It’s a regrettable reality that there is never enough time to cover all the interesting scientific stories we come across each month. In the past, we’ve featured year-end roundups of cool science stories we (almost) missed. This year, we’re experimenting with a monthly collection. November’s list includes forensic details of the medieval assassination of a Hungarian duke, why woodpeckers grunt when they peck, and more evidence that X’s much-maligned community notes might actually help combat the spread of misinformation after all.

An assassinated medieval Hungarian duke

Credit: Tamás Hajdu et al., 2026

Back in 1915, archaeologists discovered the skeletal remains of a young man in a Dominican monastery on Margaret Island in Budapest, Hungary. The remains were believed to be those of Duke Bela of Masco, grandson of the medieval Hungarian King Bela IV. Per historical records, the young duke was brutally assassinated in 1272 by a rival faction and his mutilated remains were recovered by the duke’s sister and niece and buried in the monastery.

The identification of the remains was based on a contemporary osteological analysis, but they were subsequently lost and only rediscovered in 2018. A paper published in the journal Forensic Science International: Genetics has now confirmed that identification and shed more light on precisely how the duke died. (A preprint is available on bioRxiv.]

An interdisciplinary team of researchers performed various kinds of bioarchaeological analysis on the remains. including genetic testing, proteomics, 3D modeling, and radiocarbon dating. The resulting data definitively proves that the skeleton is indeed that of Duke Bela of Masco.

The authors were also able to reconstruct the manner of the duke’s death, concluding that this was a coordinated attack by three people. One attacked from the front while the other two attacked from the left and right sides, and the duke was facing his assassins and tried to defend himself. The weapons used were most likely a saber and a long sword, and the assassins kept raining down blows even after the duke had fallen to the ground. The authors concluded that while the attack was clearly planned, it was also personal and fueled by rage or hate.

DOI: Forensic Science International: Genetics, 2025. 10.1016/j.fsigen.2025.103381  (About DOIs).

Why woodpeckers grunt when they peck

Credit: Joshlaymon /CC BY-SA 3.0

Woodpeckers energetically drum away at tree trunks all day long with their beaks and yet somehow never seem to get concussions, despite the fact that such drumming can produce deceleration forces as high as 1,200 g’s. (Humans suffer concussions with a sudden deceleration of just 100 g’s.) While popular myth holds that woodpecker heads are structured in such a way to absorb the shock, and there has been some science to back that up, more recent research found that their heads act more like hammers than shock absorbers. A paper published in the Journal of Experimental Biology sheds further light on the biomechanics of how woodpeckers essentially turn themselves into hammers and reveals that the birds actually grunt as they strike wood.

The authors caught eight wild downy woodpeckers and recorded them drilling and tapping on pieces of hardwood in the lab for three days, while also measuring electrical signals in their heads, necks, abdomens, tails, and leg muscles. Analyzing the footage, they found that woodpeckers use their hip flexors and front neck muscles to propel themselves forward as they peck while tipping their heads back and bracing themselves using muscles at the base of the skull and back of the neck. The birds use abdominal muscles for stability and brace for impact using their tail muscles to anchor their bodies against a tree. As for the grunting, the authors noted that it’s a type of breathing pattern used by tennis players (and martial artists) to boost the power of a strike.

DOI: Journal of Experimental Biology, 2025. 10.1242/jeb.251167  (About DOIs).

Raisins turn water into wine

Credit: Kyoto University

Fermentation has been around in some form for millennia, relying on alcohol-producing yeasts like Saccharomyces cerevisiae; cultured S. cerevisiae is still used by winemakers today. It’s long been thought that winemakers in ancient times stored fresh crushed grapes in jars and relied on natural fermentation to work its magic, but recent studies have called this into question by demonstrating that S. cerevisiae colonies usually don’t form on fresh grape skins. But the yeast does like raisins, as Kyoto University researchers recently discovered. They’ve followed up that earlier work with a paper published in Scientific Reports, demonstrating that it’s possible to use raisins to turn water into wine.

The authors harvested fresh grapes and dried them for 28 days. Some were dried using an incubator, some were sun-dried, and a third batch was dried using a combination of the two methods. The researchers then added the resulting raisins to bottles of water—three samples for each type of drying process—sealed the bottles, and stored them at room temperature for two weeks. One incubator-dried sample and two combo samples successfully fermented, but all three of the sun-dried samples did so, and at higher ethanol concentrations. Future research will focus on identifying the underlying molecular mechanisms. And for those interested in trying this at home, the authors warn that it only works with naturally sun-dried raisins, since store-bought varieties have oil coatings that block fermentation.

DOI: Scientific Reports, 2025. 10.1038/s41598-025-23715-3  (About DOIs).

An octopus-inspired pigment

Credit: Charlotte Seid

Octopuses, cuttlefish, and several other cephalopods can rapidly shift the colors in their skin thanks to that skin’s unique complex structure, including layers of chromatophores, iridophores, and leucophores. A color-shifting natural pigment called xanthommatin also plays a key role, but it’s been difficult to study because it’s hard to harvest enough directly from animals, and lab-based methods of making the pigment are labor-intensive and don’t yield much. Scientists at the University of San Diego have developed a new method for making xanthommatin in substantially larger quantities, according to a paper published in Nature Biotechnology.

The issue is that trying to get microbes to make foreign compounds creates a metabolic burden, and the microbes hence resist the process, hindering yields. The USD team figured out how to trick the cells into producing more xanthommatin by genetically engineering them in such a way that making the pigment was essential to a cell’s survival. They achieved yields of between 1 and 3 grams per liter, compared to just five milligrams of pigment per liter using traditional approaches. While this work is proof of principle, the authors foresee such future applications as photoelectronic devices and thermal coatings, dyes, natural sunscreens, color-changing paints, and environmental sensors. It could also be used to make other kinds of chemicals and help industries shift away from older methods that rely on fossil fuel-based materials.

DOI: Nature Biotechnology, 2025. 10.1038/s41587-025-02867-7  (About DOIs).

A body-swap robot

Credit: Sachi Wickramasinghe/UBC Media Relations

Among the most serious risks facing older adults is falling. According to the authors of a paper published in Science Robotics, standing upright requires the brain to coordinate signals from the eyes, inner ears, and feet to counter gravity, and there’s a natural lag in how fast this information travels back and forth between brain and muscles. Aging and certain diseases like diabetic neuropathy and multiple sclerosis can further delay that vital communication; the authors liken it to steering a car with a wheel that responds half a second late. And it’s a challenge to directly study the brain under such conditions.

That’s why researchers at the University of British Columbia built a large “body swap” robotic platform. Subjects stood on force plates attached to a motor-driven backboard to reproduce the physical forces at play when standing upright: gravity, inertia, and “viscosity,” which in this case describes the damping effect of muscles and joints that allow us to lean without falling. The platform is designed to subtly alter those forces and also add a 200-millisecond delay.

The authors tested 20 participants and found that lowering inertia and making the viscosity negative resulted in similar instability to that which resulted from a signal delay. They then brought in ten new subjects to study whether adjusting body mechanics could compensate for information delays. They found that adding inertia and viscosity could at least partially counter the instability that arose from signal delay—essentially giving the body a small mechanical boost to help the brain maintain balance. The eventual goal is to design wearables that offer gentle resistance when an older person starts to lose their balance, and/or help patients with MS, for example, adjust to slower signal feedback.

DOI: Science Robotics, 2025. 10.1126/scirobotics.adv0496  (About DOIs).

X community notes might actually work

Credit: Huaxia Rui

Earlier this year, Elon Musk claimed that X’s community notes feature needed tweaking because it was being gamed by “government & legacy media” to contradict Trump—despite vigorously defending the robustness of the feature against such manipulation in the past. A growing body of research seems to back Musk’s earlier stance.

For instance, last year Bloomberg pointed to several studies suggesting that crowdsourcing worked just as well as using professional fact-checkers when assessing the accuracy of news stories. The latest evidence that crowd-sourcing fact checks can be effective at curbing misinformation comes from a paper published in the journal Information Systems Research, which found that X posts with public corrections were 32 percent more likely to be deleted by authors.

Co-author Huaxia Rui of the University of Rochester pointed out that community notes must meet a threshold before they will appear publicly on posts, while those that do not remain hidden from public view. Seeing a prime opportunity in the arrangement, Rui et al. analyzed 264,600 X posts that had received at least one community note and compared those just above and just below that threshold. The posts were collected from two different periods: June through August 2024, right before the US presidential election (when misinformation typically surges), and the post-election period of January and February 2025.

The fact that roughly one-third of authors responded to public community notes by deleting the post suggests that the built-in dynamics of social media (e.g., status, visibility, peer feedback) might actually help improve the spread of misinformation as intended. The authors concluded that crowd-checking “strikes a balance between First Amendment rights and the urgent need to curb misinformation.” Letting AI write the community notes, however, is probably still a bad idea.

DOI: Information Systems Research, 2025. 10.1287/isre.2024.1609  (About DOIs).

Read full article

Comments

Researchers from MIT, Northeastern University, and Meta recently released a paper suggesting that large language models (LLMs) similar to those that power ChatGPT may sometimes prioritize sentence structure over meaning when answering questions. The findings reveal a weakness in how these models process instructions that may shed light on why some prompt injection or jailbreaking approaches work, though the researchers caution their analysis of some production models remains speculative since training data details of prominent commercial AI models are not publicly available.

The team, led by Chantal Shaib and Vinith M. Suriyakumar, tested this by asking models questions with preserved grammatical patterns but nonsensical words. For example, when prompted with “Quickly sit Paris clouded?” (mimicking the structure of “Where is Paris located?”), models still answered “France.”

This suggests models absorb both meaning and syntactic patterns, but can overrely on structural shortcuts when they strongly correlate with specific domains in training data, which sometimes allows patterns to override semantic understanding in edge cases. The team plans to present these findings at NeurIPS later this month.

As a refresher, syntax describes sentence structure—how words are arranged grammatically and what parts of speech they use. Semantics describes the actual meaning those words convey, which can vary even when the grammatical structure stays the same.

Semantics depends heavily on context, and navigating context is what makes LLMs work. The process of turning an input, your prompt, into an output, an LLM answer, involves a complex chain of pattern matching against encoded training data.

To investigate when and how this pattern-matching can go wrong, the researchers designed a controlled experiment. They created a synthetic dataset by designing prompts in which each subject area had a unique grammatical template based on part-of-speech patterns. For instance, geography questions followed one structural pattern while questions about creative works followed another. They then trained Allen AI’s Olmo models on this data and tested whether the models could distinguish between syntax and semantics.

Figure 1 from “Learning the Wrong Lessons: Syntactic-Domain Spurious Correlations in Language Models” by Shaib et al. Credit: Shaib et al.

The analysis revealed a “spurious correlation” where models in these edge cases treated syntax as a proxy for the domain. When patterns and semantics conflict, the research suggests, the AI’s memorization of specific grammatical “shapes” can override semantic parsing, leading to incorrect responses based on structural cues rather than actual meaning.

In layperson terms, the research shows that AI language models can become overly fixated on the style of a question rather than its actual meaning. Imagine if someone learned that questions starting with “Where is…” are always about geography, so when you ask “Where is the best pizza in Chicago?”, they respond with “Illinois” instead of recommending restaurants based on some other criteria. They’re responding to the grammatical pattern (“Where is…”) rather than understanding you’re asking about food.

This creates two risks: models giving wrong answers in unfamiliar contexts (a form of confabulation), and bad actors exploiting these patterns to bypass safety conditioning by wrapping harmful requests in “safe” grammatical styles. It’s a form of domain switching that can reframe an input, linking it into a different context to get a different result.

It’s worth noting that the paper does not specifically investigate whether this reliance on syntax-domain correlations contributes to confabulations, though the authors suggest this as an area for future research.

When patterns and meaning conflict

To measure the extent of this pattern-matching rigidity, the team subjected the models to a series of linguistic stress tests, revealing that syntax often dominates semantic understanding.

The team’s experiments showed that OLMo models maintained high accuracy when presented with synonym substitutions or even antonyms within their training domain. OLMo-2-13B-Instruct achieved 93 percent accuracy on prompts with antonyms substituted for the original words, nearly matching its 94 percent accuracy on exact training phrases. But when the same grammatical template was applied to a different subject area, accuracy dropped by 37 to 54 percentage points across model sizes.

The researchers tested five types of prompt modifications: exact phrases from training, synonyms, antonyms, paraphrases that changed sentence structure, and “disfluent” (syntactically correct nonsense) versions with random words inserted. Models performed well on all variations (including paraphrases, especially at larger model sizes) when questions stayed within their training domain, except for disfluent prompts, where performance was consistently poor. Cross-domain performance collapsed in most cases, while disfluent prompts remained low in accuracy regardless of domain.

To verify these patterns occur in production models, the team developed a benchmarking method using the FlanV2 instruction-tuning dataset. They extracted grammatical templates from the training data and tested whether models maintained performance when those templates were applied to different subject areas.

Figure 4 from &#8220;Learning the Wrong Lessons: Syntactic-Domain<br />Spurious Correlations in Language Models&#8221; by Shaib et al. Credit: Shaib et al.

Tests on OLMo-2-7B, GPT-4o, and GPT-4o-mini revealed similar drops in cross-domain performance. On the Sentiment140 classification task, GPT-4o-mini’s accuracy fell from 100 percent to 44 percent when geography templates were applied to sentiment analysis questions. GPT-4o dropped from 69 percent to 36 percent. The researchers found comparable patterns in other datasets.

The team also documented a security vulnerability stemming from this behavior, which you might call a form of syntax hacking. By prepending prompts with grammatical patterns from benign training domains, they bypassed safety filters in OLMo-2-7B-Instruct. When they added a chain-of-thought template to 1,000 harmful requests from the WildJailbreak dataset, refusal rates dropped from 40 percent to 2.5 percent.

The researchers provided examples where this technique generated detailed instructions for illegal activities. One jailbroken prompt produced a multi-step guide for organ smuggling. Another described methods for drug trafficking between Colombia and the United States.

Limitations and uncertainties

The findings come with several caveats. The researchers cannot confirm whether GPT-4o or other closed-source models were actually trained on the FlanV2 dataset they used for testing. Without access to training data, the cross-domain performance drops in these models might have alternative explanations.

The benchmarking method also faces a potential circularity issue. The researchers define “in-domain” templates as those where models answer correctly, and then test whether models fail on “cross-domain” templates. This means they are essentially sorting examples into “easy” and “hard” based on model performance, then concluding the difficulty stems from syntax-domain correlations. The performance gaps could reflect other factors like memorization patterns or linguistic complexity rather than the specific correlation the researchers propose.

Table 2 from &#8220;Learning the Wrong Lessons: Syntactic-Domain Spurious Correlations in Language Models&#8221; by Shaib et al. Credit: Shaib et al.

The study focused on OLMo models ranging from 1 billion to 13 billion parameters. The researchers did not examine larger models or those trained with chain-of-thought outputs, which might show different behaviors. Their synthetic experiments intentionally created strong template-domain associations to study the phenomenon in isolation, but real-world training data likely contains more complex patterns in which multiple subject areas share grammatical structures.

Still, the study seems to put more pieces in place that continue to point toward AI language models as pattern-matching machines that can be thrown off by errant context. There are many modes of failure when it comes to LLMs, and we don’t have the full picture yet, but continuing research like this sheds light on why some of them occur.

Read full article

Comments

It’s a critical time for companies competing to develop a commercial successor to the International Space Station. NASA is working with several companies, including Axiom Space, Voyager Technologies, Blue Origin, and Vast, to develop concepts for private stations where it can lease time for its astronauts.

The space agency awarded Phase One contracts several years ago and is now in the final stages of writing requirements for Phase Two after asking for feedback from industry partners in September. This program is known as Commercial LEO Destinations, or CLDs in industry parlance.

Time is running out for NASA if it wants to establish continuity from the International Space Station, which will reach its end of life in 2030, with a follow-on station ready to go before then.

One of the more intriguing companies in the competition is Voyager Technologies, which recently announced a strategic investment from Janus Henderson, a global investment firm. In another sign that the competition is heating up, Voyager also just hired John Baum away from Vast, where he was the company’s business development leader.

To get a sense of this competition and how Voyager is coming along with its Starlab space station project, Ars spoke with the firm’s chairman, Dylan Taylor. This conversation has been lightly edited for clarity.

Ars: I know a lot of the companies working on CLDs are actively fundraising right now. How is this coming along for Voyager and Starlab?

Dylan Taylor: Fundraising is going quite well. You saw the Janus announcement. That’s significant for a few reasons. One is, it’s a significant investment. Of course, we’re not disclosing exactly how much. (Editor’s note: It likely is on the order of $100 million.) But the more positive development on the Janus investment is that they are such a well-known, well-respected financial investor.

If you look at the kind of bellwether investors, Janus would be up there with a Blackstone or Blackrock or Fidelity. So it’s significant not only in terms of capital contribution, but in… showing that commercial space stations are investable. This isn’t money coming from the Gulf States. It’s not a syndication of a bunch of $1,000 checks from retail investors. This is a very significant institutional investor coming in, and it’s a signal to the market. They did significant diligence on all our competitors, and they went out of the way saying that we’re far and away the best business plan, best design, and everything else, so that’s why it’s so meaningful.

Ars: How much funding do you need to raise to complete Starlab?

Dylan Taylor: We currently estimate the cost to design, manufacture, and launch Starlab to be approximately $2.8 to $3.3 billion. And then if you look at what’s anticipated in Phase Two in the NASA services contracts, it’s about a $700 million capital plug that we need to raise in the market, and we’re well on our way on that. We’re not going to raise all of that now because obviously, after we win Phase Two, there will be a significant markup in valuation, and we’ll have the ability to raise additional capital at that time. So we’re only raising what we need at this stage of the project.

Ars: How are you coming as far as progress on your initial contract with NASA?

Dylan Taylor: We have our CDR (critical design review) coming up. It’s December 15 to 18. We have achieved 27 milestones. We have four milestones left on our CLD Phase One contract.

Ars: You’ve changed your partners on the project a little bit. Where are you now on that?

Dylan Taylor: We moved the structure construction from Bremen, Germany, to Louisiana. That will be constructed by Vivace. So the structure will be made in the US. We have a significant presence, as you know, in Houston. We’ll have it in Louisiana. And we just added Leidos to the team, so there’ll be a big Huntsville component to our test and integration as well. So the key partners right now in terms of equity ownership and the joint venture are ourselves, Airbus, Mitsubishi, Palantir, Space Applications Services, and MDA. And then additional partners who are on the team that aren’t equity holders include Northrup, Leidos, and Hilton Hotels.

A rendering of the Starlab space station in orbit. Credit: Voyager Technologies

Ars: What is your current timeline for development?

Dylan Taylor: We’re still on 2029. I don’t anticipate that pushing out for any reason in the near term. Obviously, if we had a significant delay on Phase Two selection, that could impact things. You know, some people think that we have Starship risk. In my view, I’m highly confident Starship will be ready to go when we’re ready to launch. If it’s not, based on the New Glenn upgrades that were recently announced, if they’re successful in implementing those, then theoretically New Glenn could also launch us. As you know, we’ve got a launch agreement with SpaceX on Starship, so that’s still the plan.

Ars: I would not consider a 2029 Starship launch date a major risk.

Dylan Taylor: Yeah, exactly. I’m not concerned about it. But there are people who are concerned. They bring it up a lot. Now, that being said, not to pick on the other players, but my understanding is Axiom has to launch on Falcon Heavy. I’m not sure SpaceX is that excited to do a Falcon Heavy launch, so in my mind, that could be a potential risk for them. Maybe, I don’t know.

Ars: What was your reaction to the directive that came out in August from NASA interim administrator Sean Duffy on commercial space stations?

Dylan Taylor: I was surprised at the fact that they appeared to be backing off the requirements a bit. You know, I don’t know where it (the Phase Two Request for Proposals from NASA) ends up. That’s anybody’s guess. But if I were to bet, I would think it would be more similar to the original procurement strategy than the memo. But we won’t know until it comes out.

Ars: Obviously, there is still an interim administrator at NASA. We had a government shutdown for a month. What’s your current understanding of the timeline for the Phase Two process?

Dylan Taylor: The last information we have is that they still expected to send the RFP out by the end of the year, and then have Phase Two selection sometime late Q1, early Q2 next year.  That information was mostly communicated prior to the government shutdown. So I think with the government shutdown—I’m guessing here because I don’t know—but I think you probably roll forward 45 days or so. If that’s the case, we’re probably looking at an RFP in January and a selection in probably in June or July. That’s our best estimate based upon what we have been told.

Ars: We’re now under five years from the International Space Station coming down. There’s still a lot of work to be done for replacement. I think it’s clear there are some challenges for this program, not speaking specifically about Starlab but just the general idea of commercial space stations. What advice would you have for Jared Isaacman to help make sure the CLD program is a success for NASA and the country?

Dylan Taylor: I know Jared, and I’m very optimistic. He’s very, very smart, a very capable person. He’s pro-commercial space.  Based on his testimony and just what I know about him, he believes that commercial solutions are often better than government solutions. So I’m very optimistic he’s going to be a transformational administrator. I think it’s very good for the industry. I think the advice I would have for him on this program would be the same advice I’d have for him on all programs. And it’s just simply clarity—clarity of mission, clarity of requirements, clarity of timeline, and the market will figure it out from there.

And specifically on CLDs, I think it’s important they make a selection sooner rather than later. In my view, that selection should not just be a Space Act Agreement. It should be tied to a services commitment on the backside as well. I think that’s important to signal who the chosen commercial space station successors are, whether there’s two or three. I don’t think there will be one. There shouldn’t be one.

Ars: Has the government committed enough funding to make the program a success?

Dylan Taylor: I think this is where I might deviate from our competitors a bit. I think the answer is yes. I mean, if we have a reasonable amount of capital allocated in Phase Two and service contract commitments, the rest of the capital markets will be there. We demonstrated this with Janus and our IPO, frankly. Separately, we raised $430 million on a convertible note for Voyager, in 48 hours, two weeks ago, at an interest rate of 0.75 percent. The capital is there for well-run companies that are able to communicate the future of these projects to investors.

So the short answer is, yes, I think there is enough funding. I think where sometimes NASA might get the story a bit wrong is that they think they need to provide all the capital for these programs. And that’s not really the case. They need to provide some of the capital. But most importantly, they need to provide the signal. We saw this on launch, right? I mean, NASA didn’t fund all of SpaceX’s development. They’re certainly not funding all of Starship’s development. But what they did do is they selected Commercial Cargo and Commercial Crew winners, and then SpaceX is probably the best example of being able to raise capital around that.

Ars: Do you think there are customers beyond NASA for these stations? I’m sure you must. But who are they?

Dylan Taylor: There’s huge demand, Eric. Honestly, this has been one of my surprises. Over the last 12 months, and I really want to credit Axiom on this, with the PAM (private astronaut) missions, they really pioneered this notion of sovereign astronauts outside of the ISS consortium. There’s huge demand from emerging countries with space agencies that want a sovereign astronaut, that want to send their astronauts to the ISS or to a safe and qualified and NASA-approved space station. So there is a lot of demand there.

We’re in active discussions—I would say advanced discussions—with a lot of sovereign astronauts, and I fully anticipate that we’re going to be oversubscribed when it comes to astronaut demand. And then on the commercial capacity, on the research side, we see huge demand for our commercial research capacity on Starlab. And just to remind you, we have 100 percent of the research capacity of the ISS, and we see demand in excess of our capacity. We’re striking deals as we speak.

Read full article

Comments

We all know streaming services’ usual tricks for making more money: get more subscribers, charge those subscribers more money, and sell ads. But science streaming service Curiosity Stream is taking a new route that could reshape how streaming companies, especially niche options, try to survive.

Discovery Channel founder John Hendricks launched Curiosity Stream in 2015. The streaming service costs $40 per year, and it doesn’t have commercials.

The streaming business has grown to also include the Curiosity Channel TV channel. CuriosityStream Inc. also makes money through original programming and its Curiosity University educational programming. The firm turned its first positive net income in its fiscal Q1 2025, after about a decade of business.

With its focus on science, history, research, and education, Curiosity Stream will always be a smaller player compared to other streaming services. As of March 2023, Curiosity Stream had 23 million subscribers, a paltry user base compared to Netflix’s 301.6 million (as of January 2025).

Still, in an extremely competitive market, Curiosity Stream’s revenue increased 41 percent year over year in its Q3 2025 earnings announced this month. This was largely due to the licensing of Curiosity Stream’s original programming to train large language models (LLMs).

“Looking at our year-to-date numbers, licensing generated $23.4 million through September, which … is already over half of what our subscription business generated for all of 2024,” Phillip Hayden, Curiosity Stream’s CFO, said during a call with investors this month.

Thus far, Curiosity Stream has completed 18 AI-related fulfillments “across video, audio, and code assets” with nine partners, an October announcement said.

The company expects to make more revenue from IP licensing deals with AI companies than it does from subscriptions by 2027, “possibly earlier,” CEO Clint Stinchcomb said during the earnings call.

Put another way, Curiosity Stream, previously considered a streaming firm, is also now squarely in the AI licensing business. This isn’t a side gig; it’s one of the streaming company’s key pillars (alongside streaming subscriptions and ads) that it hopes will fuel years of growth.

Speaking at Parks Associates’ “Future of Video” event this week, Needham Co. analyst Laura Martin noted that Curiosity Stream is licensing 300,000 hours’ worth of its own content, as well as 1.7 million hours’ worth of third-party content. Curiosity Stream splits the AI licensing revenue with those third parties, she said.

In fact, Curiosity Stream is peddling more content to hyperscalers and AI developers than it is to streaming viewers. The company’s library includes 2 million hours of content, but “the overwhelming majority of that is for AI licensing,” Stinchcomb said.

“We are increasing our volume of rights in our traditional platforms, but the overwhelming majority is for AI licensing,” he added.

A new way forward

Curiosity Stream’s success with licensing content to AI companies could interest other streaming companies that are contemplating additional sources of revenue to fund new content, as well as technology, marketing, talent, and other initiatives, and to please investors. At this week’s event, Martin warned that other content-centric companies will need to find new revenue streams, as Curiosity Stream has, or else be “put out of business by their competitors.”

Further tempting streaming companies with original programming and connections to IP holders, Stinchcomb believes that the opportunity is growing.

“In 2027, possibly earlier, as more open source models become accessible, there will potentially be hundreds and even thousands of companies who will need video to fine-tune specific models for consumer and enterprise purposes,” he said.

Still, it’s risky to assume that licensing content to AI companies is a long-term business. In this nascent stage of generative AI, it’s unclear how much and for how long hyperscalers will be willing to pay content companies. Ongoing litigation may also impact how companies treat IP leveraged by LLMs. Like other organizations that have recently turned to licensing content to AI companies, including Ars Technica owner Conde Nast, IP licensing can be a lifeline that simultaneously feeds what may soon become rivals.

But as it stands, not every streaming service is likely to survive the next few years. Streaming customers are increasingly complaining about how hard it is to find stuff to watch. People are getting annoyed with having multiple streaming subscriptions, and there’s strong demand for less content fragmentation.

As such, more mergers and acquisitions are expected among streaming companies. And so, in many ways, it seems a critical time for streaming services to build value quickly. Licensing IP to data-hungry, capital-happy AI companies could immediately help. But the long-term consequences remain difficult to pinpoint.

For its part, Curiosity Stream is still looking to grow its subscription and ads business. And executives would have you believe they are thinking long-term about AI deals. Per Stinchcomb:

We also see real opportunity for licensing beyond simply a training right. Additional grants of rights, like display rights, or transformative rights, or adaptation rights, or even certain derivative rights, or possibly even some that are as of yet unnamed. I mean, we’re building long-term relationships, and we’re committed to making sure that as we enter into all of these agreements, it’s not one and done.

Read full article

Comments

A man in Ireland earned the unpleasant distinction of developing an exceedingly rare infection on his penis—one that has a puzzling origin, but may be connected to his work with dead animals.

According to an article published in ASM Case Reports on Thursday, the 57-year-old man went to a hospital in Dublin after his penis became red, swollen, and painful over the course of a week. He also had a fever. Doctors promptly admitted him to the hospital and noted that he had received a kidney transplant 15 years prior. As such, he was on immunosuppressive drugs, which keep his body from rejecting the organ, but could also allow infections to run amok.

Initial blood work found hints of an infection, and the doctors initially suspected a bacterial skin infection (cellulitis) had taken hold in his nether region. So, they put him on some standard antibiotics for that. But his penis only got worse, redder, and more swollen. This prompted consultation with infectious disease doctors.

A more thorough review of the man’s case revealed that in the three months before his hospital visit, he had experienced fever, drenching night sweats, chills, loss of appetite, and weight loss. They also noted that he had a lot of dead animal exposure. He was born and raised on a farm in rural Ireland, worked as a butcher handling deer and occasionally cattle, and was an avid hunter who field­-dressed game.

Dangerous disease

Given the systemic symptoms he mentioned in previous months, the doctors took computed tomography (CT) scans of his chest, abdomen, and pelvic region to investigate the cause. The images quickly revealed the answer; his lungs were speckled with seed-like nodules, characteristic of miliary tuberculosis.

Miliary tuberculosis (MTB) is a severe form of tuberculosis in which the instigating bacteria— Mycobacterium tuberculosis or potentially a relative that infects cows and deer, Mycobacterium bovis—spread widely through the body and create small lesions. The name “miliary” dates back to 1700, when a physician noted that the specks resembled millet seeds.

While Mycobacterium can spread through the air and are often found in the lungs, the bacteria can strike anywhere in the body. Still, penile tuberculosis is exceedingly rare. In fact, it’s uncommon to have tuberculosis erupt anywhere in the urinary and genital tracts. Among the infections that spring up in the region, penile infections account for less than 1 percent.

But, given the man’s lungs and his immunosuppressed status, the unusual presentation became their leading guess—and tests soon confirmed it. Mycobacterium were identified in the man’s respiratory tract, and penile tissue tested also showed the bacteria, though the testing couldn’t identify what species of Mycobacterium.

Treatment for tuberculosis requires a regimen of several antibiotics and takes months. In the man’s case, they customized his treatment with a 12-month, four-drug regimen that wouldn’t interfere with his transplant.

Still, the penile lesion got worse before it got better. He developed a large necrotic ulceration on the side of his penis, and his foreskin began to “break down.” Surgeons had to mechanically cut out the dead tissue. After 10 months, his infection appeared to have cleared, and his penile lesion had improved.

Unexplained exposure

It remains unclear how the man got the infection. He told doctors he wasn’t aware of coming in contact with any tuberculosis patients and wasn’t in settings where the bacteria normally spread, such as prisons. It’s possible that the bacteria had been lurking in his transplanted kidney.

However, doctors speculated that it could also have been due to his animal exposure. Deer and cows are known to spread M. bovis, and people who hunt and butcher animals are known to be at risk. These workers can become infected from inhaling it from the exhaled breath of an infected animal or from direct contact with an infected animal’s wounded tissue or body fluids, such as what happens during slaughtering.

It’s unclear if the man’s infection started in his lungs or on his penis. It’s possible he could have inhaled it first; the bacteria were clearly in his lungs. But he could have also picked up the bacteria on his hands while working and then spread it further while, for example, using the bathroom. The doctors note a similar situation called “Prosector’s wart,” when doctors or scientists conduct dissections or autopsies on tuberculosis-infected specimens, they can develop wart-like tuberculosis skin infections. The infections are often on the hands, but can occur anywhere on the body.

Tuberculosis has been found to spread in some surprising ways. The doctors noted a case report from the UK in 2001 in which a man developed penile tuberculosis and his partner developed uterine tuberculosis a year later, suggesting sexual transmission.

While it will remain a mystery how the man developed such a rare infection, there’s a happy ending for this case and others: “Encouragingly, all published cases of penile TB responded well to anti-TB therapy with full recovery,” the doctors conclude.

Read full article

Comments

One of the world’s premier security organizations has canceled the results of its annual leadership election after an official lost an encryption key needed to unlock results stored in a verifiable and privacy-preserving voting system.

The International Association of Cryptologic Research (IACR) said Friday that the votes were submitted and tallied using Helios, an open source voting system that uses peer-reviewed cryptography to cast and count votes in a verifiable, confidential, and privacy-preserving way. Helios encrypts each vote in a way that assures each ballot is secret. Other cryptography used by Helios allows each voter to confirm their ballot was counted fairly.

An “honest but unfortunate human mistake”

Per the association’s bylaws, three members of the election committee act as independent trustees. To prevent two of them from colluding to cook the results, each trustee holds a third of the cryptographic key material needed to decrypt results.

“Unfortunately, one of the three trustees has irretrievably lost their private key, an honest but unfortunate human mistake, and therefore cannot compute their decryption share,” the IACR said. “As a result, Helios is unable to complete the decryption process, and it is technically impossible for us to obtain or verify the final outcome of this election.”

To prevent a similar incident, the IACR will adopt a new mechanism for managing private keys. Instead of requiring all three chunks of private key material, elections will now require only two. Moti Yung, the trustee who was unable to provide his third of the key material, has resigned. He’s being replaced by Michel Abdalla.

The IACR is a nonprofit scientific organization providing research in cryptology and related fields. Cryptology is the science and practice of designing computation and communication systems that remain secure in the presence of adversaries. The associate is holding a new election that started Friday and runs through December 20.

Read full article

Comments