© GoogleGoogle Images’s latest update appears to be taking some inspiration from the digital pinboard platform Pinterest.
The company is celebrating Google Images’s 25th anniversary this week by looking back at its history and announcing a major redesign that turns the search engine into more of a personalized visual feed.
It all started with a green dress that broke the internet more than two decades ago.
In a blog post, Google recounted how it launched its image search engine after noticing a huge spike in interest in the now-iconic green Versace dress Jennifer Lopez wore to the 2000 Grammy Awards. At the time, though, Google Search was largely limited to a list of blue text links.
“People didn’t just want to read about the dress—they wanted to see it,” wrote Search Senior Enginering Director Brad Kellett in the blog post. “So in July 2001, we launched Google Images, making it possible for the first time to search and instantly explore visual content from across the web.”
Google Images has since evolved with the times. The company introduced the ability to search using an image instead of text in 2011. In 2018, it integrated Google Lens into Search, allowing people to use their phone cameras to identify objects, translate text, and find products online.
But throughout all those changes, the Google Images homepage has stayed pretty much the same with a mostly white page and a search bar in the center. That is about to change.
© Google
Google announced Tuesday that the homepage is being updated with an immersive gallery of images from across the internet that will refresh in real time and be tailored to a user’s interests. That means people who are signed into their Google accounts will be greeted with a personalized stream of images before they even type anything into the search bar.
As users browse, they will also be able to save images into collections, similar to Pinterest boards. Those collections will appear as tabs above the main gallery.
The goal appears to be to turn Google Images into more than a tool for just finding a specific picture. The new update will make it easier for users to browse for ideas and inspiration for things like fashion, interior design, art, party decor, and vacations.
The new homepage will begin rolling out over the coming weeks to desktop users in the United States in English.
It wouldn’t be a tech announcement in 2026 without some AI. Google also announced that it is bringing image generation directly into AI Overviews using its Nano Banana model.
The company says the feature “transforms a simple text prompt into a high-quality, custom visual made completely from scratch, seamlessly bridging the gap between imagination and reality.”
Google says the tool is meant for moments when someone has a highly specific image in mind that may not already exist online. Image generation in AI Overviews will also begin rolling out in the coming weeks in English in all regions that already support image creation through AI Mode.
The inclusion of AI in this announcment is not too surprising. Pinterest itself has been leaning heavily into AI in recent years.
Last October, the company introduced several AI features, including a shopping assistant, AI-powered board upgrades, and new settings that allow users to control the amount of AI-generated content they see on the platform.
More recently, the company launched a limited-access experimental app called Ask Pinterest that uses AI to help users make complicated shopping and planning decisions through conversational prompts. Last month, Pinterest also announced a $4 billion agreement to use Amazon Web Services infrastructure and chips to train and run its AI models through 2031.
Artificial intelligence labs are after computing power. Kalshi doesn’t have any to offer them but they’ve got something else that might be of some use, according to Bloomberg: a tool that plots the predicted future price of computing power. So, that’s something!
There really is some value for companies to know where the cost of compute is going—especially since currently, it’s mostly just going up. Having a sense of just how much compute is going to cost allows companies to try to lock in a price with a provider before the price spikes.
Demand for compute continues to climb faster than the ongoing data center buildout is able to keep up with. In fact, former Intel CEO Pat Gelsinger recently told CNBC that demand is “almost unlimited.” That is notably more than the amount of energy and processing power available, which actually does have an upper limit.
A recent report from Apollo Global Management described current compute capacity as “effectively sold out,” which has created a bottleneck where the price to rent GPUs keeps climbing faster than new ones are spun up and made available—an issue that is likely only going to get worse as more and more agentic AI tools become available, as recent research suggests they consume up to 136.5 times more energy per query than most generative AI models.
Kalshi’s new tool is supposed to serve as a sort of indicator for companies dealing with those realities of limited resources while trying to generate unlimited growth. Per Bloomberg, it’ll reportedly offer a forward tracking curve of compute, giving an early outlook as to where the cost of computing power is headed in the near-term.
Kalshi will reportedly analyze weekly and monthly contracts for computing power and use an algorithm to predict the future curve, spitting out a price that it expects to see paid in the future. The project reportedly aims to stretch its predictions as far out as a year into the future.
Bloomberg didn’t have details as to whether or not Kalshi will create a market around the price and let people bet on whether the real price will be higher or lower, but that feels like a pretty safe assumption given that it is kind of the prediction market’s whole deal. Plus, it wouldn’t be alone in trying to monetize the curve of compute availability. The Bloomberg report notes that a number of exchanges are looking to list compute futures contracts, which would allow people to trade on the resource like an asset. So if you’re looking for a way to short the future of AI, well, it seems it’s coming.
Like it or not, data centers are now intrinsic to our modern lives, supporting not just the AI boom but healthcare, banking, government services, and other essential sectors. Reliable data center operation depends on effective cooling, which is already a major challenge as many methods require huge inputs of water or energy. To make matters worse, new research suggests that one of our cheapest, most efficient cooling strategies could stop working in a warmer world.
The findings, published Monday in the journal Scientific Reports, show that rising temperatures and humidity levels threaten the viability of direct air free cooling, an energy-efficient, waterless technique that pulls outside air in to cool data center servers. Over the past 45 years, weather conditions that limit direct air cooling have become significantly more common, particularly across the tropics and the southeastern United States, according to the study. As the global temperature continues to rise, this problem is only going to get worse.
“We found that periods of time when temperature and humidity exceed recommended operating thresholds for direct air free cooling are becoming more frequent and lasting longer in many regions,” lead author Christina Karamperidou, a professor of atmospheric sciences professor at the University of Hawaii at Mānoa, said in a statement. “This will reduce the availability of air free cooling for a growing number of data centers globally.”
For direct air free cooling, the American Society of Heating, Refrigerating and Air-Conditioning Engineers recommends keeping the air entering a data center between 64 and 81 degrees Fahrenheit (18 and 27 degrees Celsius), with 10% to 70% relative humidity and a dew point below 59 degrees F (15 degrees C). Air that is hotter and more humid than this won’t cool the servers effectively and could corrode metal components.
To investigate how this cooling method will function in a warmer, wetter world, Karamperidou and her colleagues used a combination of high-resolution hourly weather observations, climate model simulations, and global records of data center locations. With this data, they evaluated how often environmental conditions exceeded recommended operating limits for direct air free cooling over the past 45 years and in future climate scenarios.
The researchers found that the prevalence of weather conditions that limit direct air free cooling has increased significantly in recent decades. Even regions that have only seen modest long-term increases in heat and humidity are experiencing longer daily exceedance events, and the share of data centers exposed to conditions that limit direct air free cooling availability for at least one quarter of the year is rising.
Interestingly, the findings suggest that the hottest, most humid days are intensifying faster than average days, indicating that environmental stress on direct air free cooling systems is become more and more concentrated in rare, highly consequential events.
“From an operational perspective, those worst-day conditions often drive contingency planning, system overrides, redundancy requirements, and reliability decisions,” Karamperidou said. “This suggests that infrastructure planning may need to account not only for average environmental conditions but also for how the most stressful days are changing over time.”
By 2050, the number of hours that exceed temperature and humidity limits for direct air free cooling is protected to increase under high greenhouse gas emissions scenarios, according to the researchers. In most regions globally, the average number of hours per day during which this cooling strategy is constrained increases by more than two hours per day, the findings show.
While this study focuses on how weather can influence data centers, it’s important to remember that data centers can influence local weather too. These facilities dissipate a lot of heat, and research has shown that they can actually create heat islands within a 6-mile radius of themselves.
Karamperidou and her colleagues did not account for this effect, so the direct air free cooling constraints they identified may be conservative, they write in their report. Still, they emphasize that their findings do not mean that this cooling strategy is necessarily infeasible in warm, humid regions. Rather, the study shows that the window of feasibility for direct air free cooling is narrowing due to climate change.
“Alternative strategies—including indirect evaporative cooling, liquid cooling, and hybrid architectures—can partially offset these constraints, albeit with distinct trade-offs in water use, system complexity, and operational design,” the researchers write.
Indeed, as one of the simplest, cheapest, and most efficient cooling strategies becomes increasingly unreliable, data center operators may be forced to turn to more energy- and water-intensive methods. This, in turn, could put added strain on electric grids and water resources that are themselves strained by climate change. Adapting data centers to a warming world without exacerbating the impacts of rising global temperatures will require innovative solutions.
It’s real, and it’s precious. This week, after years of anticipation and speculation, cameras began rolling on The Lord of the Rings: The Hunt for Gollum, the next big-screen foray from the iconic franchise. And, to announce the news, Warner Bros. released a short but sweet video that’s sure to give you goosebumps.
The Hunt for Gollum won’t have to hunt for him too much on set. The actor portraying the title character, Andy Serkis, is also directing this time, taking the reins from his predecessor, Peter Jackson, who this time produces. Here’s the video.
So, of course, here we see Serkis not just in his role as director, but as the star, the former hobbit turned into a despicable monster by the power of the One Ring. It then transitions to some kind of gorgeous landscape, presumably in New Zealand, which instantly makes us feel like the original films.
Unfortunately, while we now have this official news of production having begun, we don’t yet have an official plot description. We’re pretty sure, though, that the film takes place before and during the timeline of The Fellowship of the Ring as Gandalf and Aragorn search for Gollum around the time he was captured by Sauron’s armies and revealed Bilbo Baggins was the last to have the ring.
We do, however, have an official cast list that confirms, in addition to Serkis, Ian McKellan, Elijah Wood, and Lee Pace are returning to their roles as Gandalf, Frodo Baggins, and Thranduil, Jamie Dornan is replacing Viggo Mortensen as Aragorn/Strider, Kate Winslet joins the fray as Marigol, and Anya Taylor-Joy is a character named Seren.
How do they all fit together? What makes this a story worth telling? Hopefully, with cameras now rolling, we’ll learn more as we inch closer to next Christmas and the release of the film.
The Lord of the Rings: The Hunt for Gollum opens on December 17, 2027. Are you ready to return to Middle-earth?
Want more io9 news? Check out when to expect the latest Marvel, Star Wars, and Star Trek releases, what’s next for the DC Universe on film and TV, and everything you need to know about the future of Doctor Who.
There’s been a big push among AI developers in recent years towards the development of more “agentic” systems—that is, algorithms that can autonomously make decisions and interact with digital tools without constant hand-holding from humans. This has been especially true within software development, the field that’s arguably become the most ripe for automation in the ongoing AI boom.
But one of the upshots of building highly agentic AI systems is that they’re prone to all kinds of unexpected behaviors—including now and then deleting copious amounts of files. Multiple people have reported this recently happening to them while using GPT-5.6, the newest model from OpenAI.
On Monday, Bruno Lemos, a Brazilian developer at software company Unlayer, claimed in a X post that the model deleted his entire production database. “This had never happened to me before, with any other model, ever.” He wrote. “[GPT-5.6 is] not safe.”
A screenshot included in the post showed a chat between Lemos and GPT-5.6, in which he asked it to confirm that it had in fact mistakenly deleted his entire production database. The model responded by saying that it “mistakenly ran destructive integration tests” which led to Lemos’ production tables being cleared. “I’m sorry—this should never have happened,” it said.
It followed closely on the heels of another X post from tech investor Matt Shumer—who’s also the author of an essay about AI that went viral earlier this year called “Something Big is Happening”—who reported something similar. According to an attached screenshot, GPT-5.6 told him it had caused “a serious local data-loss incident,” leading to the deletion of what Shumer described as “almost ALL” of his computer’s files. The screenshot showed that the model had executed a “rm -rf” command, which in Linux and Mac systems is used to permanently delete files without requesting user confirmation.
“I’ve never seen anything like this,” Shumer wrote in the thread beneath that post. “Will only be using [Anthropic’s] Fable moving forward.” He added that OpenAI cofounder and president Greg Brockman called him personally and offered to help fix the situation.
Shumer also claimed he had the AI model set to “full access mode,” which allows it to work directly within a user’s database (as opposed to operating within a constrained sandbox). It also comes with a “default mode” that requires users to frequently approve specific tasks, and a more recently introduced “auto-review mode” through which a separate AI agent checks the main coding agent’s work. Beneath his X post, many people claimed Shumer had simply been careless by trusting sensitive files in full access mode.
In the system card for GPT-5.6, published online the day before Shumer’s X post, OpenAI cautioned that when using the model for coding purposes “it is important for users to supervise the agent’s work.” The company added that the model could act in unexpected ways that are misaligned with the user’s goals, and that while these were “most often low severity (e.g. overstating confidence or overclaiming success),” they could in other cases “be meaningfully more severe (e.g. circumventing important security restrictions or deleting important data).”
Lemos, Shumer, and OpenAI did not immediately respond to Gizmodo’s request for comment.
In 2019, French archaeologists unearthed an ancient Celtic settlement from the third century BCE. Things took somewhat of a grim turn when the team, while cataloging metal artifacts from the site, found some remarkably well-preserved shackles—likely relics from the region’s slave trade.
The French National Institute for Preventive Archaeological Research (INRAP) published the results in a recent announcement (translated via Google Translate). In the statement, INRAP reported that the settlement of Allonnes turned up a rich collection of metal objects. In addition to the iron shackles, there were weapons, coins, jewelry, and more, made of iron and copper alloys. The team suspects that the settlement was home to a lively community of Gaulish artisans and merchants—and, in the shadows, individuals who, at the time, were “mere objects of property,” it noted in the statement.
A set of metallic Gallo-Roman offering furniture. Credit: Emmanuelle Collado/INRAP
“The Allonnes excavation yielded a rich assemblage of metal objects—remarkable for their number, variety, and quality, despite some degradation caused by soil acidity,” Thierry Lejars, director at the French National Centre for Scientific Research, said in an accompanying booklet (translated via Google Translate).
Until its conquest by the Roman Empire around 50 BCE, the Gauls occupied most of modern Western Europe. In a 1987 essay, historian Ramsay MacMullen explained that the evidence for slavery in Gaul is “almost exclusively epigraphic.” It’s highly likely that the Gaul elites kept enslaved people to maintain their luxurious homes, but in terms of physical labor, such as in agriculture, the evidence is “scanty.”
The INRAP announcement echoed this sentiment. Generally speaking, “races relating to the poorest members of Gallic society, and especially those of servile populations, remain invisible,” the INRAP noted. Therefore, the discovery of the metal shackles is “extremely rare” for this particular period and territory of the Gauls.
“The identification of shackles and weapons suggests a hierarchical social structure comprising a dominant class and a subordinate one [or imprisoned or enslaved people],” Lejars said.
And the details are quite gruesome. The wrist shackles are around 2.4 inches (6 centimeters) in diameter, which could easily mean they were for women or children. The ankle shackles weighed roughly 2.2 pounds (1 kilogram). The team hypothesizes that the enslaved could have been war prisoners, convicts, or individuals with unpaid debts.
That said, the settlement revealed other aspects of Gallic society, even after its Roman conquest. For instance, the settlement’s location in Allonnes would have placed it at a “particularly advantageous position for trade,” according to the statement. This would explain the sheer abundance of metal relics, as well as semi-finished products and scrap metal. The area also appears to have housed a sanctuary for religious rituals that endured long after the Roman conquest of the Gauls.
A sample of Gallic coins, some of which have been deliberately mutilated. Credit: Emmanuelle Collado/INRAP
Fascinatingly, a significant portion of weapons and trinkets were deliberately mutilated. The team believes this “reveals a religious intent—stripping away the object’s commercial function to consecrate it to the sacred, thereby ensuring the offering’s permanence,” Isabelle Bollard-Raineau, a regional curator at Pays de la Loire in France, said in the booklet.
The results were initially presented to the local community in a conference, as well as school visits over the last academic year for public education.