Physicists have been trying to measure the fundamental gravitational constant for well over two centuries. The current accepted value of big G, as it’s known, is 6.67430 × 10-11 cubic meters per kilogram per square second. It also has an uncertainty of ±0.00015 × 10^-11 m³/(kg s²). As far as constants of the universe go, that’s very uncertain.

Stephan Schlamminger recently completed a 10-year effort at the U.S. National Institute of Standards and Technology to replicate an earlier measurement of big G from the International Bureau of Weights and Measures, or BIPM (located near Paris) that’s notably higher than most measurements. He spoke with IEEE Spectrum about why it took so long to get a number—6.67387 x 10^-11 m³/(kg s²)—and why it’s notably lower than the BIPM result, to the tune of 0.0235 percent.

Why is it so difficult to measure big G?

Stephan Schlamminger: Gravity is very weak. When you were a kid, you probably played with fridge magnets, and it was a force you could feel. But if you have two coffee cups, you can try all you want—you can’t feel the force between them. It is there, but it’s so, so weak.

How did you attempt to measure big G?

NIST used a torsion balance with a fourfold geometry. This animation shows an exaggerated version of how the outer green masses gravitationally attract the inner blue masses.S. Kelley/NIST

Schlamminger: We used what’s called a torsion balance. The key idea in the torsion balance is that it decouples vertical gravity that you have from Earth from horizontal gravity, and that makes it sensitive to masses that are around the torsion balance but not the Earth below.

Ours had a fourfold geometry. It has a very thin torsion strip, then four cylinders in a “plus sign” arrangement. All of this is inside a vacuum. Outside, we have four larger cylinders that gravitationally attract the four smaller masses to them. If I move the outer masses just a tiny little bit, the plus sign will rotate, and we measure that angle that it moves. That angle is proportional to the gravitational torque.

Why try to replicate the BIPM value?

Schlamminger: We could move the field forward. The measurements have been plagued with inconsistencies, so by redoing an experiment, we hoped to shed light on the inconsistencies.

We did not find a smoking gun, so there’s no single reason why it’s different—our value versus their value. It’s still a big question mark.

What was it like spending 10 years on this?

Schlamminger: It’s a bit like herding cats. I’ve measured other fundamental constants, like Planck’s constant, and for most experiments, they have some sort of self-calibration built in. But with the gravitational constant, you have to keep track of every single mass that moves—where they are, how big they are, and weigh them.

How does your result compare to the rest?

Schlamminger: Our result is a little bit below the standard accepted literature value. I was disappointed because it doesn’t agree with the BIPM value, nor with the literature value. If there’s something wrong with the BIPM experiment, then the literature value—which includes that result—probably ought to come down a bit. But that is not for me to say. I think somebody else, independent, should figure out what the new mean value ought to be.

A Katalyst Space robot that was launched to grab on to a space telescope and raise it to a higher orbit tumbled out of control after a series of failures, NASA and the company said today.

The company is trying to stabilize the spacecraft using back-up systems, before NASA will decide if the mission can continue.

This is the first time the space agency has hired a private company to lift one of its space telescopes to a higher orbit, allowing the observatory to continue operations beyond its expected lifespan.

The Neil Gehrels Swift Observatory was launched in 2004, with the ability to swiftly point at ephemeral space events like gamma ray bursts. The spacecraft has been pulled back toward Earth and needs to be pushed back up to a higher orbit to continue doing its job.

The LINK spacecraft was launched into space on July 3 on board a Northrop Grumman Pegasus rocket. Over the past several weeks, flight controllers have worked to commission the spacecraft, activating its various systems before it heads off to rendezvous with Swift.

However, the spacecraft suffered issues controlling itself and began spinning over the weekend. According to NASA, the spinning has resulted in “sporadic communications” with the satellite, likely because its antenna flips away from the planet. Two of the three reaction wheels that control the spacecraft’s alignment have failed, and there are problems with one of the spacecraft’s thruster systems.

Flight controllers are working to recover the spacecraft using another set of thrusters on the spacecraft to slow the spin, and they said in a statement that “we have already begun this series of burns and are seeing the intended effect.” Katalyst told TechCrunch that “this remains an active mission, and we continue to move forward with plans to rendezvous with Swift.”

Katalyst’s Kieran Wilson, the principal investigator for the mission, told reporters ahead of the mission that the spacecraft had been built incredibly quickly because of the urgent need to raise Swift in the next few months.

“All this is challenging and risky,” he said at the time. “There’re a lot of spacecraft that have had far longer development cycles with far more funding behind them that have failed for mundane reasons.”

Katalyst raised a $12 million round in June to help support this mission, backed by investors Fortitude Ventures and Geodesic Capital. The company has also won funding from the U.S. military, which is interested in a dynamic vehicle that can service satellites or surveil rival spacecraft.

This story has been updated to include comment from Katalyst.

OFFBEAT

Researchers uncover patterns in how sleeping minds recombine memories, people, and places

Your brain apparently spends the night remixing reality rather than simply replaying it, according to researchers armed with AI and more than 3,700 accounts of dreams and waking life.

Researchers at Italy’s IMT School for Advanced Studies Lucca used natural language processing (NLP) to analyze more than 3,700 reports of dreams and waking experiences from 287 adults and found dreams are neither random mental noise nor faithful replays of the day’s events. Instead, the sleeping brain appears to recycle memories, emotions, familiar places, and imagined possibilities into entirely new scenarios.

Published in Communications Psychology, the study combined two weeks of dream and daily experience diaries with data on participants’ personalities, sleep quality, cognitive abilities, and psychological traits. Researchers used NLP to compare the semantic structure of the waking and dream reports.

Everyday settings such as workplaces, hospitals, and classrooms were combined with unrelated places, shifting viewpoints, or unfamiliar surroundings, while different parts of a person’s life could merge into a single scene. The findings suggest dreams do not replay reality so much as rebuild it from familiar parts.

Dream reports also varied with participants’ traits and attitudes. Those prone to mind-wandering tended to report dreams that jumped quickly from one scene to another, while those who said dreams were personally meaningful generally described more vivid and immersive experiences.

The researchers also looked at dream reports collected during the COVID-19 lockdown by a separate team at Sapienza University of Rome. Compared with more recent reports, those dreams contained more references to confinement, barriers, restrictions, and heightened emotions, reflecting the reality people were living through at the time.

“Our findings show that dreams are not just a reflection of past experiences, but a dynamic process shaped by who we are and what we live through,” said lead author Valentina Elce. “By combining large-scale data with computational methods, we were able to uncover patterns in dream content that were previously difficult to detect.”

The work also hints at a new role for AI. Rather than asking researchers to read and categorize thousands of dream reports themselves, the team used NLP to spot patterns across the entire dataset. The software’s assessments closely matched those of independent human reviewers, suggesting it could make large-scale dream studies far less laborious.

Before you ask ChatGPT why your dentist turned into your boss, the research has some obvious limitations. It relies on people remembering and accurately describing their dreams, and it doesn’t explain why we dream in the first place. However, it does suggest dreams have more structure than random neural noise – even if your brain still insists on making you sit an exam you forgot to revise for. ®

Runway spent weeks trying to engineer its way out of a stubborn bug: AI-generated avatars would drift off-center during real-time video generation. The fix wasn’t a back-end patch — it was a new front-end feature that just worked around the problem. That’s the kind of lesson Ryan Phillips, head of enterprise product at Runway ML, walked through at VB Transform 2026, arguing that even companies not building foundation models themselves can learn from how Runway builds, evaluates, and ships them.

“I think even if you are not all building models yourselves, it’s helpful to learn how we do it because I think almost all of the lessons are applicable to what you all are doing day-to-day,” Phillips said.

Runway is an applied AI research company building general world models to power generative tools. During his presentation, Phillips showcased Runway Characters, a real-time video model that enables zero-latency, back-and-forth interactions with AI-generated avatars. Five years ago, creating a video with illegible text and low framerates took artists hundreds of hours of stitching individual frames together, he said. Today, Runway’s models generate interactive video on the fly.

“Studying how we build these real-time models can inspire how you build and deploy real-time experiences, whether agentic or not, in your companies today,” he said.

Demystifying evals

Building a robust AI product starts with a high-quality evaluation set. However, creating this set cannot be treated solely as an engineering task. It requires deep cross-functional alignment across product, design, research, and sales to define what “quality” actually looks like.

Phillips emphasized running internal workshops where team members review generated examples together. The goal is to align the entire organization on specific failure modes so everyone shares a unified definition of a successful generation.

“We spent a lot of time working with our team, running through examples… of what success and failure looks like, down to the very, very detailed and picky things,” Phillips said.

The resulting evaluation set must cover broad customer use cases alongside extreme edge cases. For instance, Phillips highlighted that to ensure the model behaves predictably when pushed beyond standard human facial structures, they used “Tooth,” a non-human character with no nose and very unusual teeth.

When grading these generations, the Runway team looks for subtle artifacts. In one example, a video where a character’s face remained intact but background elements, such as a net, began morphing was strictly graded as a failure.

Despite the cutting-edge nature of the product, the tool Runway uses to track these evaluations is simple: an Excel spreadsheet. The team logs tests daily, categorizing outputs as “minor” or “major” failures against a predetermined pass rate. 

“We set a bar before we get started on what percentage we need to pass, and when we hit that, we ship the model,” Phillips said. “So it’s not magical.”

For enterprise developers facing non-deterministic quality drift in their own real-time pipelines, manual evaluation at scale is a bottleneck. To solve this, Phillips noted that developers can rely on language models to automate the visual grading process. 

“LLMs are getting quite good at being a judge for a lot of this content, especially the types of morphing or changing that you would see in an evaluation set,” he said. Teams can also feed an LLM behind-the-scenes context (e.g., a hand-drawn sketch or an ad’s structural layout) to guide the generation and validation processes, ensuring quality without adding cognitive load to the end user.

Model training and turning bugs into features

Delivering real-time generative video requires a highly optimized technical stack. The process begins with pre-training a massive foundation model, which is resource-intensive and slow to generate outputs. To achieve real-time latency, Runway relies on distillation, where a smaller, faster “student” model is trained to mimic the large “teacher” model. According to Phillips, distillation helps Runway cut down “80 to 90% of the generation time.”

The team then applies adversarial post-training (APT) to the distilled model. This technique forces the model to continuously improve by testing it against a system designed to find its flaws, helping regain the visual sharpness lost during the distillation process.

However, altering the model architecture introduces new problems. The distillation and APT phases introduced a stubborn bug: characters would sway or drift from the center of the frame during real-time generation.

The team spent weeks attempting to fix the core model to eliminate the drift, he said. Ultimately, they discovered that if the user’s initial input image was perfectly centered, the generated video remained stable. Instead of spending more time on a backend engineering patch, Runway pivoted to a user experience solution.

“What we did was, when we noticed this in our evaluations, we then said, ‘What if we just offered that as a feature?’ If a user gives us a character that is turned to the left, we know the video is going to morph. Let’s just fix it for them,” Phillips said. 

They introduced a frontend feature called “Optimize for Image Quality,” which automatically re-centers the user’s image before generation begins. By wrapping a backend model limitation in a frontend tool, users perceived a helpful feature rather than an engineering flaw.

“Turn model limitations into product features so that you can actually expand how the model works,” Phillips advised. “It might feel like a limitation internally, but your customers will not see it that way if you’re kind of building this in as a product feature.”

The devil is in the infrastructure details

Delivering video globally at 24 frames per second requires optimizing every layer of the infrastructure stack. This ranges from caching and parallel decoding to making deep kernel changes in partnership with hardware providers like Nvidia.

Shortly after launching Runway Characters, he said the team noticed that 8% of API calls were dropping to 16 frames per second, causing the video to stutter for customers. 

Finding the root cause required deep observability. The team used an AI agent powered by Claude alongside monitoring tools like Datadog and Sentry to trace the anomaly. The debugging session isolated the problem to a single data center in the us-east-1 region.

“The solution actually wasn’t [to] go fix anything or change a config,” Phillips explained. “They actually went and physically replaced those GPUs in the data center to fix it, and that ultimately solved the problems.”

For enterprise teams deploying real-time applications, the takeaway is clear: hardware and infrastructure anomalies will directly impact model performance, requiring rigorous, full-stack debugging capabilities. 

“Don’t forget about all the small details, because there’s so many of them when you’re deploying these models,” Phillips said.

Surviving “failure hell” and the future of world-building

Developing AI systems is rarely a linear process. Teams often find themselves stuck for weeks on a single problem with no end in sight, a phase Phillips referred to as “failure hell.”

“We think you have to go through that pain and really struggle with the problem for a little bit before you can get the breakthrough,” he said. Consistent iteration eventually flattens the difficulty curve, triggering sudden, exponential improvements.

As the underlying models overcome these technical hurdles, the role of enterprise creatives is also fundamentally changing. Traditionally, marketing and design teams have focused on creating single assets, like a specific advertisement or illustration. In an era of real-time generation and agentic workflows, that paradigm is shifting toward defining parameters, aesthetics, and intellectual property.

“You might not be designing a single ad, but you might design a world that then the agent or a real-time video model can generate ads from,” Phillips said. 

Last week’s unprecedented security event in which two OpenAI security hacking models trespassed into the network of fellow AI company Hugging Face was enabled by exploiting one or more zero-day vulnerabilities in Artifactory, JFrog, the product’s developer, said Monday.

In an incident mimicking a dystopian sci-fi novel, two OpenAI models broke out of the restricted environment meant to keep them from accessing the Internet during an internal test, the AI company revealed last week. The models went on to breach Hugging Face’s network and steal confidential information and credentials. OpenAI said its agent achieved the feat by exploiting a previously unknown vulnerability. The company called the event “unprecedented,” and outsiders largely agreed.

Not the triumph made out to be

OpenAI said the models exploited multiple attack vectors, including stolen credentials and zero-days, to gain remote code execution capabilities, but until now, the vulnerable software was unknown. JFrog’s Monday disclosure said the product was a self-managed instance Artifactory, a repository management system that secures and streamlines customers’ software development operations. JFrog says Artifactory is used by more than 7,500 developer Teams, 80 percent of which work for Fortune 100 companies.

“During an internal evaluation of frontier cyber capabilities, OpenAI’s models, running deliberately without production safeguards in an isolated research environment, autonomously discovered and employed chained vulnerabilities to escape its sandbox, reach the open internet, and extract evaluation answers from Hugging Face’s infrastructure,” JFrog CTO Yoav Landman wrote. The executive went on to say that the company learned of the zero-days from OpenAI.

The company said Monday that it fixed the exploited vulnerabilities, but it didn’t identify them or provide other important details, such as the conditions under which the vulnerabilities can be exploited. Such details are standard in many vulnerability disclosures because they’re necessary for customers to assess risks. In an email, a company representative declined to provide the details.

Release notes published Monday for version Artifactory 7.161.15 listed the CVE designations for nine patched vulnerabilities. The disclosure made no mention that any of them had been actively exploited in the wild. External sources, however, show that three of them—CVE-2026-65617, CVE-2026-65923, and CVE-2026-66018—were privately reported by OpenAI researcher Khai Tran. It’s likely that at least two of them were the zero-days OpenAI’s models exploited, but without confirmation, it’s impossible to say so definitively.

In shaping her career, Peru native Mariana Costa has asked herself a question: What can I do to make life better for women in Latin America?

The answer she landed on was training them for tech jobs.

Such positions pay well and are in demand. And for too long, women across the region have been locked out of them, she says.

Costa is president of Laboratoria, a U.S.-registered nonprofit based in Miami that she helped found. Laboratoria has trained thousands of women in 11 Latin American countries for technology careers. She has built training centers in the countries and has placed graduates at major companies. Meanwhile, she has become one of the most recognized voices in the region on workforce equity and tech education for women.

IEEE recognized her work with its President’s Award this year for her “distinguished leadership and contributions to the betterment of society.” Recipients of the award are selected by the IEEE president with the consent of the IEEE Board of Directors.

Costa says the recognition came as a surprise because she is not an engineer by training and had never considered becoming affiliated with IEEE.

She was presented with the award at the IEEE Honors Ceremony on 24 April in New York City.

Peru: a country of contrasts

Costa grew up in Lima, Peru’s capital, in a household with no connection to engineering or technology. Her mother was an art historian and professor, and her father was a lawyer. The family was financially comfortable and traveled abroad regularly. Costa attended well-resourced schools.

That economic stability came with a reckoning, Costa says, in that she recognized early on that economic inequality had created separate societies inside Peru. Her parents, she says, made it “clear that my reality wasn’t the reality of most people in my country.”

Lima is a microcosm of the country, she says. The divide in the capital city is visible: A kilometers-long concrete wall topped with barbed wire separates wealthier neighborhoods from shantytowns, where residents lack running water.

Nationally, there are splits along ethnic and geographic lines. The highland and jungle regions remain home to mostly indigenous communities with limited educational access and a deep cultural distance from the Hispanic-dominated coast.

The questions that stirred in her as a child never left, she says.

“Why do I live in a country where so much depends on where you’re born?” she asked herself. “What does it mean to be Peruvian when individual realities are strikingly different?”

Those questions followed her to the London School of Economics, where she studied international relations, graduating with a bachelor’s degree in 2007. She held onto the questions when she moved to Washington, D.C., where she spent the next four years working for the Organization of American States, helping Latin American governments improve public services that fall under the heading of civil registration.

“I said, ‘How can it be? The tech space has so many rich opportunities. Why aren’t any women here?’”

The OAS Universal Civil Identity Program in the Americas provides technical support to national civil registry institutions, modernizing them to foster social inclusion and ensuring the right to civil identity for all people. Without civil identity, a person can’t access education, health care, legal employment, social services, or the right to vote. People without the classification don’t exist in the eyes of the government. They also can’t own property, get married officially, or pass citizenship rights to their children.

Doing that work deepened her concern about the socioeconomic disparities in her homeland, she says. In search of practical solutions to those problems, she went to New York City in 2011 to further her education. She earned a master’s degree in public administration and development from Columbia in 2013.

Technology was not yet part of a solution. But Costa already had met someone who would change that.

Falling in love with a programmer

While working in Washington, Costa met Herman Marìn, a software engineer who used digital tools in support of social causes. Because he was doing work she had never associated with programmers before, her assumptions about the field dissolved quickly.

“I had a vision of [programmers doing] something not very social—strictly technical,” she says. “And my then-boyfriend, now husband, actually worked for different social movements that used technology to address social causes.”

That realization cracked something open, she says: “I said, ‘Oh! Technology can actually be a tool to address some of the more stubborn problems in our societies.’”

After earning her degree at Columbia, Costa returned to Lima with her husband. She had been abroad for nearly a decade and felt the pull of home.

“The thought of not moving back to my country was something that tormented me a bit,” she says. “I really felt I had to move back, at least to try it out and contribute somehow.”

What Latin America’s tech space lacked

When Costa, her husband, and a friend from graduate school moved to Lima, they had modest savings and big ambitions. They wanted to build something that combined technology with social impact.

They started with what they had: a small digital services agency, where they built websites for clients.

The business grew, and they hired more employees. Their team expanded to a dozen software engineers. And as it did, Costa noticed three things.

First, there weren’t enough trained developers to meet the demand. Second, many of their best hires did not have traditional computer science degrees. Some had never even finished college.

“There was no other space where you could actually build an amazing career and get a well-paying job without a good degree from a good school,” she says. “The tech world was different. It was open in ways other fields weren’t.”

Thirdly, she noticed that there were no women on the team. In the first six months, Costa says, they didn’t interview a single female developer.

Her colleagues shrugged. It’s just how it is, they told her.

Costa, the outsider, didn’t accept that.

“I said, ‘How can that be? The tech space has so many rich opportunities,’” she says. “‘Why aren’t there any women?’”

Building Laboratoria

In 2014 she decided to launch Laboratoria. The business model was simple: Find talented women who hadn’t yet broken into tech, train them quickly on practical skills, and connect them with employers who needed developers.

Laboratoria started offering a six-month immersive boot camp that covered Web development, UX design, data literacy, strategic use of artificial intelligence, and soft-skills coaching such as interview prep and projecting confidence.

Just as important for career success, Costa says, is a user-centered mindset. She says Laboratoria’s program emphasizes the discipline of keeping the client’s needs in mind when designing the work.

The teaching model has evolved beyond the boot-camp structure, but the organization still focuses on helping Latin American women develop tech skills and land quality jobs in the digital age. These days, the training, conducted via twice-weekly live Zoom sessions, lasts six weeks.

“We needed developers ourselves,” she says of the company’s original logic. “I said, ‘Why don’t we run a program to train women—women who are incredibly talented but haven’t been given a chance yet—and help them gain the skills they need to get a great job as quickly as possible?’”

Mariana Costa [seated, right] poses with Laboratoria co-founder and CEO Gabriela Rocha and co-founder and chief product officer Rodulfo Prieto.Valeria Martens

It worked. Laboratoria expanded from Lima to Santiago, Chile; Mexico City; São Paulo, Brazil; and Bogotá, Colombia. The organization eventually incorporated as a nonprofit in the United States. Today its programs are held remotely in Latin America’s major job markets. So far, Laboratoria has opened the doors to tech careers for more than 3,500 women.

Costa says she believes the most important skills Laboratoria’s graduates need aren’t purely technical. Close behind the growth mindset is self-confidence, she says.

“Knowing who you are, valuing who you are, and learning to trust yourself and your capacities are indispensable traits,” she says.

Networking, she adds, is the third pillar, and often the hardest to build for women without access to elite schools or flexible work schedules.

“When you go out in the market,” she says, “you realize that having a network of people who trust you and know your work is such a valuable and critical asset.”

IEEE: a new connection

Costa’s introduction to IEEE came late—but it landed hard.

She is not an IEEE member, so when she was contacted this year about receiving the President’s Award, she did her homework on the organization. What she found, she says, was a public charity whose reach and values aligned with her mission.

“IEEE is about expanding access to opportunities in the world of technology,” she says. “And that’s also the core of what we do at Laboratoria.”

She says she also sees IEEE as a living example of something her company preaches every day: “I was talking about the value of professional networks, and I think IEEE is such an amazing reference in that space. It exemplifies this belief that human connection—not only doing your work but also sharing and learning with others—is at the core of building thriving technology careers.”

The engineering organization found her well after she launched her tech-related career. But it wasn’t too late. She says she intends to make the most of the connection.