Like an AGN, a black-hole star is an accreting black hole wrapped in a gas cocoon. But this cocoon’s density is much greater. So great, in fact, that no radiation emitted by the black hole’s infalling matter—not light, not X-rays —can ever directly escape. Instead, the radiation is absorbed by the innermost layer of gas, causing it to emit new radiation, at longer wavelengths, which the next layer will absorb. After many such stages, say the idea’s proponents, the light would emerge from an outer layer similar to a star’s glowing surface, possessing a distinctive red tinge. Such objects would thus look like stars from the outside, but powered by growing black holes instead of nuclear fusion. Although both proposals go some way to explaining the dots’ unique characteristics, neither does so satisfactorily. Critics of the AGN scenario contend that it requires a string of assumptions to explain their unusual appearance. But the advocates of black-hole stars have to explain why a handful of dots have displayed light leaking out of the gas cocoon, something their model forbids. Which side of the debate sounds more convincing can come down to taste. As Jenny Greene, an astronomer at Princeton University, puts it: “People see what they want to see.” What could settle the matter? The answer may lie in the dots’ mass. The two proposals predict wildly different numbers from the same data: AGN-like

objects can be several orders of magnitude heavier than black-hole stars. Writing in Nature, in May, a team led by Ignas Juodzbalis, then at the University of Cambridge, published the first mass measurement of a little red dot. They calculated the speeds with which different patches of the gas cocoon moved, and used them to infer the dot’s mass. The result, about 50m solar masses, was a point for team AGN. But the work has proved controversial. Dr Juodzbalis himself points out that it provides only one data point. What’s more, “I think they can’t do what they said they did,” says Dr Greene. She contends that the authors’ technique works only when measuring the speeds of clearly defined objects, such as two stars, and not the motion within the messy gas cocoons that surround little red dots. More than academic vindication is at stake. Astronomers do not know how the first supermassive black holes put on their weight, for example, nor how that shaped the galaxies in which they sit. The modest masses of black-hole stars would suggest that they started small and grew quickly by feeding on their gas cocoons. But if the AGN advocates are right, somehow the first black holes were massive at birth. On such questions, the little red dots punch far above their weight. ■ This article was downloaded by zlibrary from https://www.economist.com/science-and-technology/2026/11/04/astronomers-are- mystified-by-little-red-dots-in-the-sky

Science & technology | Weights and measures AI models can tinker with one another’s brains For now, the tweaks are relatively small-scale Nov 5th 2026 Numbers such as -0.180, -0.080 and 0.191 do not mean very much to the untrained human eye. Even if a specialist were told that these were an AI model’s “weights”—as its internal parameters are called—they would struggle to reconstruct what sort of model it was, or what it was trained to do. But it seems that a cutting-edge large language model (LLM), such as OpenAI’s GPT-6 Astra, would have no such difficulty. So said Rikiya Takehi, a PhD student at the Massachusetts Institute of Technology, and his collaborators in a blog post published on November 2nd. They claimed that Astra figured out that a second AI model had been trained to play Pong, a basic video game, just by looking at its weights. More impressive still, it fixed a known bug in a small LLM by tweaking its weights, thereby sextupling its score on a popular maths test, and manually

programmed the weights for an AI model capable of playing Go, a board game. If ordinary model design is akin to natural selection, in which external conditions are set to allow the right weights to emerge on their own, the tweaks Astra made were closer to genetic engineering. These claims rapidly caught the attention of others in the field. Some observers speculated that a future AI could tweak its own weights, potentially releasing a runaway train of recursive self-improvement. Others hoped that AI systems could help them peer inside future models to catch possible misbehaviour early. The blog post is no guarantee that either of these will materialise. It shows only relatively simple examples of models finding and fixing flaws in other models that are still thousands of times smaller than Astra itself. What’s more, says Fazl Barez, an AI researcher at Oxford University who was not involved in the work, it is likely that the model’s abilities emerged only because it had been trained on previous human-led experiments to do the same thing. A comprehensive understanding of frontier models remains a long way off. But if it happens, says Dr Barez, “It’ll be the happiest day of my life.”■ This article was downloaded by zlibrary from https://www.economist.com/science-and-technology/2026/11/04/ai-models-can-tinker- with-one-anothers-brains

Science & technology | Ears in the sky Drones don’t just blow up the enemy They can also pick up on radio transmissions to identify future targets Nov 5th 2026 In the David and Goliath war between Ukraine and Russia, drones have proved a great equaliser. Ukraine’s short-range aircraft are cutting Russia’s supply lines to its occupying troops, while long-range versions attack oil refineries, airfields and warehouses as far afield as Siberia. Not all military drones, however, carry warheads. Many are used instead for spying and target location. A lot of these gather their data at optical or infrared frequencies. But some, involved in the field of signals intelligence, or SIGINT, listen out for radio waves. And within that category are those which seek not to understand what is being broadcast, but merely to identify and locate its source, the better to attack it—a field of endeavour called electronic support measures (ESM).

According to Thomas Withington, an electronic-warfare expert at the Royal United Services Institute, a think-tank, the result has been an arms race in which the radio frequencies used to control drones have migrated ever further up and down the spectrum, in attempts to evade detection. Frequency-hopping radios, which skip from one wavelength to another thousands of times a second, are getting faster, so a transmitter may be visible only momentarily; signals are increasingly disguised to merge into background noise. Dr Withington notes the advantages drones bring to ESM. A lorry-mounted system with a ten-metre antenna can detect transmitters roughly 13km away. The same receiver on a drone at an altitude of 1,000 metres can see for 130km. Having antennae on drones is also more useful when an attack may come from any direction. It avoids blind spots caused by topography. The Ukrainians are reluctant to discuss what they are up to. But not all their research is being done in-house. At Eurosatory, a European defence exhibition held in June, Aaronia, a German manufacturer, unveiled a SIGINT electronics package that can be mounted on a drone. The developers say three such drones working in tandem can pinpoint transmitters to within 25 metres from up to 75km away. At 5kg, Aaronia’s product is inconveniently heavy. Sky Spy, based in Delaware, recently demonstrated Agent 001, a 500-gram electronics package that can be mounted on a small Orqa quadcopter. This proved able to detect, identify and locate “hostile” radios as part of an exercise in Estonia. Agent 001 also has on-board processing, linked to the drone’s camera, so it can move towards a transmitter and capture video for its operator to confirm the identity of the source. One limitation of small drones, says Dr Withington, is that the amount of processing capacity they can carry is restricted. Such capacity is essential for sifting through signal data. Drone-control signals may, for example, be hidden in Wi-Fi. New processors and AI software are, however, making small drones vastly more capable. One bit of kit that can be mounted onto a drone, known as Barn Owl, has been developed by NxGenComm of North Carolina. David

Gross, NxGenComm’s marketing director, says Barn Owl follows all traffic in the Wi-Fi range, analysing changes to pick out the signatures of concealed military signals. He claims the process is fast enough to keep up with the most nimble frequency-hopping radios, and powerful enough to scan and analyse an 800MHz-wide swathe of radio signals. Intelligence can go to commanders, or directly to other units. In one proposed set-up, a parliament of Barn Owls would be linked directly to interceptor drones. When they detected an incoming drone, they would launch an interceptor and guide it to the target’s location until it was close enough for visual confirmation by the operator. NxGenComm hopes to deploy Barn Owl to Ukraine in the next few months, and is in discussion with American military customers. As well as locating enemy drones, ESM can also help identify their control frequencies so that they can be jammed. Both sides get round this by flying some drones via fibre-optic control cables, which unspool as they fly. Such drones, which can carry 50km of fibre, emit no radio signals. Similarly, some AI-enabled drones can operate incommunicado, flying and striking targets without operator input. In the meantime, however, the vast majority of drones still rely on radio, and the electronic-warfare arms race continues. Falling behind can be disastrous: if an enemy can jam communications and knock drones out of the sky while remaining invisible, heavy losses are sure to result. ■ This article was downloaded by zlibrary from https://www.economist.com/science-and-technology/2026/11/04/drones-dont-just-blow- up-the-enemy

Science & technology | Show your working OpenAI’s approach to mathematics does not add up Two months after the firm’s initial advances, its lack of transparency is still causing problems Nov 5th 2026 On October 29th leading mathematicians gathered at the Royal Society in London to discuss the future of their field. There was plenty of excitement at the progress artificial intelligence is making, leavened with nostalgia for the good old days of early 2026 when human-led mathematics was the only game in town. Many attendees were also critical of how AI labs have been handling their models’ breakthroughs. In recent months labs have flooded the field with papers that are almost impenetrable to human minds, and kept the details of new results under wraps. Such secrecy, say mathematicians, increases the