An illustration of what the Amniote (early reptile) would look like from 350 million years ago. CREDIT: Martin Ambrozik. Get the Popular Science daily newsletter💡 Breakthroughs, discoveries, and DIY tips sent every weekday. One of the most impactful stories in evolution is getting a rewrite, thanks to the exciting discovery of the earliest known set of reptile footprints. Craig A. Eury and John Eason, two amateur paleontologists exploring the fossil-rich Snowy Plains Formation in Australia, found a rock with an intriguing set of fossilized prints. They brought the intriguing specimen to professional paleontologists, who soon discovered that the roughly 356 million-year-old fossilized claw prints likely belong to an amniote–an early reptile relative.  Though small in stature, amniotes were a large evolutionary leap forward towards land-dwelling, four-limbed animals called tetrapods. The age of these prints suggest that amniotes evolved millions of years earlier than expected, according to a study published May 14 in the journal Nature.  “I’m stunned,” Per Ahlberg, a paleontologist at Uppsala University in Sweden who coordinated the study, said in a statement. “A single track-bearing slab, which one person can lift, calls into question everything we thought we knew about when modern tetrapods evolved.” When fish grew legs Tetrapods include all vertebrates with four limbs that primarily live on land, including everything from frogs to turtles to eagles, to tigers to humans. Their story began as fish left the water between 390 and 360 million years ago. Their descendants began to diversify into the ancestors of modern amphibians and amniotes–the group that includes birds, reptiles, and mammals. Originally, the timeline for how this massive diversification of life occurred was fairly clear-cut. The first tetrapods evolved roughly 390 million years ago during the Devonian period. Amniotoes and the earliest members of the modern groups of animals we see today followed fishapods during the Carboniferous period. Previously, the earliest amniote fossils dated back to  about 320 million years old to the late Carboniferous. Based on this new evidence, researchers concluded the start of the point on the evolutionary tree where the ancestors of amphibians and amniotes split actually happened in the earliest days of the Carboniferous or 356 million years ago.   Proof in the prints The newly discovered 356 million-year-old sandstone slab from this new study potentially changes this entire timeline by about 35 to 40 million years. The well-preserved footprints of long-toed feet with distinct claw impressions at the tips dot the stone and are the earliest known clawed footprints. Two sets of tracks were identified on the stone, seemingly from the same animal.  Footprints are important for paleontologists, as they can indicate the types of behaviors an extinct animal may have exhibited. The team compared the ancient tracks with a modern water monitor (Varanus salvator) lizard, since they have similarly shaped feet to what is seen on the footprints. They examined the spacing between the front and hind footprints against that living lizard’s feet. With these measurements, the team estimates that the ancient amniote may have been around 2.5 feet long, but that the exact proportions of the animal are still unknown. “Claws are present in all early amniotes, but almost never in other groups of tetrapods,” Ahlberg said. “The combination of the claw scratches and the shape of the feet suggests that the track maker was a primitive reptile.” If this new interpretation is correct, it pushes the origin of reptiles, and thus amniotes as a whole, back by roughly 40 million years to the earliest Carboniferous. A new set of fossil reptile footprints uncovered all the way across the globe in Poland are also detailed in the study and bolster the evidence. These European footprints are not as old as the ones from Australia, but are still older than previous specimens.  “The implications of this discovery for the early evolution of tetrapods are profound,” John Long, a study co-author and paleontologist at Flinders University in Australia, said. “All stem-tetrapod and stem-amniote lineages must have originated during the Devonian period – but tetrapod evolution proceeded much faster, and the Devonian tetrapod record is much less complete than we have believed.” A place on the evolutionary tree According to the team, this recalibration of the origin of reptiles has a ripple effect on the whole timeline of tetrapod evolution. Tetrapods must be older than even the earliest amniotes, since it has a deeper branching point on the evolutionary tree.  “It’s all about the relative length of different branches in the tree,” said Ahlberg. “In a family tree based on DNA data from living animals, branches will have different lengths reflecting the number of genetic changes along each branch segment. This does not depend on fossils, so it’s really helpful for studying phases of evolution with a poor fossil record.” The fossil trackways with the different tracks on it highlighted. CREDIT: Flinders University The team believes that amphibians and ammonites split apart further into the Devonian period and were likely a contemporary of the primitive, transitional “fishapod” called Tiktaalik. This evidence indicates that a diverse group of tetrapods existed when only transitional “fishapods” were believed to be dragging themselves around muddy shorelines and starting to explore the land.  If this new theory holds, it is likely that the evolution of tetrapods from aquatic creatures to those fully living on land may have occurred even faster than previously believed.   “The Australian footprint slab is about 50 cm [1.5 feet] across,” said Ahlberg, “and at present it represents the entire fossil record of tetrapods from the earliest Carboniferous of Gondwana – a gigantic supercontinent comprising Africa, South America, Antarctica, Australia and India. Who knows what else lived there?”   “The most interesting discoveries are yet to come and that there is still much to be found in the field,” added study co-author and paleontologist Grzegorz Niedźwiedzki. “These footprints from Australia are just one example of this.” 

Archaeopteryx lived about 150 million years ago during the Jurassic Period. Although the first Archaeopteryx fossil was discovered more than 160 years ago and the prehistoric bird has been well studied, an excellent specimen has yielded new information about the species, including why it could fly so well, according to a report in the journal Nature. Painstaking preparation made the “Chicago Archaeopteryx” (named because it resides in the city’s Field Museum) the best-preserved example of the species. That amount of care meant that CT scans and imaging via UV light have yielded the most vivid Archaeopteryx pictures yet. Scientists visualized its smallest bones as well as soft tissue, including some feathers.New Archaeopteryx Flight Details EmergeIllustration showing Archaeopteryx in life, including its tertial feathers that would have helped it fly. (Image Credit: Michael Rothman)Seeing Archaeopteryx in such detail revealed a long set of feathers called tertials on the its upper arm. Those feathers also place the species more accurately on the evolutionary tree among winged dinosaurs as well as a predecessor to contemporary birds, which also have tertials. “Archaeopteryx isn’t the first dinosaur to have feathers, or the first dinosaur to have ‘wings,’" Jingmai O’Connor a curator at the museum and an author of the paper, said in a press release. “But we think it’s the earliest known dinosaur that was able to use its feathers to fly.”Ancient FlightThe scientists now think they know why Archaeopteryx was such a good flier. Archaeopteryx’s upper arm bone is much longer than those of contemporary birds. That length creates a gap between the long primary feathers and others, as well as between the wings and the body. Air passing through that gap generates lift. Contemporary birds don’t need such long arms, because evolution solved their need to create lift another way: a set of tertial feathers between the bird’s body and the rest of its wing.“Our specimen is the first Archaeopteryx that was preserved and prepared in such a way that we can see its long tertial feathers,” O’Connor said in the release. “These feathers are missing in feathered dinosaurs that are closely related to birds but aren’t quite birds. Their wing feathers stop at the elbow. That tells us that these non-avian dinosaurs couldn’t fly, but Archaeopteryx could. This also adds to evidence that suggests dinosaurs evolved flight more than once — which I think is super exciting.”Other Bird Evolutionary FeaturesImaging also revealed some other key evolutionary features. For instance, seeing details of the tiny bones in the roof of its mouth are helping scientists understand how modern birds eventually developed the ability to move their beaks independent of the part of the skull that holds their brain.“That might not sound exciting, but to people who study bird evolution, it’s really important, because it’s been hypothesized that being able to evolve specialized skulls for different ecological niches might have helped birds evolve into more than 11,000 species today,” O’Connor said in the release.Also, because the scientists got a detailed look at some of the Archaeopteryx’s soft tissues they hypothesize that it may have spent a lot of time walking on the ground, and may even have climbed trees.The Chicago Archaeopteryx fossil went on display at the Field Museum in 2024. Its tiny hollow bones are embedded in a slab of limestone. At about the size of a pigeon, it is among the smallest known fossils of the species.Article SourcesOur writers at Discovermagazine.com use peer-reviewed studies and high-quality sources for our articles, and our editors review for scientific accuracy and editorial standards. Review the sources used below for this article:Before joining Discover Magazine, Paul Smaglik spent over 20 years as a science journalist, specializing in U.S. life science policy and global scientific career issues. He began his career in newspapers, but switched to scientific magazines. His work has appeared in publications including Science News, Science, Nature, and Scientific American.

Google DeepMind has once again used large language models to discover new solutions to long-standing problems in math and computer science. This time the firm has shown that its approach can not only tackle unsolved theoretical puzzles, but improve a range of important real-world processes as well. Google DeepMind's new tool, called AlphaEvolve, uses the Gemini 2.0 family of large language models (LLMs) to produce code for a wide range of different tasks. LLMs are known to be hit and miss at coding. The twist here is that AlphaEvolve scores each of Gemini’s suggestions, throwing out the bad and tweaking the good, in an iterative process, until it has produced the best algorithm it can. In many cases, the results are more efficient or more accurate than the best existing (human-written) solutions. “You can see it as a sort of super coding agent,” says Pushmeet Kohli, a vice president at Google DeepMind who leads its AI for Science teams. “It doesn’t just propose a piece of code or an edit, it actually produces a result that maybe nobody was aware of.” In particular, AlphaEvolve came up with a way to improve the software Google uses to allocate jobs to its many millions of servers around the world. Google DeepMind claims the company has been using this new software across all of its data centers for more than a year, freeing up 0.7% of Google's total computing resources. That might not sound like much, but at Google’s scale it’s huge. Jakob Moosbauer, a mathematician at the University of Warwick in the UK, is impressed. He says the way AlphaEvolve searches for algorithms that produce specific solutions—rather than searching for the solutions themselves—makes it especially powerful. “It makes the approach applicable to such a wide range of problems,” he says. “AI is becoming a tool that will be essential in mathematics and computer science.” AlphaEvolve continues a line of work that Google DeepMind has been pursuing for years. Its vision is that AI can help to advance human knowledge across math and science. In 2022, it developed AlphaTensor, a model that found a faster way to solve matrix multiplications—a fundamental problem in computer science—beating a record that had stood for more than 50 years. In 2023, it revealed AlphaDev, which discovered faster ways to perform a number of basic calculations performed by computers trillions of times a day. AlphaTensor and AlphaDev both turn math problems into a kind of game, then search for a winning series of moves. FunSearch, which arrived in late 2023, swapped out game-playing AI and replaced it with LLMs that can generate code. Because LLMs can carry out a range of tasks, FunSearch can take on a wider variety of problems than its predecessors, which were trained to play just one type of game. The tool was used to crack a famous unsolved problem in pure mathematics. AlphaEvolve is the next generation of FunSearch. Instead of coming up with short snippets of code to solve a specific problem, as FunSearch did, it can produce programs that are hundreds of lines long. This makes it applicable to a much wider variety of problems.     In theory, AlphaEvolve could be applied to any problem that can be described in code and that has solutions that can be evaluated by a computer. “Algorithms run the world around us, so the impact of that is huge,” says Matej Balog, a researcher at Google DeepMind who leads the algorithm discovery team. Survival of the fittest Here’s how it works: AlphaEvolve can be prompted like any LLM. Give it a description of the problem and any extra hints you want, such as previous solutions, and AlphaEvolve will get Gemini 2.0 Flash (the smallest, fastest version of Google DeepMind’s flagship LLM) to generate multiple blocks of code to solve the problem. It then takes these candidate solutions, runs them to see how accurate or efficient they are, and scores them according to a range of relevant metrics. Does this code produce the correct result? Does it run faster than previous solutions? And so on. AlphaEvolve then takes the best of the current batch of solutions and asks Gemini to improve them. Sometimes AlphaEvolve will throw a previous solution back into the mix to prevent Gemini from hitting a dead end. When it gets stuck, AlphaEvolve can also call on Gemini 2.0 Pro, the most powerful of Google DeepMind’s LLMs. The idea is to generate many solutions with the faster Flash but add solutions from the slower Pro when needed. These rounds of generation, scoring, and regeneration continue until Gemini fails to come up with anything better than what it already has. Number games The team tested AlphaEvolve on a range of different problems. For example, they looked at matrix multiplication again to see how a general-purpose tool like AlphaEvolve compared to the specialized AlphaTensor. Matrices are grids of numbers. Matrix multiplication is a basic computation that underpins many applications, from AI to computer graphics, yet nobody knows the fastest way to do it. “It’s kind of unbelievable that it’s still an open question,” says Balog. The team gave AlphaEvolve a description of the problem and an example of a standard algorithm for solving it. The tool not only produced new algorithms that could calculate 14 different sizes of matrix faster than any existing approach, it also improved on AlphaTensor’s record-beating result for multipying two four-by-four matrices. AlphaEvolve scored 16,000 candidates suggested by Gemini to find the winning solution, but that’s still more efficient than AlphaTensor, says Balog. AlphaTensor’s solution also only worked when a matrix was filled with 0s and 1s. AlphaEvolve solves the problem with other numbers too. “The result on matrix multiplication is very impressive,” says Moosbauer. “This new algorithm has the potential to speed up computations in practice.” Manuel Kauers, a mathematician at Johannes Kepler University in Linz, Austria, agrees: “The improvement for matrices is likely to have practical relevance.” By coincidence, Kauers and a colleague have just used a different computational technique to find some of the speedups AlphaEvolve came up with. The pair posted a paper online reporting their results last week. “It is great to see that we are moving forward with the understanding of matrix multiplication,” says Kauers. “Every technique that helps is a welcome contribution to this effort.” Real-world problems Matrix multiplication was just one breakthrough. In total, Google DeepMind tested AlphaEvolve on more than 50 different types of well-known math puzzles, including problems in Fourier analysis (the math behind data compression, essential to applications such as video streaming), the minimum overlap problem (an open problem in number theory proposed by mathematician Paul Erdős in 1955), and kissing numbers (a problem introduced by Isaac Newton that has applications in materials science, chemistry, and cryptography). AlphaEvolve matched the best existing solutions in 75% of cases and found better solutions in 20% of cases.   Google DeepMind then applied AlphaEvolve to a handful of real-world problems. As well as coming up with a more efficient algorithm for managing computational resources across data centers, the tool found a way to reduce the power consumption of Google’s specialized tensor processing unit chips. AlphaEvolve even found a way to speed up the training of Gemini itself, by producing a more efficient algorithm for managing a certain type of computation used in the training process. Google DeepMind plans to continue exploring potential applications of its tool. One limitation is that AlphaEvolve can’t be used for problems with solutions that need to be scored by a person, such as lab experiments that are subject to interpretation.    Moosbauer also points out that while AlphaEvolve may produce impressive new results across a wide range of problems, it gives little theoretical insight into how it arrived at those solutions. That’s a drawback when it comes to advancing human understanding.   Even so, tools like AlphaEvolve are set to change the way researchers work. “I don't think we are finished,” says Kohli. “There is much further that we can go in terms of how powerful this type of approach is.”

Piri piri or African bird’s eye chilli peppersSteidi/Alamy Accidentally made your food too spicy? One day, you might be able to reach for an “anti-spice” condiment to tame the heat of a dish, thanks to the discovery of chemicals found in chilli peppers that counteract their spicy sensation. A chilli’s heat comes from compounds called capsaicinoids, which bind to receptors on nerve fibres within your mouth, sending impulses to the brain that create a burning sensation similar to those caused by a fire’s heat or a sting. Advertisement Chilli aficionados compare the heat of different strains using the Scoville scale, which is based on the concentration of capsaicinoids, yet some varieties aren’t quite as hot as their Scoville rating suggests they should be. To investigate, Devin Peterson at the Ohio State University and his colleagues used an analytical method known as liquid chromatography-mass spectrometry to determine how much of two capsaicinoids – capsaicin and dihydrocapsaicin – were in powdered samples of 10 types of peppers, including chile de árbol, African bird’s eye and Scotch bonnet. Then they gave samples of tomato juice containing powder from the different chillis to a panel of tasters. Each contained the same amount of capsaicin and dihydrocapsaicin – which should have been enough to give all the samples a relatively mild kick of 800 Scoville units. But the tasters perceived the heat from the 10 peppers as different, so Peterson and his colleagues performed additional chemical analyses. This identified three compounds in the chilli powder – capsianoside I, roseoside and gingerglycolipid A – that were present in high quantities in the chillies that weren’t as intense as they should have been according to the Scoville scale. All three compounds are glucosides, molecules that contain the sugar glucose. The latest science news delivered to your inbox, every day. Sign up to newsletter A group of 37 tasters then tested two samples at once, one containing these suspected spice-killing compounds and one without, placed on either side of their tongue to stop an enflamed tongue affecting a second taste test. Their feedback revealed that the compounds decreased the chilli intensity by between 0.7 and 1.2 points on a 15-point scale, on average. “They are effectively anti-spice compounds,” says Peterson. He isn’t yet sure how they work, but says they could be altering receptors on nerve fibres in the mouth in a way that results in a reduction in their burning signals. Knowing about these anti-spice chemicals could allow growers to breed or genetically modify plants so they lack them to create valuable, even hotter chillis, or breed very mild fruit with higher amounts of anti-spice, says Peterson. Using the compounds could also enable the creation of a household ingredient to tone down excessive heat in dishes, he says, or they could work as relievers for severe pain by blocking pain signals. “Sometimes when I’ve ordered food with my kids and it’s too spicy, that’s a no-go,” says Peterson. “So, the idea of having some kind of a natural compound to dial it back may be appealing.” “The way the study was done using the half tongue was very clever,” says Barry Smith at the University of London’s School of Advanced Study, adding that it shows the Scoville scale isn’t a very precise instrument for describing how hot a chilli is. Smith also wonders if the perceived intensity of menthol or mint, which cause a cooling sensation in a similar way to how capsaicinoids cause a burning one, could also be muted by these kinds of compounds. Journal reference:Journal of Agricultural and Food Chemistry DOI: 10.1021/acs.jafc.5c01448 Topics:food and drink

As long as you keep up with regular cleaning and follow the recommended maintenance tips we’ve rounded up here, you shouldn’t have any issues.Is it expensive to maintain quartz countertops?According to Booth, not in the slightest. “Quartz is nonporous,” she says. “So there is no need to routinely seal the surface.” In comparison, countertops such as quartzite, granite, and marble do need to be routinely sealed to maintain their appearance. That being said, you may face some hefty costs if you don’t follow the general recommendations for preventing long-term damage to your countertops, like avoiding extreme heat or scratching them with knives or other sharp objects.Can you use Clorox wipes on quartz countertops?The short answer? Probably not. The long answer? Right this way. Many Clorox wipes are actually bleach-free. You can use them occasionally for deep cleaning, but they shouldn’t be the first thing you grab. You shouldn’t be using Clorox wipes daily, or even often, on quartz countertops because they have citric acid, which can be very harsh on your counter. Long-term use could lead to discoloration of your quartz surface.Can I use Windex on quartz countertops?Windex is considered a base and has a pretty high pH level, which means it could be damaging to quartz. Some professionals say it’s okay to use in small quantities, while others warn against it. For the safest and most sure option to keeping your quartz clean, skip Windex and opt for a mild cleaner like dish soap.In general you should be very careful when using this cleaner. Some Windex formulas rely on the cleaning agent ammonia, which can create very harmful mixtures when combined with bleach—like sodium hypochlorite. This combo can create chloramine vapor, a noxious fume that can irritate your eyes, nose, throat, and airway. In some cases, large quantities can even be fatal.What should you not clean quartz countertops with?In addition to avoiding Clorox and Windex, you shouldn’t use anything with abrasive or harsh chemicals to clean your stone countertops. That means skipping bleach, degreaser, oven cleaner, scouring pads, or other alkaline cleaning solutions or disinfecting formulas with high pH levels. These highly acidic cleaners will damage the countertop and ruin its integrity. If you‘re using them on other items, make sure the spray doesn’t land on the stone surface.To prevent stubborn gunk from building up, it’s best to stick to mild detergents, soapy water and gentle cleaning products. Keep up with a daily maintenance routine rather than trying to address the dirt with a deep cleaning method or harsh cleaners later on. If you happen to have accidents involving nail polish or permanent marker stains, try DIY methods, like using watered-down lemon juice or a dose of rubbing alcohol in a spray bottle.It’s also important to keep quartz surfaces safe from hot pans because they are sensitive to extreme heat. You can use trivets and hot pads to protect granite countertops (which have quartz) from off-the-stove dishes. For food prep that uses sharp objects, especially knives for chopping, cutting boards protect the surface from scratches and divots. And it’s always best to keep coasters around for hot mugs.How to remove stubborn stains from quartz countertopsTo prevent stubborn gunk from building up on quartz countertops, it’s best to stick to mild detergents, soapy water, and gentle cleaning products. Keep up with a daily maintenance routine rather than trying to go back in with a deep cleaning method or harsh cleaners later on. As a general rule of thumb for any stains you come across, Booth recommends first trying mild cleaning solutions before reaching for the stronger, harsher options; it’s better to start slow than risk causing irreversible damage. “I suggest starting with mild dish soap and a microfiber cloth,” she says. “Rub the stain in circular motion and then rinse with clean water.”

Google DeepMind has once again used large language models to discover new solutions to long-standing problems in math and computer science. This time the firm has shown that its approach can not only tackle unsolved theoretical puzzles, but improve a range of important real-world processes as well. Google DeepMind's new tool, called AlphaEvolve, uses the Gemini 2.0 family of large language models (LLMs) to produce code for a wide range of different tasks. LLMs are known to be hit and miss at coding. The twist here is that AlphaEvolve scores each of Gemini’s suggestions, throwing out the bad and tweaking the good, in an iterative process, until it has produced the best algorithm it can. In many cases, the results are more efficient or more accurate than the best existing (human-written) solutions. “You can see it as a sort of super coding agent,” says Pushmeet Kohli, a vice president at Google DeepMind who leads its AI for Science teams. “It doesn’t just propose a piece of code or an edit, it actually produces a result that maybe nobody was aware of.” In particular, AlphaEvolve came up with a way to improve the software Google uses to allocate jobs to its many millions of servers around the world. Google DeepMind claims the company has been using this new software across all of its data centers for more than a year, freeing up 0.7% of Google's total computing resources. That might not sound like much, but at Google’s scale it’s huge. Jakob Moosbauer, a mathematician at the University of Warwick in the UK, is impressed. He says the way AlphaEvolve searches for algorithms that produce specific solutions—rather than searching for the solutions themselves—makes it especially powerful. “It makes the approach applicable to such a wide range of problems,” he says. “AI is becoming a tool that will be essential in mathematics and computer science.” AlphaEvolve continues a line of work that Google DeepMind has been pursuing for years. Its vision is that AI can help to advance human knowledge across math and science. In 2022, it developed AlphaTensor, a model that found a faster way to solve matrix multiplications—a fundamental problem in computer science—beating a record that had stood for more than 50 years. In 2023, it revealed AlphaDev, which discovered faster ways to perform a number of basic calculations performed by computers trillions of times a day. AlphaTensor and AlphaDev both turn math problems into a kind of game, then search for a winning series of moves. FunSearch, which arrived in late 2023, swapped out game-playing AI and replaced it with LLMs that can generate code. Because LLMs can carry out a range of tasks, FunSearch can take on a wider variety of problems than its predecessors, which were trained to play just one type of game. The tool was used to crack a famous unsolved problem in pure mathematics. AlphaEvolve is the next generation of FunSearch. Instead of coming up with short snippets of code to solve a specific problem, as FunSearch did, it can produce programs that are hundreds of lines long. This makes it applicable to a much wider variety of problems.     In theory, AlphaEvolve could be applied to any problem that can be described in code and that has solutions that can be evaluated by a computer. “Algorithms run the world around us, so the impact of that is huge,” says Matej Balog, a researcher at Google DeepMind who leads the algorithm discovery team. Survival of the fittest Here’s how it works: AlphaEvolve can be prompted like any LLM. Give it a description of the problem and any extra hints you want, such as previous solutions, and AlphaEvolve will get Gemini 2.0 Flash (the smallest, fastest version of Google DeepMind’s flagship LLM) to generate multiple blocks of code to solve the problem. It then takes these candidate solutions, runs them to see how accurate or efficient they are, and scores them according to a range of relevant metrics. Does this code produce the correct result? Does it run faster than previous solutions? And so on. AlphaEvolve then takes the best of the current batch of solutions and asks Gemini to improve them. Sometimes AlphaEvolve will throw a previous solution back into the mix to prevent Gemini from hitting a dead end. When it gets stuck, AlphaEvolve can also call on Gemini 2.0 Pro, the most powerful of Google DeepMind’s LLMs. The idea is to generate many solutions with the faster Flash but add solutions from the slower Pro when needed. These rounds of generation, scoring, and regeneration continue until Gemini fails to come up with anything better than what it already has. Number games The team tested AlphaEvolve on a range of different problems. For example, they looked at matrix multiplication again to see how a general-purpose tool like AlphaEvolve compared to the specialized AlphaTensor. Matrices are grids of numbers. Matrix multiplication is a basic computation that underpins many applications, from AI to computer graphics, yet nobody knows the fastest way to do it. “It’s kind of unbelievable that it’s still an open question,” says Balog. The team gave AlphaEvolve a description of the problem and an example of a standard algorithm for solving it. The tool not only produced new algorithms that could calculate 14 different sizes of matrix faster than any existing approach, it also improved on AlphaTensor’s record-beating result for multipying two four-by-four matrices. AlphaEvolve scored 16,000 candidates suggested by Gemini to find the winning solution, but that’s still more efficient than AlphaTensor, says Balog. AlphaTensor’s solution also only worked when a matrix was filled with 0s and 1s. AlphaEvolve solves the problem with other numbers too. “The result on matrix multiplication is very impressive,” says Moosbauer. “This new algorithm has the potential to speed up computations in practice.” Manuel Kauers, a mathematician at Johannes Kepler University in Linz, Austria, agrees: “The improvement for matrices is likely to have practical relevance.” By coincidence, Kauers and a colleague have just used a different computational technique to find some of the speedups AlphaEvolve came up with. The pair posted a paper online reporting their results last week. “It is great to see that we are moving forward with the understanding of matrix multiplication,” says Kauers. “Every technique that helps is a welcome contribution to this effort.” Real-world problems Matrix multiplication was just one breakthrough. In total, Google DeepMind tested AlphaEvolve on more than 50 different types of well-known math puzzles, including problems in Fourier analysis (the math behind data compression, essential to applications such as video streaming), the minimum overlap problem (an open problem in number theory proposed by mathematician Paul Erdős in 1955), and kissing numbers (a problem introduced by Isaac Newton that has applications in materials science, chemistry, and cryptography). AlphaEvolve matched the best existing solutions in 75% of cases and found better solutions in 20% of cases.   Google DeepMind then applied AlphaEvolve to a handful of real-world problems. As well as coming up with a more efficient algorithm for managing computational resources across data centers, the tool found a way to reduce the power consumption of Google’s specialized tensor processing unit chips. AlphaEvolve even found a way to speed up the training of Gemini itself, by producing a more efficient algorithm for managing a certain type of computation used in the training process. Google DeepMind plans to continue exploring potential applications of its tool. One limitation is that AlphaEvolve can’t be used for problems with solutions that need to be scored by a person, such as lab experiments that are subject to interpretation.    Moosbauer also points out that while AlphaEvolve may produce impressive new results across a wide range of problems, it gives little theoretical insight into how it arrived at those solutions. That’s a drawback when it comes to advancing human understanding.   Even so, tools like AlphaEvolve are set to change the way researchers work. “I don't think we are finished,” says Kohli. “There is much further that we can go in terms of how powerful this type of approach is.”

Remedy Entertainment has a history of being experimental with genre favorites like Alan Wake 2 and Control, but largely sticking to the (comparatively) safe waters of premium single-player games. Now Remedy is using the setting of Control as a launching pad for a live service co-op shooter called FBC Firebreak, their first crack at an online co-op game about contending with The Oldest House following a new lockdown.Producing an online live service game these days is an increasingly risky business that many developers have struggled with. A key factor is that playing a live service game and keeping up with the content and other players is time-consuming—and only so many games can operate within that space at a given time. With that, the developers at Remedy Entertainment have built their take on an online live service game that not only allows players to tune the activities to match their pace, but also offer a way for players to take in as much as they want when they want.In a chat with Game Developer, game director Mike Kayatta explained that Remedy is expanding the 'Remedy Connected Universe' further, while also building an online co-op game that's all about getting players right into the action and progressing at their own pace—all to build a sustainable live service that can attract players without plenty of time on their hands, and how they're ensuring players who don't have a lot of free time won't feel left behind.Related:An online game built around the player's scheduleIn FBC Firebreak, squads of 3 players are tasked with completing various missions within The Oldest House—the extradimensional office building from the original Control—using jerry-rigged equipment and weapons to contain rogue Objects of Power or halt the invasion of the otherworldly monsters known as the Hiss. FBC Firebreak stands out in the live service space by providing more surreal objectives, like a mission called "Paper Chase," where players hunt supernatural sticky notes.According to Kayatta, FBC Firebreak and its first-person, co-op-driven gameplay have been long brewing at Remedy. The concept itself stretches back to the original's conceit of contending with a setting that's actively working against players."One of my favorite pieces of Control concept art is of an FBC employee trying to play a game of pool, but the floor has shifted one end of the table just enough to ruin his game. This is pretty much how we think about the Oldest House in FBC: Firebreak," said the game director. "It's a powerful and mysterious building, yes, but there's also a certain level of canonical 'grounded absurdity,' it produces that's perfect for our type of experience."Related:Image via Remedy Entertainment."But even aside from those tonal opportunities, the House's shifting and unknowable nature, and the fact that it can connect to other dimensions, gives us all the room we'll ever need to do what's best for Firebreak without violating what makes Control, Control."On the surface, the concept of FBC Firebreak recalls similar efforts from other developers attempting a stand-alone multiplayer game based on a single-player experience, another example being the upcoming Elden Ring: Nightreign. What makes FBC Firebreak particularly interesting is that it's not only leveraging the weirdness of the Control setting, but it's also an online game that seeks to remedy (pun intended) a growing problem with live-service games. Simply put, there's never enough time to play all the games out there while meeting the investments they expect from players.Kayatta stated that FBC Firebreak was all about removing the familiar pain points and sense of FOMO found in online games. To them, the Control spin-off aimed to be a game that players could play "on their terms, and not on the games."Related:Amusingly, Kayatta said the team at Remedy took inspiration from their own busy schedules as game developers, studying the pace of how they kept up with modern online games.FBC Firebreak lets players set their own progression paceThe developers of FBC Firebreak seek to solve the live-service dilemma with two solutions. First, all players can unlock key weapons and upgrades for their Crisis Kits (the game's take on class types), all from playing the game's set of missions and collecting resources. As players rank up their Requisitions—the game's take on the unlock system—they can spend points to unlock whatever they like, which includes new weapons, kit upgrades, and other add-ons to modify their loadouts.The second solution, and most interesting to FBC Firebreak's structure, is the customizable objectives and encounter difficulties. Before each job, the team leader can tune the enemy variety and spawn rate to different intensity levels while also setting the number of objectives needed to clear the mission.This means players can can have a mission with limited objectives but high enemy frequency, presenting a one-stage encounter with an onslaught of enemies. Conversely, they can customize the same mission to have a larger set of objectives, but with a lower frequency of enemies, making it more of an atmospheric, task-focused mission that culminates in a boss encounter.Image via Remedy Entertainment."With a multiplayer game, it's important to consider how your game slots into people's lives," he said. "In co-op, especially, friend groups are stepping in with an unpredictable set of needs: How experienced is each group member? How much time do they have? Are they treating the game session like a phone call, where they just want something to do in the background while talking about something personal?""And you can't solve this for a player forever because the same person who wants to zone out one night for thirty minutes sometimes turns around and wants to hyper-coordinate for two hours the next night. This goes back to putting the game on the players' terms instead of ours; we wanted to give you those controls so that Firebreak can meet you where you are on any given day."FBC Firebreak's approach to challenge and tuning of the experience is flexible by design, which feels like an oddity among other games that push players towards timed battle passes or monthly events.According to Kayatta, building the game's structure and more flexible approach to progression was all about removing potential barriers that so many other live service games have, to provide an online game that players of whatever time investment can get something out of. Remedy axed other popular live service features like dense statistics (think damage, buff, and debuff percentages) because it made players care about "min-maxing and tracking metas.""That in turn, can create barriers to people who don't want to be pressured into thinking about anything that's not directly in front of them," he said.It's an unconventional strategy to drive retention by asking players for less engagement, but Kayatta said targeting those busy everyday players will be key to FBC Firebreak's success."We just wanted to make a game that players without a lot of time could enjoy, which feels like it describes a lot of gamers these days," he concluded. "Some people lose their time to work or to their kids. Others are among the hundreds of incredible games destroying their backlogs. Some poor suckers, like me, lose time to both.""Too many games ask too much of their players, especially in the multiplayer space."

Cool Finds Archaeologists Say They’ve Discovered a Hidden Chamber Where Elites Used Hallucinogens 2,500 Years Ago Unearthed in Peru, the small underground room may have been used for rituals involving psychoactive drugs. New research suggests these “exclusive” events were reserved for the elite An artistic rendering of the stone chamber where the tubes were discovered Daniel Contrera Long before the rise of the Inca empire, a group called the Chavín people thrived in ancient Peru. They’re known for their elaborate stone structures, which were located at an archaeological site now known as Chavín de Huántar in the Andean highlands. During recent excavations at the site, researchers found dozens of hollow bones packed with sediment. They think the artifacts could be ancient drug paraphernalia. “The tubes are analogous to the rolled-up bills that high-rollers snort cocaine through in the movies,” Daniel Contreras, an archaeologist at the University of Florida and a co-author of the study, tells Live Science’s Kristina Killgrove. The site of Chavín de Huántar, located in Peru at an elevation of 10,000 feet, holds the ruins of several monumental buildings. Daniel Contreras The tubes were found in small underground chambers, where experts think they may have been used by Chavín elites, according to a recent study published in the journal PNAS. Chemical and microscopic analyses of the tubes revealed traces of nicotine and vilca bean, a hallucinogen related to the drug DMT. Vilca was commonly used among pre-Columbian populations of the Andes. When smoked or consumed, vilca would bring about an altered mental state. Many ancient cultures used hallucinogens communally, but the Chavín rituals appear to have been “exclusive,” according to a statement from the university. These rituals may have taken place in rooms that held “only a handful of participants at a time, creating an air of mystique and control.” “This is compelling evidence that psychoactive plants were part of formalized and tightly-controlled rituals rather than individual vision-quests or shamanic healing practices,” Contreras tells CBS News’ Emily Mae Czachor. “As such, they seem to have been an important element in the long-term transition from small egalitarian societies to large stratified ones, where social, political and economic inequality were thought of as normal and to be expected rather than unusual.” Snuff tubes made from hollowed bones may have been used to inhale hallucinogenic drugs. Daniel Contreras The Chavín society flourished between roughly 900 and 200 B.C.E. Chavín de Huántar is known for its intricate stone carvings, which often depicted “animal-human hybrids or transformations of human into beast,” and its large network of tunnels, writes Science’s Lizzie Wade. The area is full of imported seashells and obsidian, and Chavín art has been discovered throughout the Andes and on the Peruvian coast, suggesting a “broad cultural reach.” “Chavín was part of the first big moment in Andean prehistory when people, ideas and goods were circulating quite extensively,” Contreras tells Science. Depictions of psychoactive effects in Chavín iconography PNAS For many years, historians assumed the Chavín people conducted rituals using drugs. “What’s exciting about this paper is that, for the first time, we have actual evidence,” José Capriles, an archaeologist at Pennsylvania State University who studies ancient psychoactive drug use but wasn’t involved in the study, tells Science. The researchers think the drugs may have influenced the Chavín society’s class system. By restricting access to these “profound, even terrifying” experiences, Chavín rulers could have convinced their people that “leadership was intertwined with mystical power and part of the natural order,” per the statement. “Taking psychoactives was not just about seeing visions,” says Contreras in the statement. “It was part of a tightly controlled ritual, likely reserved for a select few, reinforcing the social hierarchy.” Get the latest stories in your inbox every weekday.

While EA Motive has been quiet about its ongoing development of a game based on the Iron Man franchise, a new job listing has been spotted on EA’s jobs website that indicates that single-player will be a major component of the game. The job listing is for a Level Designer that would be in charge of designing “major singleplayer levels”. Along with designing levels, the responsibilities for the job posting include bringing the levels through every phase of development, from concept all the way up to polishing. The level designer will also need to incorporate feedback from research and playtest data, while also communicating with the rest of the team when it comes to feedback on the levels they design. According to the posting, the position requires a potential level designer to have experience with single-player action RPG titles, as well as creative skills when it comes to conceptualising and designing environments. The posting also requires familiarity with Epic’s Unreal Engine. The last time EA Motive spoke about Iron Man was with general manager Patrick Klaus revealing that the team in the studio working on the game had hit a major internal milestone. In his update, Klaus mentioned that the project, led by Olivier Proulx and Ian Frazier, has laid “a robust foundation for the journey ahead.” The major internal milestone would likely have to do with the pre-production phase of the development of Iron Man. Back in October 2023, Klaus had revealed that EA Motive was still in the early stages of the game’s development, having started work on the title after wrapping up the Dead Space remake earlier that year. “Since [Dead Space’s launch], our focus has switched to the Iron Man project, giving us the opportunity to work (and play!) with another extraordinary property,” Klaus wrote back then. “We’re still early in pre-production and taking our time to make sure we set the best possible foundation for development. But already, we’re finding so many opportunities to explore story, design and gameplay with Iron Man, and it’s exciting to see the fantasy start coming alive.” While we don’t yet know too much about the Iron Man project, back in 2023, author Ryan North had announced that he was joining the writing team for the game. Nolan, who has worked on a host of comic books (Jughead, Adventure Time, The Unbeatable Squirrel Girl) as well as 2022 video game Lost in Random, said that “You are not ready for how amazing this game is going to be.” Owing to the general silence regarding the game, EA had confirmed that development on Iron Man, along with another project based on superhero Black Panther, hadn’t been cancelled. The statement came in light of lay-offs at the time affecting 670 employees throughout the company. Currently, no further details about EA Motive’s Iron Man game have been revealed, including what platforms it would be coming to, or even a potential launch window. Going by EA’s general multiplatform release strategy, however, the title is likely being made for PC and consoles.