Stay Ahead, Stay ONMINE

OpenAI overrode concerns of expert testers to release sycophantic GPT-4o

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More It’s been a bit of a topsy-turvy week for the number one generative AI company in terms of users. OpenAI, creator of ChatGPT, released and then withdrew an updated version of the underlying multimodal (text, image, […]

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More


It’s been a bit of a topsy-turvy week for the number one generative AI company in terms of users.

OpenAI, creator of ChatGPT, released and then withdrew an updated version of the underlying multimodal (text, image, audio) large language model (LLM) that ChatGPT is hooked up to by default, GPT-4o, due to it being too sycophantic to users. The company recently reported at least 500 million active weekly users of the hit web service.

A quick primer on the terrible, no good, sycophantic GPT-4o update

OpenAI began updating GPT-4o to a newer model it hoped would be more well-received by users on April 24th, completed the updated by April 25th, then, five days later, rolled it back on April 29, after days of mounting complaints of users across social media — mainly on X and Reddit.

The complaints varied in intensity and in specifics, but all generally coalesced around the fact that GPT-4o appeared to be responding to user queries with undue flattery, support for misguided, incorrect and downright harmful ideas, and “glazing” or praising the user to an excessive degree when it wasn’t actually specifically requested, much less warranted.

In examples screenshotted and posted by users, ChatGPT powered by that sycophantic, updated GPT-4o model had praised and endorsed a business idea for literal “shit on a stick,” applauded a user’s sample text of schizophrenic delusional isolation, and even allegedly supported plans to commit terrorism.

Users including top AI researchers and even a former OpenAI interim CEO said they were concerned that an AI model’s unabashed cheerleading for these types of terrible user prompts was more than simply annoying or inappropriate — that it could cause actual harm to users who mistakenly believed the AI and felt emboldened by its support for their worst ideas and impulses. It rose to the level of an AI safety issue.

OpenAI then released a blog post describing what went wrong — “we focused too much on short-term feedback, and did not fully account for how users’ interactions with ChatGPT evolve over time. As a result, GPT‑4o skewed towards responses that were overly supportive but disingenuous” — and the steps the company was taking to address the issues. OpenAI’s Head of Model Behavior Joanne Jang also participated in a Reddit “Ask me anything” or AMA forum answering text posts from users and revealed further information about the company’s approach to GPT-4o and how it ended up with an excessively sycophantic model, including not “bak[ing] in enough nuance,” as to how it was incorporating user feedback such as “thumbs up” actions made by users in response to model outputs they liked.

Now today, OpenAI has released a blog post with even more information about how the sycophantic GPT-4o update happened — credited not to any particular author, but to “OpenAI.”

CEO and co-founder Sam Altman also posted a link to the blog post on X, saying: “we missed the mark with last week’s GPT-4o update. what happened, what we learned, and some things we will do differently in the future.”

What the new OpenAI blog post reveals about how and why GPT-4o turned so sycophantic

To me, a daily user of ChatGPT including the 4o model, the most striking admission from OpenAI’s new blog post about the sycophancy update is how the company appears to reveal that it did receive concerns about the model prior to release from a small group of “expert testers,” but that it seemingly overrode those in favor of a broader enthusiastic response from a wider group of more general users.

As the company writes (emphasis mine):

“While we’ve had discussions about risks related to sycophancy in GPT‑4o for a while, sycophancy wasn’t explicitly flagged as part of our internal hands-on testing, as some of our expert testers were more concerned about the change in the model’s tone and style. Nevertheless, some expert testers had indicated that the model behavior “felt” slightly off…

We then had a decision to make: should we withhold deploying this update despite positive evaluations and A/B test results, based only on the subjective flags of the expert testers? In the end, we decided to launch the model due to the positive signals from the users who tried out the model.

Unfortunately, this was the wrong call. We build these models for our users and while user feedback is critical to our decisions, it’s ultimately our responsibility to interpret that feedback correctly.”

This seems to me like a big mistake. Why even have expert testers if you’re not going to weight their expertise higher than the masses of the crowd? I asked Altman about this choice on X but he has yet to respond.

Not all ‘reward signals’ are equal

OpenAI’s new post-mortem blog post also reveals more specifics about how the company trains and updates new versions of existing models, and how human feedback alters the model qualities, character, and “personality.” As the company writes:

“Since launching GPT‑4o in ChatGPT last May, we’ve released five major updates focused on changes to personality and helpfulness. Each update involves new post-training, and often many minor adjustments to the model training process are independently tested and then combined into a single updated model which is then evaluated for launch.

To post-train models, we take a pre-trained base model, do supervised fine-tuning on a broad set of ideal responses written by humans or existing models, and then run reinforcement learning with reward signals from a variety of sources.

During reinforcement learning, we present the language model with a prompt and ask it to write responses. We then rate its response according to the reward signals, and update the language model to make it more likely to produce higher-rated responses and less likely to produce lower-rated responses.

Clearly, the “reward signals” used by OpenAI during post-training have an enormous impact on the resulting model behavior, and as the company admitted earlier when it overweighted “thumbs up” responses from ChatGPT users to its outputs, this signal may not be the best one to use equally with others when determining how the model learns to communicate and what kinds of responses it should be serving up. OpenAI admits this outright in the next paragraph of its post, writing:

“Defining the correct set of reward signals is a difficult question, and we take many things into account: are the answers correct, are they helpful, are they in line with our Model Spec⁠, are they safe, do users like them, and so on. Having better and more comprehensive reward signals produces better models for ChatGPT, so we’re always experimenting with new signals, but each one has its quirks.”

Indeed, OpenAI also reveals the “thumbs up” reward signal was a new one used alongside other reward signals in this particular update.

“the update introduced an additional reward signal based on user feedback—thumbs-up and thumbs-down data from ChatGPT. This signal is often useful; a thumbs-down usually means something went wrong.”

Yet critically, the company doesn’t blame the new “thumbs up” data outright for the model’s failure and ostentatious cheerleading behaviors. Instead, OpenAI’s blog post says it was this combined with a variety of other new and older reward signals, led to the problems: “…we had candidate improvements to better incorporate user feedback, memory, and fresher data, among others. Our early assessment is that each of these changes, which had looked beneficial individually, may have played a part in tipping the scales on sycophancy when combined.”

Reacting to this blog post, Andrew Mayne, a former member of the OpenAI technical staff now working at AI consulting firm Interdimensional, wrote on X of another example of how subtle changes in reward incentives and model guidelines can impact model performance quite dramatically:

Early on at OpenAI, I had a disagreement with a colleague (who is now a founder of another lab) over using the word “polite” in a prompt example I wrote.

They argued “polite” was politically incorrect and wanted to swap it for “helpful.”

I pointed out that focusing only on helpfulness can make a model overly compliant—so compliant, in fact, that it can be steered into sexual content within a few turns.

After I demonstrated that risk with a simple exchange, the prompt kept “polite.”

These models are weird.

How OpenAI plans to improve its model testing processes going forward

The company lists six process improvements for how to avoid similar undesirable and less-than-ideal model behavior in the future, but to me the most important is this:

“We’ll adjust our safety review process to formally consider behavior issues—such as hallucination, deception, reliability, and personality—as blocking concerns. Even if these issues aren’t perfectly quantifiable today, we commit to blocking launches based on proxy measurements or qualitative signals, even when metrics like A/B testing look good.”

In other words — despite how important data, especially quantitative data, is to the fields of machine learning and artificial intelligence — OpenAI recognizes that this alone can’t and should not be the only means by which a model’s performance is judged.

While many users providing a “thumbs up” could signal a type of desirable behavior in the short term, the long term implications for how the AI model responds and where those behaviors take it and its users, could ultimately lead to a very dark, distressing, destructive, and undesirable place. More is not always better — especially when you are constraining the “more” to a few domains of signals.

It’s not enough to say that the model passed all of the tests or received a number of positive responses from users — the expertise of trained power users and their qualitative feedback that something “seemed off” about the model, even if they couldn’t fully express why, should carry much more weight than OpenAI was allocating previously.

Let’s hope the company — and the entire field — learns from this incident and integrates the lessons going forward.

Broader takeaways and considerations for enterprise decision-makers

Speaking perhaps more theoretically, for myself, it also indicates why expertise is so important — and specifically, expertise in fields beyond and outside of the one you’re optimizing for (in this case, machine learning and AI). It’s the diversity of expertise that allows us as a species to achieve new advances that benefit our kind. One, say STEM, shouldn’t necessarily be held above the others in the humanities or arts.

And finally, I also think it reveals at its heart a fundamental problem with using human feedback to design products and services. Individual users may say they like a more sycophantic AI based on each isolated interaction, just like they also may say they love the way fast food and soda tastes, the convenience of single-use plastic containers, the entertainment and connection they derive from social media, the worldview validation and tribalist belonging they feel when reading politicized media or tabloid gossip. Yet again, taken all together, the cumulation of all of these types of trends and activities often leads to very undesirable outcomes for individuals and society — obesity and poor health in the case of fast food, pollution and endocrine disruption in the case of plastic waste, depression and isolation from overindulgence of social media, a more splintered and less-informed body public from reading poor quality news sources.

AI model designers and technical decision-makers at enterprises would do well to keep this broader idea in mind when designing metrics around any measurable goal — because even when you think you’re using data to your advantage, it could backfire in ways you didn’t fully expect or anticipate, leaving your scrambling to repair the damage and mop up the mess you made, however inadvertently.

Shape
Shape
Stay Ahead

Explore More Insights

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

Shape

Pantheon of college football gets a Wi-Fi upgrade

Notre Dame has fully adopted mobile ticketing and introduced grab-and-go concession stands, with plans to expand them further. Alcohol sales were recently approved, prompting efforts to support new services like mobile carts. In premium areas, fans can stream various games during events. Notre Dame also tested mobile ordering for concessions

Read More »

The U.S. leads the world in AI (job) anxiety

The Americans have the highest search volume with a population-adjusted value of 440,000 search queries on the topic of AI job loss, while their attitude towards AI is moderately positive at 54.5%. The intensity score of 3 for the U.S. shows that the concern of losing jobs to AI is

Read More »

Tigera extends cloud-native networking with Calico 3.30

This logging capability is exposed through two new components: Goldmane: A gRPC-based API endpoint that aggregates flow logs from Calico’s Felix component, which runs on each node. Whisker: A web-based visualization tool built with React and TypeScript that connects to the Goldmane API. The combination of these components provides detailed

Read More »

Energy Secretary Visits Appliance Manufacturing Facility in Georgia to Mark 100 Days of Unleashing American Energy

GRIFFIN, GA— U.S. Secretary of Energy Chris Wright today visited Rinnai America Corporation’s manufacturing facility for non-condensed tankless gas water heaters in Griffin, Georgia, to celebrate the first 100 days of the Trump Administration’s efforts to unleash American energy and innovation. The visit underscored the Department of Energy’s commitment to protecting consumer freedom, defending American manufacturing jobs, and restoring American energy dominance. “President Trump was elected to bring back common sense—to get the barriers out of the way and let Americans pursue their own dreams,” said Secretary Wright. “In these first 100 days, that’s exactly what we’ve started doing. Rinnai America is a perfect example of what’s at stake when Washington pushes reckless regulations, bureaucrats tried to end hundreds of jobs with a rule no one asked for, slipped in the day after Christmas—without a single consideration for the people it would hurt. But these workers stood strong. They didn’t back down and because of their courage and hard work, we won this battle together. What they build here changes lives—millions of people choose these products to make their lives better. That’s worth fighting for. Let’s stand together for our dreams, for the American Dream.” Rinnai is the only company manufacturing non-condensing tankless water heaters in the United States—an energy-efficient technology targeted by a Biden-era rule that would have effectively banned the product and forced this facility to shutter its doors. The Department’s swift action to halt the rule has saved more than 200 Georgia jobs and preserved an affordable, high-efficiency option for American families.  “We were honored to host Secretary Wright and are grateful for the Department’s decisive actions to support American manufacturing and consumer freedom,” said Frank Windsor, President of Rinnai America Corporation. “By preserving access to high-efficient, cost-effective tankless water heaters, the Department is helping companies like ours continue

Read More »

Crude Drops Nearing 2021 Lows Amid Supply Surge

Oil slumped as OPEC+ discussed making a second major production increase, inflaming concerns about swelling global supplies that have dragged down crude prices this year. West Texas Intermediate futures fell 1.6% to settle near $58 a barrel, down more than 7% for the week, with prices holding near the lowest since early 2021. Key OPEC+ nations are considering another production increase of about 400,000 barrels a day in June ahead of a meeting the group pushed forward two days to May 3. Another aggressive supply boost from the cartel threatens to batter a market already pressured by soft Chinese demand and plentiful output from outside the group. The increase would be in line with figures previously telegraphed by the group and roughly matches last month’s shock hike, which was seen as a bid to discipline over-producing members. “OPEC’s decision framework appears to be fueled by the persistent cheating, particularly from the likes of Iraq, Kazakhstan, Russia among others,” TD Cowen strategists including Dan Ghali and Bart Melek said in a note to clients. Inventories may increase by about 200 million barrels over the next three quarters, which could drop crude toward the low $50s, they wrote. Brent prompt spread — the difference between its two nearest contracts — has narrowed to 36 cents a barrel in backwardation, compared with a gap of $1.07 four days ago. The narrowing spread signals expectations that near-term supplies will be readily available. Crude has shed about 19% this year — and briefly touched a four-year low last month — as the Trump administration’s tariffs fan concerns that energy demand will fall. The drop in prices is already showing signs of squeezing a key industry that US President Donald Trump pledged to help. Some of the biggest US shale-oil producers plan to slash about 4%

Read More »

Energy Department Lifts Regulations on Miscellaneous Gas Products

WASHINGTON— The U.S. Department of Energy (DOE) today announced the withdrawal of the determination of miscellaneous gas products as a covered consumer product under the Energy Policy and Conservation Act (EPCA). This action is yet another step toward President Trump’s pledge to lower costs for the American people by expanding choice and cutting red tape. By withdrawing this rule, DOE will exempt miscellaneous gas products—a category that includes decorative hearths and outdoor heaters—from a range of unnecessary regulations on their manufacture and sale.  “Under President Trump’s leadership, the Department of Energy is returning to common sense – and that means giving the American people the ability to choose which heaters they use in their own backyards,” U.S. Secretary of Energy Chris Wright said. “To date, rescinding or delaying unnecessary consumer regulations such as this have saved the taxpayers nearly $24 billion – and we’re just getting started.”   “Previous DOE rulemaking on this subject lumped together several products that are dissimilar in form and function, subjecting manufacturers to an awkward and unnecessary regulatory framework,” Principal Deputy Assistant Secretary for Energy Efficiency and Renewable Energy Lou Hrkman said. “By withdrawing the previous determination and repealing these unclear definitions, the Trump Administration is sending a clear signal that these markets will be allowed to thrive without fear of undue government interference.” Prior to today’s action, miscellaneous gas products were classified as covered products under Part A of Title III of the EPCA, and therefore potentially subject to burdensome standards for energy conservation. The withdrawal of this classification, along with the repeal of the definitions for “miscellaneous gas products,” “decorative hearth product,” and “outdoor heater” from the Code of Federal Regulations, will allow the market for these products to freely develop without needing to account for new conservation standards from DOE. In addition to today’s action, DOE has officially withdrawn four

Read More »

DOE Announces New Leadership to Tackle Challenges of Growing Energy Demand

WASHINGTON—The Department of Energy (DOE) today announced new leadership to tackle the challenge of strengthening and securing the U.S. energy system and ensuring America can lead the global race for AI leadership. To unleash American Energy Dominance, the systems and infrastructure that produce and deliver energy to the American people must be reliable, resilient, and secure. As energy demand continues to grow, the U.S. needs to upgrade both existing energy infrastructure and build new infrastructure – all of which must be done with resilience and security as priorities.  To advance these goals, today DOE is announcing that the Office of Cybersecurity, Energy Security, and Emergency Response (CESER) will be led by DOE Chief of Staff Alex Fitzsimmons. Carl Coe, who currently leads the Department of Government Efficiency (DOGE) at DOE, will assume the role of DOE Chief of Staff. “The race for global leadership in AI is the new Manhattan Project, and winning this race depends on our ability to increase access to abundant supplies of reliable, affordable energy and build secure infrastructure,” said U.S. Secretary of Energy Chris Wright. “The Department of Energy is focused on the need to meet growing energy demand while strengthening the resilience and security of U.S. energy infrastructure against all threats and hazards.   “Alex has served as a critical leader across the Department in our first 100 days, and his expertise and ability to take on complex problems make him the right person to spearhead this important office. I am grateful for his ongoing leadership within the Department, and I look forward to continuing to work with Carl Coe in his new role as Chief of Staff.” As Chief of Staff to the Secretary, Alex Fitzsimmons led the DOE beach-head team on day one and through the first 100 days of the Administration.

Read More »

Chevron Cuts Buybacks and Exxon Sits Tight as Oil Plunges

Chevron Corp. will reduce share buybacks this quarter after oil prices tumbled, indicating that President Donald Trump’s trade war is hurting a key US industry he pledged to help. The Houston-based company said Friday it will repurchase about $2.75 billion of stock in the second quarter, about 30% less than it bought in the first three months of the year. It comes despite Chevron beating earnings estimates on more low-cost production from Kazakhstan and the Permian Basin.  Exxon Mobil Corp., which also reported earnings Friday, is sticking to its plan to buy back about $5 billion in shares per quarter. And Shell Plc said it has the financial wherewithal to keep repurchasing upwards of $3 billion of shares each quarter even if crude plunges as low as $50 a barrel. Big Oil is finding it increasingly difficult to maintain share buybacks as Brent crude slumped 17% this year to about $62 a barrel at the close Thursday. Trump’s tariffs are poised to slow demand growth for crude and increase the cost of steel and other materials needed to produce oil and gas. At the same time, OPEC and its allies surprised markets last month with a plan to increase oil supplies more than expected later this year.  “Oil prices have changed,” Chief Financial Officer Eimear Bonner said in an interview. “The market, from a supply and demand perspective, appears to be softening.” The downturn in oil prices is starting to show the relative strength and weakness between the world’s supermajors. BP plc and Chevron cut their buybacks while Exxon, Shell and TotalEnergies SE maintained their payouts. Still, with debt levels rising across the group, it remains to be seen which is the right approach — especially if crude prices continue to decline. Brent crude futures slipped about 0.3% Friday, to

Read More »

Scottish government offers up £8.5m for next round of just transition fund

The next tranche of the Just Transition Fund is set to open, as £8.5 million in public funds comes up for grabs for firms in Moray and the north-east. The Scottish Government is set to receive applications on Tuesday 6 May from businesses looking to take advantage of funds set aside to create jobs in low-carbon industries and enhance green and net zero skills. Speaking to Energy Voice, acting net zero secretary Gillian Martin said: “This time around, we are really focusing in on the key economic benefits to the region. “We looked at how the last tranche was done and looked at where we were getting the most bang for our buck, and that was in the development of the skills and the creation of jobs and recognising that there was economic opportunity that was out there to be grasped.” Martin appeared in Aberdeen to announce the opening of the next wave of government investment as she visited Sarens PSG, which received backing during the last Just Transition Fund tranche. Jack Anderson, project engineer at Sarens PSG, benefited from the £175,000 previously awarded to the Altens-based business as it used the government cash to upgrade training facilities. Anderson explained: “In the two years I’ve been here, I’ve received a lot of technical training, which has enabled me to progress my career in engineering. “It has enabled me to focus on the future of the huge projects that we’re looking at, along with looking at delivering Sarens PSG’s biggest project to date in the Moray West project.” In November, the 60th and final turbine was installed at the 882MW Moray West offshore wind farm. Operator Ocean Winds expects the project that sits off the east coast of Scotland to become fully operational this year. More wind projects needed to ‘derisk

Read More »

ExtraHop looks to eliminate ‘extra hops’ in NDR stack

This deep visibility allows ExtraHop to provide insights across the entire network stack, from basic connectivity to application-level transactions. “The benefit of going all the way through Layer 7 is I can actually see a database transaction going through on the wire,” Vasani said. “If you have application teams complaining about database query latency, we can map it to what session was that tied to and what flows was it tied to from a network perspective and is this really an app server issue, or is it a network issue, or is it an endpoint issue?” The new sensor integrates with ExtraHop’s RevealX platform, feeding telemetry into the company’s cloud-scale ML/AI engine that powers its detection and analysis capabilities. “The sensor collects the telemetry, feeds it into an ML/AI engine that sits in the cloud, and then we layer in workflow engines on top to enable the various use cases,” Vasani said. In modern distributed enterprise environments, network visibility must extend beyond traditional data centers. ExtraHop’s all-in-one sensor is designed to address this reality with deployment options that span physical appliances, virtual machines and cloud environments. ExtraHop has both virtual and physical hardware appliances for sensor deployment. ExtraHop sensors can plug into a network through multiple methods including, Network Tap, SPAN (Switched Port Analyzer) port, packet broker or a cloud provider’s vTAP capabilities.

Read More »

AI’s energy appetite drives interest in nuclear power

In its new report, Deloitte said that its analysis of figures from the World Nuclear Association, the American Nuclear Society, the U.S. Department of Energy, and others showed that new nuclear power could potentially meet about 10% of the projected increase in data center demand over the next decade, assuming capacity is also significantly expanded by between 35GW and 62GW, and 30% of the expansion is earmarked for data centers. “Nuclear energy presents a potential solution for meeting some of the growing electricity demands of data centers, with its reliable and clean energy profile,” Deloitte’s report said, noting five key advantages of the technology: Reliable baseload power: Nuclear reactors operate 24/7, regardless of the weather, providing the reliable power so important to data centers. In addition, Deloitte said, “Their capacity factor, exceeding 92.5%, outperforms other sources like natural gas (56%) and renewables like wind (35%) and solar (25%).” High energy density: A small amount of fuel generates a lot of power, which minimizes the need for fuel storage and transportation. “This efficiency can translate to a smaller physical footprint and enhanced sustainability,” Deloitte said. Scalable power output: A full-sized reactor typically generates 800 megawatts (MW) or more of electricity, which accommodates the needs of large data centers. Low carbon emissions: Nuclear power plants produce virtually no greenhouse gas emissions during operation. Enhanced land use efficiency: Compared to other energy sources, nuclear power plants require relatively little land. Gartner’s Johnson echoed these advantages, and also predicted that nuclear energy, and small modular reactors (SMRs) in particular, will “provide a viable answer” to the question of what to do when electricity demand exceeds supply. They can, he said, “ensure independence from grid power fluctuations by providing dedicated on-site power for large data centers.” However, both Gartner and Deloitte also highlighted challenges in

Read More »

Nvidia AI supercluster targets agents, reasoning models on Oracle Cloud

Oracle has previously built an OCI Supercluster with 65,536 Nvidia H200 GPUs using the older Hopper GPU technology and no CPU that offers up to 260 exaflops of peak FP8 performance. According to the blog post announcing the availability, the Blackwell GPUs are available via Oracle’s public, government, and sovereign clouds, as well as in customer-owned data centers through its OCI Dedicated Region and Alloy offerings. Oracle joins a growing list of cloud providers that have made the GB200 NVL72 system available, including Google, CoreWeave and Lambda. In addition, Microsoft offers the GB200 GPUs, though they are not deployed as an NVL72 machine.

Read More »

Deep Data Center: Neoclouds as the ‘Picks and Shovels’ of the AI Gold Rush

In 1849, the discovery of gold in California ignited a frenzy, drawing prospectors from around the world in pursuit of quick fortune. While few struck it rich digging and sifting dirt, a different class of entrepreneurs quietly prospered: those who supplied the miners with the tools of the trade. From picks and shovels to tents and provisions, these providers became indispensable to the gold rush, profiting handsomely regardless of who found gold. Today, a new gold rush is underway, in pursuit of artificial intelligence. And just like the days of yore, the real fortunes may lie not in the gold itself, but in the infrastructure and equipment that enable its extraction. This is where neocloud players and chipmakers are positioned, representing themselves as the fundamental enablers of the AI revolution. Neoclouds: The Essential Tools and Implements of AI Innovation The AI boom has sparked a frenzy of innovation, investment, and competition. From generative AI applications like ChatGPT to autonomous systems and personalized recommendations, AI is rapidly transforming industries. Yet, behind every groundbreaking AI model lies an unsung hero: the infrastructure powering it. Enter neocloud providers—the specialized cloud platforms delivering the GPU horsepower that fuels AI’s meteoric rise. Let’s examine how neoclouds represent the “picks and shovels” of the AI gold rush, used for extracting the essential backbone of AI innovation. Neoclouds are emerging as indispensable players in the AI ecosystem, offering tailored solutions for compute-intensive workloads such as training large language models (LLMs) and performing high-speed inference. Unlike traditional hyperscalers (e.g., AWS, Azure, Google Cloud), which cater to a broad range of use cases, neoclouds focus exclusively on optimizing infrastructure for AI and machine learning applications. This specialization allows them to deliver superior performance at a lower cost, making them the go-to choice for startups, enterprises, and research institutions alike.

Read More »

Soluna Computing: Innovating Renewable Computing for Sustainable Data Centers

Dorothy 1A & 1B (Texas): These twin 25 MW facilities are powered by wind and serve Bitcoin hosting and mining workloads. Together, they consumed over 112,000 MWh of curtailed energy in 2024, demonstrating the impact of Soluna’s model. Dorothy 2 (Texas): Currently under construction and scheduled for energization in Q4 2025, this 48 MW site will increase Soluna’s hosting and mining capacity by 64%. Sophie (Kentucky): A 25 MW grid- and hydro-powered hosting center with a strong cost profile and consistent output. Project Grace (Texas): A 2 MW AI pilot project in development, part of Soluna’s transition into HPC and machine learning. Project Kati (Texas): With 166 MW split between Bitcoin and AI hosting, this project recently exited the Electric Reliability Council of Texas, Inc. planning phase and is expected to energize between 2025 and 2027. Project Rosa (Texas): A 187 MW flagship project co-located with wind assets, aimed at both Bitcoin and AI workloads. Land and power agreements were secured by the company in early 2025. These developments are part of the company’s broader effort to tackle both energy waste and infrastructure bottlenecks. Soluna’s behind-the-meter design enables flexibility to draw from the grid or directly from renewable sources, maximizing energy value while minimizing emissions. Competition is Fierce and a Narrower Focus Better Serves the Business In 2024, Soluna tested the waters of providing AI services via a  GPU-as-a-Service through a partnership with HPE, branded as Project Ada. The pilot aimed to rent out cloud GPUs for AI developers and LLM training. However, due to oversupply in the GPU market, delayed product rollouts (like NVIDIA’s H200), and poor demand economics, Soluna terminated the contract in March 2025. The cancellation of the contract with HPE frees up resources for Soluna to focus on what it believes the company does best: designing

Read More »

Quiet Genius at the Neutral Line: How Onics Filters Are Reshaping the Future of Data Center Power Efficiency

Why Harmonics Matter In a typical data center, nonlinear loads—like servers, UPS systems, and switch-mode power supplies—introduce harmonic distortion into the electrical system. These harmonics travel along the neutral and ground conductors, where they can increase current flow, cause overheating in transformers, and shorten the lifespan of critical power infrastructure. More subtly, they waste power through reactive losses that don’t show up on a basic utility bill, but do show up in heat, inefficiency, and increased infrastructure stress. Traditional mitigation approaches—like active harmonic filters or isolation transformers—are complex, expensive, and often require custom integration and ongoing maintenance. That’s where Onics’ solution stands out. It’s engineered as a shunt-style, low-pass filter: a passive device that sits in parallel with the circuit, quietly siphoning off problematic harmonics without interrupting operations.  The result? Lower apparent power demand, reduced electrical losses, and a quieter, more stable current environment—especially on the neutral line, where cumulative harmonic effects often peak. Behind the Numbers: Real-World Impact While the Onics filters offer a passive complement to traditional mitigation strategies, they aren’t intended to replace active harmonic filters or isolation transformers in systems that require them—they work best as a low-complexity enhancement to existing power quality designs. LoPilato says Onics has deployed its filters in mission-critical environments ranging from enterprise edge to large colos, and the data is consistent. In one example, a 6 MW data center saw a verified 9.2% reduction in energy consumption after deploying Onics filters at key electrical junctures. Another facility clocked in at 17.8% savings across its lighting and support loads, thanks in part to improved power factor and reduced transformer strain. The filters work by targeting high-frequency distortion—typically above the 3rd harmonic and up through the 35th. By passively attenuating this range, the system reduces reactive current on the neutral and helps stabilize

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 »