Enlarge (credit: hrui)

Earlier this week, the US Senate passed what's being called the ADVANCE Act, for Accelerating Deployment of Versatile, Advanced Nuclear for Clean Energy. Among a number of other changes, the bill would attempt to streamline permitting for newer reactor technology and offer cash incentives for the first companies that build new plants that rely on one of a handful of different technologies. It enjoyed broad bipartisan support both in the House and Senate and now heads to President Biden for his signature.

Given Biden's penchant for promoting his bipartisan credentials, it's likely to be signed into law. But the biggest hurdles nuclear power faces are all economic, rather than regulatory, and the bill provides very little in the way of direct funding that could help overcome those barriers.

Incentives

For reasons that will be clear only to congressional staffers, the Senate version of the bill was attached to an amendment to the Federal Fire Prevention and Control Act. Nevertheless, it passed by a margin of 88-2, indicating widespread (and potentially veto-proof) support. Having passed the House already, there's nothing left but the president's signature.

Enlarge (credit: Aurich Lawson | Getty Images)

It's one of the world's worst-kept secrets that large language models give blatantly false answers to queries and do so with a confidence that's indistinguishable from when they get things right. There are a number of reasons for this. The AI could have been trained on misinformation; the answer could require some extrapolation from facts that the LLM isn't capable of; or some aspect of the LLM's training might have incentivized a falsehood.

But perhaps the simplest explanation is that an LLM doesn't recognize what constitutes a correct answer but is compelled to provide one. So it simply makes something up, a habit that has been termed confabulation.

Figuring out when an LLM is making something up would obviously have tremendous value, given how quickly people have started relying on them for everything from college essays to job applications. Now, researchers from the University of Oxford say they've found a relatively simple way to determine when LLMs appear to be confabulating that works with all popular models and across a broad range of subjects. And, in doing so, they develop evidence that most of the alternative facts LLMs provide are a product of confabulation.

Enlarge (credit: IURII BUKHTA)

A key aspect of humans' evolutionary success is the fact that we don't have to learn how to do things from scratch. Our societies have developed various ways—from formal education to YouTube videos—to convey what others have learned. This makes learning how to do things far easier than learning by doing, and it gives us more space to experiment; we can learn to build new things or handle tasks more efficiently, then pass information on how to do so on to others.

Some of our closer relatives, like chimps and bonobos, learn from their fellow species-members. They don't seem to engage in this iterative process of improvement—they don't, in technical terms, have a cumulative culture where new technologies are built on past knowledge. So, when did humans develop this ability?

Based on a new analysis of stone toolmaking, two researchers are arguing that the ability is relatively recent, dating to just 600,000 years ago. That's roughly the same time our ancestors and the Neanderthals went their separate ways.

Enlarge (credit: NASA, ESA, CSA, Joseph Olmsted (STScI))

Supermassive black holes appear to be present at the center of every galaxy, going back to some of the earliest galaxies in the Universe. And we have no idea how they got there. It shouldn't be possible for them to grow from supernova remnants to supermassive sizes as quickly as they do. And we're not aware of any other mechanism that could form something big enough that extreme growth wouldn't be necessary.

The seeming impossibility of supermassive black holes in the early Universe was already a bit of a problem; the James Webb Space Telescope has only made it worse by finding ever-earlier instances of galaxies with supermassive black holes. In the latest example, researchers have used the Webb to characterize a quasar powered by a supermassive black hole as it existed approximately 750 million years after the Big Bang. And it looks shockingly normal.

Looking back in time

Quasars are the brightest objects in the Universe, powered by actively feeding supermassive black holes. The galaxy surrounding them feeds them enough material that they form bright accretion disks and powerful jets, both of which emit copious amounts of radiation. They're often partly shrouded in dust, which glows from absorbing some of the energy emitted by the black hole. These quasars emit so much radiation that they ultimately drive some of the nearby material out of the galaxy entirely.

Enlarge / This is an intact phage. A tailocin looks like one of these with its head cut off. (credit: iLexx)

Long before humans became interested in killing bacteria, viruses were on the job. Viruses that attack bacteria, termed "phages" (short for bacteriophage), were first identified by their ability to create bare patches on the surface of culture plates that were otherwise covered by a lawn of bacteria. After playing critical roles in the early development of molecular biology, a number of phages have been developed as potential therapies to be used when antibiotic resistance limits the effectiveness of traditional medicines.

But we're relative latecomers in terms of turning phages into tools. Researchers have described a number of cases where bacteria have maintained pieces of disabled viruses in their genomes and converted them into weapons that can be used to kill other bacteria that might otherwise compete for resources. I only just became aware of that weaponization, thanks to a new study showing that this process has helped maintain diverse bacterial populations for centuries.

Evolving a killer

The new work started when researchers were studying the population of bacteria associated with a plant growing wild in Germany. The population included diverse members of the genus Pseudomonas, which can include plant pathogens. Normally, when bacteria infect a new victim, a single strain expands dramatically as it successfully exploits its host. In this case, though, the Pseudomonas population contained a variety of different strains that appeared to maintain a stable competition.

Enlarge (credit: DrAfter123)

The development of gene editing tools, which enable the specific targeting and correction of mutations, hold the promise of allowing us to correct those mutations that cause genetic diseases. However, the technology has been around for a while now—two researchers were critical to its development in 2020—and there have been only a few cases where gene editing has been used to target diseases.

One of the reasons for that is the challenge of targeting specific cells in a living organism. Many genetic diseases affect only a specific cell type, such as red blood cells in sickle-cell anemia, or specific tissue. Ideally, to limit potential side effects, we'd like to ensure that enough of the editing takes place in the affected tissue to have an impact, while minimizing editing elsewhere to limit side effects. But our ability to do so has been limited. Plus, a lot of the cells affected by genetic diseases are mature and have stopped dividing. So, we either need to repeat the gene editing treatments indefinitely or find a way to target the stem cell population that produces the mature cells.

On Thursday, a US-based research team said that they've done gene editing experiments that targeted a high-profile genetic disease: cystic fibrosis. Their technique largely targets the tissue most affected by the disease (the lung), and occurs in the stem cell populations that produce mature lung cells, ensuring that the effect is stable.

Enlarge (credit: Buena Vista Images)

Lots of animals communicate with each other, from tiny mice to enormous whales. But none of those forms of communication share even a small fraction of the richness of human language. Still, finding new examples of complex communications can tell us things about the evolution of language and what cognitive capabilities are needed for it.

On Monday, researchers report what may be the first instance of a human-like language ability in another species. They report that elephants refer to each other by individual names, and the elephant being referred to recognizes when it's being mentioned. The work could be replicated with a larger population and number of calls, but the finding is consistent with what we know about the sophisticated social interactions of these creatures.

What’s in a name?

We use names to refer to each other so often that it's possible to forget just how involved their use is. We recognize formal and informal names that refer to the same individual, even though those names often have nothing to do with the features or history of that person. We easily handle hundreds of names, including those of people we haven't interacted with in decades. And we do this in parallel with the names of thousands of places, products, items, and so on.

Enlarge (credit: Uma Shankar sharma)

June 2023 did not seem like an exceptional month at the time. It was the warmest June in the instrumental temperature record, but monthly records haven't exactly been unusual in a period where the top 10 warmest years on record have all occurred within the last 15 years. And monthly records have often occurred in years that are otherwise unexceptional; at the time, the warmest July on record had occurred in 2019, a year that doesn't stand out much from the rest of the past decade.

But July 2023 set another monthly record, easily eclipsing 2019's high temperatures. Then August set yet another monthly record. And so has every single month since, a string of records that propelled 2023 to the warmest year since we started keeping track.

Yesterday, the European Union's Copernicus Earth-monitoring service announced that we've now gone a full year where every single month has been the warmest version of that month since we've had enough instruments in place to track global temperatures.

Enlarge / A slowly rotating neutron star is still our best guess as to the source of the mystery signals. (credit: Nazarii Neshcherenskyi)

Roughly a year ago, astronomers announced that they had observed an object that shouldn't exist. Like a pulsar, it emitted regularly timed bursts of radio emissions. But unlike a pulsar, those bursts were separated by over 20 minutes. If the 22-minute gap between bursts represents the rotation period of the object, then it is rotating too slowly to produce radio emissions by any known mechanism.

Now, some of the same team (along with new collaborators) are back with the discovery of something that, if anything, is acting even more oddly. The new source of radio bursts, ASKAP J193505.1+214841.0, takes nearly an hour between bursts. And it appears to have three different settings, sometimes producing weaker bursts and sometimes skipping them entirely. While the researchers suspect that, like pulsars, this is also powered by a neutron star, it's not even clear that it's the same class of object as their earlier discovery.

How pulsars pulse

Contrary to the section heading, pulsars don't actually pulse. Neutron stars can create the illusion by having magnetic poles that aren't lined up with their rotational pole. The magnetic poles are a source of constant radio emissions, but as the neutron star rotates, the emissions from the magnetic pole sweep across space in a manner similar to the light from a rotating lighthouse. If Earth happens to be caught up in that sweep, then the neutron star will appear to blink on and off as it rotates.