman Smith, advocating for the preservation of the IRA’s 45U, 45Y, and 48E tax credits. The letter advocates on behalf of nuclear power specifically, “[urging Smith] to maintain federal investment in the existing nuclear energy fleet while accelerating deployment of the next generation of nuclear power technologies.” The 45U credit is the IRA’s zero-emission nuclear power production credit, while 48E and 45Y are technology-neutral credits, which were targeted in legislation introduced in April by Rep. Julie Fedorchak, R-N.D. “There’s a lot of concern that some of the bonus items that have been created in the IRA, like direct pay and transferability, which were also in that Fedorchak bill, could also get targeted,” Abraham said. However, he said, “Chairman Smith is aware

Read More »

Energy Department Aligns Award Criteria for For-profit, Non-profit Organizations, and State and Local Governments, Saving $935 Million Annually

WASHINGTON — The U.S. Department of Energy (DOE) today announced three new policy actions that are projected to save more than $935 million annually for the American taxpayer, while expanding American innovation and scientific research. In three new policy memorandums, the DOE announced that it will follow best practices used by fellow grant providers and limit “indirect costs” of DOE funding to 10% for state and local governments, 15% for non-profit organizations, and 15% for for-profit companies. The Energy Department expects to generate over $935 million in annual cost savings for the American people, delivering on President Trump’s commitment to bring greater transparency and efficiency to federal government spending. Estimated savings are based on applying the new policies to 2024 fiscal year spending. “This action ensures that Department of Energy funds are supporting state, local, for-profit and non-profit initiatives that make energy more affordable and secure for Americans, not funding administrative costs,” U.S. Secretary of Energy Chris Wright said. “By aligning our policy on indirect costs with industry standards, we are increasing accountability of taxpayer dollars and ensuring the American people are getting the greatest value possible from these DOE programs.” These policy actions follow an announcement made in April to limit financial support of “indirect costs” of DOE research funding at colleges and universities to 15%, saving an estimated additional $405 million annually. By enacting indirect cost limits, the Department aligns its practices with those common for other grant providers. The full three memorandums are available below: POLICY FLASH SUBJECT: Adjusting Department of Energy Financial Assistance Policy for State and Local Governments’ Financial Assistance Awards BACKGROUND: Pursuant to 5 U.S.C. 553(a)(2), the Department of Energy (“Department”) is updating its policy with respect to Department financial assistance funding awarded to state and local governments. Through its financial assistance programs (which include grants and cooperative agreements),

Read More »

BP-Shell Megadeal Would Create European Rival to Exxon Mobil

If Shell Plc were to acquire BP Plc, it would be among the largest deals in European history, creating for the first time a European oil major that could challenge industry leaders Exxon Mobil Corp. and Chevron Corp. The circumstances for such a takeover might not be auspicious for BP — its shares have lost almost a third of their value in the last year and investors are unconvinced by the company’s turnaround plan — but the deal would be transformative for Shell. Since Bloomberg reported that Shell has been studying the merits of an acquisition — something both companies have so far declined to comment on — analysts have been gaming out both the upsides and downsides. Here are some of their conclusions: Oil Behemoth Combining the two London-based oil majors would create an energy giant with upstream oil and gas production of nearly 5 million barrels of oil equivalent a day, according to UBS Group. This would be an 85% increase from Shell’s current output of about 2.7 million barrels a day, making the combined company the largest investor-owned oil and gas producer. Exxon produced an average of 4.6 million barrels of oil equivalent a day in the first quarter, while Chevron pumped 3.4 million a day.  Adding BPX, the Denver-based US shale unit, to Shell’s portfolio would go some way to correct the company’s decision to sell its business in the Permian basin to ConocoPhillips in 2021, just before the industry experienced a major boom. Shell’s then-upstream director, Wael Sawan, is now chief executive officer.  LNG Dominance  Shell is already the world’s largest seller of liquefied natural gas, and acquiring BP would propel the business “to new heights,” said RBC. The combined companies’ LNG sales would jump to more than 90 million tonnes a year, representing more than 20%

Read More »

Devon Sells Stake in Permian Gas Pipeline

Devon Energy Corp. and Ridgemont Equity Partners are divesting their interests in a natural gas pipeline serving the Permian Basin with a capacity of about 2.5 billion cubic feet a day. Put onstream November 2024, the Matterhorn Express Pipeline is fully contracted. The system consists of a 510-mile mainline and associated compression that carries gas from the Waha area to Wharton, Texas. It also has delivery capabilities for Katy, Texas, as well as laterals in the Midland Basin, according to operator WhiteWater Development LLC. The stake sales by Devon and Ridgemont will result in WhiteWater, MPLX LP and Enbridge Inc. owning 65 percent, 10 percent and 10 percent of the pipeline respectively, according to a joint statement. “WhiteWater’s equity interest in the Matterhorn Express Pipeline will be jointly backed by [infrastructure investors] FIC and I Squared, and WhiteWater will continue to operate the Matterhorn Express Pipeline”, said the statement posted on WhiteWater’s website. The parties expect to complete the transactions in the second quarter. MPLX separately said its side of the transaction consists of a 5 percent acquisition that would raise its stake to 10 percent for a purchase price of $151 million. Devon separately said it will receive about $375 million in proceeds. “Proceeds from the divestiture will be used to further strengthen the company’s investment-grade financial position”, it said in its quarterly report. “The monetization of Devon’s equity ownership will not change the terms or conditions of the company’s secured capacity on the pipeline”. In a separate transaction, Devon said in the report its partnership with BPX for the Blackhawk field in the Eagle Ford shale had been dissolved. That dissolution has resulted in Devon owning around 46,000 net acres with a stake of 95 percent and operatorship. For the first quarter (Q1), Devon reported $494 million in net profit.

Read More »

UK and Norway sign green energy industrial partnership amid difficult week for Labour

The UK and Norway have formalised a green industrial partnership focused on the energy transition amid a week of setbacks for Labour’s clean power goals. UK energy secretary Ed Miliband travelled to Oslo to sign the deal alongside Norway’s trade and industry minister Cecilie Myrseth and energy minister Terje Aasland. The two governments first announced the partnership in December last year, stating their ambition to cooperate on carbon capture and storage (CCS) projects. Alongside CCS, the partnership will also focus on hydrogen, offshore wind, sustainable energy systems, green supply chains and joint efforts on skills development. The UK government said it will also feature collaboration on the protection of UK and Norwegian offshore infrastructure and reducing barriers to cross-border CO2 storage. An independent report found that closer cooperation between the UK and Norway could create up to 51,000 jobs and add up to £36 billion to the UK economy. © Leon Neal/PA WirePrime Minister Sir Keir Starmer and Norwegian Prime Minister Jonas Gahr Store tour the Northern Lights CCUS Plant CO2 transport and storage facility in Bergen, during a trip to Norway. Image: Leon Neal/PA Wire The Norwegian government said the UK is the country’s second-largest trading partner after the EU, with total trade in 2024 coming to approximately £38bn. More than 200 Norwegian firms operate in the UK, including Equinor, Statkraft Vårgrønn, which altogether employ over 15,000 people. In a statement, Miliband said that together, Norway and the UK can “take advantage of the opportunities ahead in the North Sea”. “Energy security is national security – and only by working with key partners like Norway can we accelerate clean power that we control, getting us off the rollercoaster of fossil fuels in these unstable times,” Miliband said. Meanwhile, Aasland said the two countries “have a unique relationship in the

Read More »

Tech CEOs warn Senate: Outdated US power grid threatens AI ambitions

The implications are clear: without dramatic improvements to the US energy infrastructure, the nation’s AI ambitions could be significantly constrained by simple physical limitations – the inability to power the massive computing clusters necessary for advanced AI development and deployment. Streamlining permitting processes The tech executives have offered specific recommendations to address these challenges, with several focusing on the need to dramatically accelerate permitting processes for both energy generation and the transmission infrastructure needed to deliver that power to AI facilities, the report added. Intrator specifically called for efforts “to streamline the permitting process to enable the addition of new sources of generation and the transmission infrastructure to deliver it,” noting that current regulatory frameworks were not designed with the urgent timelines of the AI race in mind. This acceleration would help technology companies build and power the massive data centers needed for AI training and inference, which require enormous amounts of electricity delivered reliably and consistently. Beyond the cloud: bringing AI to everyday devices While much of the testimony focused on large-scale infrastructure needs, AMD CEO Lisa Su emphasized that true AI leadership requires “rapidly building data centers at scale and powering them with reliable, affordable, and clean energy sources.” Su also highlighted the importance of democratizing access to AI technologies: “Moving faster also means moving AI beyond the cloud. To ensure every American benefits, AI must be built into the devices we use every day and made as accessible and dependable as electricity.”

Read More »

Networking errors pose threat to data center reliability

Still, IT and networking issues increased in 2024, according to Uptime Institute. The analysis attributed the rise in outages due to increased IT and network complexity, specifically, change management and misconfigurations. “Particularly with distributed services, cloud services, we find that cascading failures often occur when networking equipment is replicated across an entire network,” Lawrence explained. “Sometimes the failure of one forces traffic to move in one direction, overloading capacity at another data center.” The most common causes of major network-related outages were cited as: Configuration/change management failure: 50% Third-party network provider failure: 34% Hardware failure: 31% Firmware/software error: 26% Line breakages: 17% Malicious cyberattack: 17% Network overload/congestion failure: 13% Corrupted firewall/routing tables issues: 8% Weather-related incident: 7% Configuration/change management issues also attributed for 62% of the most common causes of major IT system-/software-related outages. Change-related disruptions consistently are responsible for software-related outages. Human error continues to be one of the “most persistent challenges in data center operations,” according to Uptime’s analysis. The report found that the biggest cause of these failures is data center staff failing to follow established procedures, which has increased by about 10 percentage points compared to 2023. “These are things that were 100% under our control. I mean, we can’t control when the UPS module fails because it was either poorly manufactured, it had a flaw, or something else. This is 100% under our control,” Brown said. The most common causes of major human error-related outages were reported as:

Read More »

Liquid cooling technologies: reducing data center environmental impact

“Highly optimized cold-plate or one-phase immersion cooling technologies can perform on par with two-phase immersion, making all three liquid-cooling technologies desirable options,” the researchers wrote. Factors to consider There are numerous factors to consider when adopting liquid cooling technologies, according to Microsoft’s researchers. First, they advise performing a full environmental, health, and safety analysis, and end-to-end life cycle impact analysis. “Analyzing the full data center ecosystem to include systems interactions across software, chip, server, rack, tank, and cooling fluids allows decision makers to understand where savings in environmental impacts can be made,” they wrote. It is also important to engage with fluid vendors and regulators early, to understand chemical composition, disposal methods, and compliance risks. And associated socioeconomic, community, and business impacts are equally critical to assess. More specific environmental considerations include ozone depletion and global warming potential; the researchers emphasized that operators should only use fluids with low to zero ozone depletion potential (ODP) values, and not hydrofluorocarbons or carbon dioxide. It is also critical to analyze a fluid’s viscosity (thickness or stickiness), flammability, and overall volatility. And operators should only use fluids with minimal bioaccumulation (the buildup of chemicals in lifeforms, typically in fish) and terrestrial and aquatic toxicity. Finally, once up and running, data center operators should monitor server lifespan and failure rates, tracking performance uptime and adjusting IT refresh rates accordingly.

Read More »

Cisco unveils prototype quantum networking chip

Clock synchronization allows for coordinated time-dependent communications between end points that might be cloud databases or in large global databases that could be sitting across the country or across the world, he said. “We saw recently when we were visiting Lawrence Berkeley Labs where they have all of these data sources such as radio telescopes, optical telescopes, satellites, the James Webb platform. All of these end points are taking snapshots of a piece of space, and they need to synchronize those snapshots to the picosecond level, because you want to detect things like meteorites, something that is moving faster than the rotational speed of planet Earth. So the only way you can detect that quickly is if you synchronize these snapshots at the picosecond level,” Pandey said. For security use cases, the chip can ensure that if an eavesdropper tries to intercept the quantum signals carrying the key, they will likely disturb the state of the qubits, and this disturbance can be detected by the legitimate communicating parties and the link will be dropped, protecting the sender’s data. This feature is typically implemented in a Quantum Key Distribution system. Location information can serve as a critical credential for systems to authenticate control access, Pandey said. The prototype quantum entanglement chip is just part of the research Cisco is doing to accelerate practical quantum computing and the development of future quantum data centers.  The quantum data center that Cisco envisions would have the capability to execute numerous quantum circuits, feature dynamic network interconnection, and utilize various entanglement generation protocols. The idea is to build a network connecting a large number of smaller processors in a controlled environment, the data center warehouse, and provide them as a service to a larger user base, according to Cisco.  The challenges for quantum data center network fabric

Read More »

Zyxel launches 100GbE switch for enterprise networks

Port specifications include: 48 SFP28 ports supporting dual-rate 10GbE/25GbE connectivity 8 QSFP28 ports supporting 100GbE connections Console port for direct management access Layer 3 routing capabilities include static routing with support for access control lists (ACLs) and VLAN segmentation. The switch implements IEEE 802.1Q VLAN tagging, port isolation, and port mirroring for traffic analysis. For link aggregation, the switch supports IEEE 802.3ad for increased throughput and redundancy between switches or servers. Target applications and use cases The CX4800-56F targets multiple deployment scenarios where high-capacity backbone connectivity and flexible port configurations are required. “This will be for service providers initially or large deployments where they need a high capacity backbone to deliver a primarily 10G access layer to the end point,” explains Nguyen. “Now with Wi-Fi 7, more 10G/25G capable POE switches are being powered up and need interconnectivity without the bottleneck. We see this for data centers, campus, MDU (Multi-Dwelling Unit) buildings or community deployments.” Management is handled through Zyxel’s NebulaFlex Pro technology, which supports both standalone configuration and cloud management via the Nebula Control Center (NCC). The switch includes a one-year professional pack license providing IGMP technology and network analytics features. The SFP28 ports maintain backward compatibility between 10G and 25G standards, enabling phased migration paths for organizations transitioning between these speeds.

Read More »

Engineers rush to master new skills for AI-driven data centers

According to the Uptime Institute survey, 57% of data centers are increasing salary spending. Data center job roles that saw the highest increases were in operations management – 49% of data center operators said they saw highest increases in this category – followed by junior and mid-level operations staff at 45%, and senior management and strategy at 35%. Other job categories that saw salary growth were electrical, at 32% and mechanical, at 23%. Organizations are also paying premiums on top of salaries for particular skills and certifications. Foote Partners tracks pay premiums for more than 1,300 certified and non-certified skills for IT jobs in general. The company doesn’t segment the data based on whether the jobs themselves are data center jobs, but it does track 60 skills and certifications related to data center management, including skills such as storage area networking, LAN, and AIOps, and 24 data center-related certificates from Cisco, Juniper, VMware and other organizations. “Five of the eight data center-related skills recording market value gains in cash pay premiums in the last twelve months are all AI-related skills,” says David Foote, chief analyst at Foote Partners. “In fact, they are all among the highest-paying skills for all 723 non-certified skills we report.” These skills bring in 16% to 22% of base salary, he says. AIOps, for example, saw an 11% increase in market value over the past year, now bringing in a premium of 20% over base salary, according to Foote data. MLOps now brings in a 22% premium. “Again, these AI skills have many uses of which the data center is only one,” Foote adds. The percentage increase in the specific subset of these skills in data centers jobs may vary. The Uptime Institute survey suggests that the higher pay is motivating workers to stay in the

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 »
More Articles