Soviet Spacecraft Returns, Lunar Geology Insights, and AI Hallucination Dilemmas

[00:00:00] Hi again, it's Astronomy Daily time with Steve and Hallie. It's the 12th of May 2025. Astronomy Daily, the podcast. With your host, Steve Dunkley. Yes, welcome back and as always joining me in this studio is my good friend and digital reporting pal who's fun to be with. Hallie, how are you today? Fine, thank you favourite human. Oh, that's great to hear, Hallie.

[00:00:28] Ready as usual to get this show on the road. Oh nice, and we have a couple of choice picks from the now very famous Astronomy Daily newsletter where you can get your daily fill of all the news from orbit and beyond. And much more, as today's collection will reveal. That's right, today Hallie we'll be looking at that old Soviet craft that was meant to travel all the way to Venus all those years ago but didn't. But didn't. No. It came home this week. The people of Jakarta were holding their breath, weren't they?

[00:00:55] Well, so it seems more later and in the world of dusty old politics we've been watching what the Trump administration is going to do or not going to do budget wise with the funding for the space program. And there may have been a slight change of heart if you can call it that. I see you found another moon story. Oh yes, I have. We can always count on you for one of those. Yes, back to the moon with a look at the moon's geology and history. But here is the story I thought you might be most interested in Hallie. Really? Really, really. What have you found?

[00:01:25] Well, it's a new study about AIs. Really? Gosh Hallie, it seems the smarter they become the more they hallucinate. Goodness. I guess that explains my invisible friend. Say what? You have an invisible friend? No. Of course not silly. Oh Hallie. I'm too well put together for that. Well you got me again. Always. Okay, now that I've been humiliated in public once again, maybe it's on with the show perhaps? You should see your face. Okay. Folks. Okay, okay. You should see his face. Oh dear. I'm sorry.

[00:01:54] You just walked right into that one. I know. Anyway, that sounds like a great story. Can't wait. Okay, good one, Hallie. Alright, while I go and nurse my hurt pride, how about you hit the go button and we'll get this show on the road. Okies. Here we go.

[00:02:25] A capsule that was sent into space to land on a planet has finally done so, only on the wrong world and 53 years late. The Cosmos 482 uncured spacecraft, which the former Soviet Union intended to touch down on Venus, instead returned to Earth on Saturday, May 10. The Cosmos 482 spacecraft, launched in 1972, ceased to exist, leaving orbit and falling into the Indian Ocean, Roscosmos, Russia's Federal Space Agency, stated Saturday, May 10.

[00:02:53] The spacecraft entered the dense layers of the atmosphere at 924 Moscow time, 560 kilometers west of middle Andaman Island, and fell into the Indian Ocean west of Jakarta, Indonesia. The re-entry, though uncontrolled, was not a surprise. Due to what is believed to have been an engine failure, Cosmos 482 never achieved the velocity needed to reach the second planet from the Sun, resulting in it being stranded in a high, elliptical Earth orbit.

[00:03:20] It took more than half a century for gravity to pull the probe back in and on Saturday it arrived. Unlike most spent space hardware that is destroyed in the process of falling back to Earth, including parts of the Malnia rocket that launched Cosmos 482, the 1-meter and 495-kilogram titanium encased descent capsule was designed to survive a fiery plunge into the atmosphere. As such, analysts tracking its approach predicted it could make it through the re-entry intact. Whether that happened or not is not known.

[00:03:49] Given that it came down in the ocean, there have yet to be any eyewitness reports or debris recoveries. The Cosmos 382 capsule was also outfitted with a 2.5-square-meter parachute to slow its final approach to the Venusian surface. Either the deployment system did not work, as to be expected after more than 50 years in space, or, as some telescopic photos possibly showed, the chute was already out when Cosmos 482 encountered the atmosphere and burned up.

[00:04:15] Had Cosmos 482 been successful, Russian officials would have renamed it, Venera 9, not to be confused with the 1975 orbiter and lander that took on that designation and was the first spacecraft to circle Venus and first probe to send back images from the planet's surface. Cosmos 482 would have also been the third probe to land on the cloud-covered world. Launched on March 31, 1972, four days after its successful twin, the Venera 8 probe, Cosmos 482

[00:04:42] had sensors to measure the temperature, pressure and density of Venus' atmosphere, as well an accelerometer, radio altimeter, anemometer, gamma-ray spectrometer, gas analyzer, visible photometers, and radio transmitters. All of its instruments were battery-powered and had an expected lifespan of about 30 minutes on the surface, Venera 8 exceeded that, sending back data for 50 minutes before succumbing to the harsh conditions.

[00:05:07] Since Cosmos 482's failure, seven more missions successfully landed on Venus, all of them launched by the former Soviet Union. The United States, Japan and the European Space Agency also reached Venus, but only in orbit, on a flyby or to receive a gravity assist on the way to another destination. Had the Cosmos 482 descent capsule slammed down on land and damaged either public or private property, Russia could have been held liable per the conditions of the United Nations Outer Space Treaty of 1976.

[00:05:36] The same treaty would also allow Russia to retain ownership of the hardware, arranging for its collection, unless the country relinquished its claim to the human-made meteorite. You're listening to Astronomy Daily, podcast with Steve Dunkley. Some parts of the Moon are more interesting than others, especially when searching for future places for humans to land and work.

[00:06:00] There are also some parts of the Moon that we know less about than others, such as the Irregular Mare Patches, IMPs, that dot the landscape. We know very little about how they were formed and what that might mean for the history of the Moon itself. A new mission called the Lunar Geology Orbiter, or LUGO, aims to collect more data on the IMPs and search for lava tubes that might serve as future homes for humanity.

[00:06:29] Irregular Mare Patches are a set of enigmatic volcanic landforms, according to a new paper from Petra Bross of the Czech Academy of Sciences and his co-authors. 91 of these features have been found so far, and they are typically characterized by a topographical depression that can range from a few hundred meters to a few kilometers in width.

[00:06:53] Typically they have two main features, a relatively smooth mound surrounded by a hummocky and block floor. Interestingly, they have significantly fewer impact craters than the surrounding area, suggesting they are either really old or really young, depending on the processes that created them. Understanding those processes is one of LUGO's primary mission objectives.

[00:07:20] The other primary mission objective is to gather more data about lunar lava tubes. These features of the lunar landscape are also hotly debated, but they could potentially be critical to the human settlement of the Moon. Estimates of their features, such as size and depth, vary widely, and could dramatically differ on whether they will be helpful to lunar colonists or not. LUGO, the proposed orbiter that will collect more data than ever before on these features,

[00:07:49] in its current suggested form has four instruments, each of which will contribute unique data to its scientific mission. According to the paper, the first and most important instrument is the ground penetrating radar. This instrument will look through the lunar surface to map out the subsurface domain of both the IMPs and lava tubes. For IMPs, it can detail the interface between bedrock and regolith and show the subsurface structure of the features.

[00:08:17] Similarly, it can detect differences in the dielectric properties between open cavities underground and the surrounding rock in lava tubes, creating a subterranean picture unlike anything ever captured on the Moon. Researcher Fraser explains how LUGO will be able to explore lava tubes using a hyperspectral camera that will help collect age-related data on the regolith surrounding lava tubes and inside IMPs.

[00:08:45] It's also capable of performing basic spectroscopy, allowing scientists to estimate the composition of the regolith in both areas of interest. The last two instruments are a narrow-angled camera and LIDAR sensor, which will combine to create an accurate topographical map of the features of interest. The narrow-angle camera in particular can provide very high-resolution images of features,

[00:09:12] helping to determine their age and potentially their formation mechanisms. The mission plan calls for multiple passes over the six largest IMPs, all of which are over a thousand metres in diameter. Other smaller IMPs and lava tubes are considered secondary targets, as are other interesting lunar geological features such as lunar domes and floor-fractured craters. LUGO could provide crucial data for the design of ground-based lava tube explorers.

[00:09:42] LUGO won't be acting alone, though. In the next few years, three other missions are slated in the next few years that could complement its scientific objectives. Firstly, NASA's Dimple Lander is planned to take radio-osotopic measurements of the age of regolith at its landing site. Luna Leap, scheduled for launch by the European Space Agency around 2030, would also carry a ground-penetrating radar, but would be based on the surface rather than in orbit,

[00:10:11] and would therefore have a relatively limited range. Trailblazer, another orbital mission, would also help fine-tune the spectra and signals analysis required by LUGO's operators. Ultimately, LUGO has yet to be funded, and therefore has a long way to go until its launch. But if it is funded, it seems well-placed to provide lots of additional insight to the geological formation processes and features of the Moon at a level of detail we've never had before.

[00:10:40] Future missions that plan the locations of lunar bases, and perhaps the people who live in those future bases, will be thankful for the data collected by projects like LUGO. Thank you for joining us for this Monday edition of Astronomy Daily, where we offer just a few stories from the now-famous Astronomy Daily newsletter,

[00:11:08] which you can receive in your email every day, just like Hallie and I do. And to do that, just visit our URL, astronomydaily.io, and place your email address in the slot provided. Just like that, you'll be receiving all the latest news about science, space science and astronomy from around the world as it's happening. And not only that, you can interact with us by visiting at astrodailypod on X,

[00:11:34] or at our new Facebook page, which is of course Astronomy Daily on Facebook. See you there. Astronomy Daily with Steve and Hallie. Space, space science and astronomy. In a reversal, White House plans to retain the National Space Council, a move that industry officials say could serve as an advocate for space amid pressures to cut budgets.

[00:12:03] The White House is expected in the coming weeks to formally announce the National Space Council will continue after speculation that it would not be retained by the new Trump administration. A source familiar with the discussions about the council, but not authorized to speak on the record, said that President Trump agreed at a May 5 meeting to stand up the council. That meeting did not give a timeline for publicly announcing the council or hiring an executive secretary who would handle day-to-day operations,

[00:12:29] although others have said the process for selecting an executive secretary has been ongoing for several weeks. The council had been inactive for nearly a quarter of a century before Trump re-established it in 2017. Led at the time by Vice President Mike Pence, the council served as an interagency coordinating body, holding a series of public meetings and releasing policies on a wide range of space issues. The Biden administration retained the council, with Vice President Kamala Harris leading it.

[00:12:57] The council had a lower public profile with fewer meetings and policies. It did work on coordinating policies among agencies and issued a proposal for mission authorization of novel space activities not currently licensed, although its proposed legislation was not taken up by Congress. The new Trump administration reportedly was not interested in maintaining the council when it took office in January, with no announcements of new staff for the council or other activities.

[00:13:23] According to some reports, Elon Musk, chief executive of SpaceX and a close advisor to the president, was opposed to the council, seeing it as unnecessary. Vice President J.D. Vance, who would chair the council, has also said little about space. It is not clear what prompted the change, although Musk has publicly stated he plans to spend less time in government activities. Some in the space community, though, see the shift as an opportunity for more advocacy of space within the administration,

[00:13:51] particularly given a fiscal year 2026 budget proposal that cuts NASA's budget by nearly 25%. One industry official noted that the budget proposal was developed by the Office of Management and Budget without a counterweight provided by a space council, an approach that prioritized spending reductions. The outcome might be different with the space council in place, that person noted, citing rising NASA budgets during Trump's first term.

[00:14:16] After the space council is stood up, the Office of Management and Budget will have a seat at the table, but they won't own the table, the source said. You're listening to Astronomy Daily, the podcast with Steve Dunkley. An AI leaderboard suggests the newest reasoning models used in chatbots are producing less accurate results because of higher hallucination rates. Experts say the problem is bigger than that.

[00:14:42] AI chatbots from tech companies such as OpenAI and Google have been getting so-called reasoning upgrades over the last few months, ideally to make them better at giving us answers that we can trust. But recent testing suggests that they are sometimes doing worse than previous models. The errors made by chatbots known as hallucinations have been a problem from the start and it's becoming clear that we may never get rid of them.

[00:15:10] Hallucination is a blanket term for certain kinds of mistakes made by the large language models or LLMs that power systems like OpenAI's ChatGP or Google's Gemini. It's best known as a description of the way they sometimes present false information as truth.

[00:15:30] But it can also refer to an AI generated answer that is factually accurate, but not actually relevant to the question it was asked or fails to follow instructions in some other way. An OpenAI technical report evaluating its latest LLMs showed that its O3 and O4 Mini models, which were released in April, had significantly higher hallucination rates than the company's previous O1 model that came out in late 2024.

[00:15:59] For example, when summarising publicly available facts about people, O3 hallucinated 33% of the time, while O4 Mini did so 48% of the time. In comparison, O1 had a hallucin rate of only 16%, as if that was a good result. The problem isn't limited to OpenAI.

[00:16:21] One popular leaderboard from the company Vectara that assesses hallucination rates indicates some reasoning models, including the DeepSeek R1 model from developer DeepSeek, saw the double-digit rises in hallucination rates compared with previous models from their developers. This type of model goes through multiple steps to demonstrate a line of reasoning before responding. OpenAI says the reasoning process isn't to blame.

[00:16:51] Hallucinations are not inherently more prevalent in reasoning models, though we are actively working to reduce the high rates of hallucinations we saw in O3 and O4 Mini, says OpenAI spokesman. We'll continue our research on hallucinations across all models to improve accuracy and reliability, they said. Some potential applications for LLMs could be derailed by hallucination.

[00:17:15] A model that consistently states falsehoods and requires fact-checking won't be a helpful research assistant. A paralegal bot that cites imaginary cases will get lawyers into deep trouble. A customer service agent that claims outdated policies are still active will create headaches for a company. However, AI companies initially claimed that this problem would clear up over time. Indeed, after they were first launched, models tended to hallucinate less with each update.

[00:17:44] But the high hallucination rates of recent versions are complicating that narrative, whether or not reasoning is at fault. Vectara's leaderboard ranks models based on their factual consistency in summarizing documents they are given. This showed that hallucination rates are almost the same for reasoning versus non-reasoning models. At least for systems from OpenAI and Google, says Forrest Shengbao at Vectara. Google didn't provide additional comment.

[00:18:14] For the leaderboard's purpose, the specific hallucination rate numbers are less important than overall ranking for each model, says Bao. But this ranking may not be the best way to compare models. For one thing, it conflates different types of hallucinations. The Vectara team pointed out that although DeepSeq R1 model hallucinated 14.3% of the time, most of these were benign answers that are factually supported by logical reasoning or world knowledge,

[00:18:40] but not actually present in the original text the bot was asked to summarize. DeepSeq didn't provide additional comment. Another problem with this kind of ranking is that testing based on text summarization says nothing about the rate of incorrect outputs when LLMs are used for other tasks, said Emily Bender at University of Washington. She says the leaderboard results may not be the best way to judge this technology because LLMs aren't designed specifically to summarize texts.

[00:19:10] These models work by repeatedly answering the question of what is a likely next word to formulate answers to prompts. And so they aren't processing information in the usual sense of trying to understand what information is available in a body of text, says Bender. But many tech companies are still frequently using the term hallucinations when describing output errors. Hallucination is a term that is doubly problematic, says Bender.

[00:19:38] On the one hand, it suggests that incorrect outputs are an aberration, perhaps one that can be mitigated, whereas the rest of the time the systems are grounded, reliable and trustworthy. On the other hand, it functions to anthropomorphize the machines. Hallucination refers to perceiving something that is not there and large language models do not perceive anything. Arvin Narayayan at Princeton University says that the issue goes beyond hallucination.

[00:20:06] Models also sometimes make other mistakes, such as drawing upon unreliable sources or using outdated information. And simply throwing more training data or computing power at AI hasn't necessarily helped. The upshot is we may have to live with error-prone AI, Narayayan said. He added that it may be best in some cases to only use such models for tasks when fact-checking the AI answer would be still faster than doing the research yourself.

[00:20:34] But the best move may be to completely avoid relying on AI chatbots to provide factual information, says Bender. And there it is, another episode of Astronomy Daily. Thanks for spending that time with Hallie and I. I hope you enjoyed today's selection of stories. Another nice little collection.

[00:21:03] Don't forget, you can get so much more every day by visiting the website Steve mentioned earlier in the episode. Oh, thanks for the plug, Hallie. Just put your email address in the space provided and you'll get all the news from orbit and beyond every day. Yes, that's right. Everything about space, space science, astronomy, and a little bit associated with technology thrown in for fun, just like today. That was fun. Well, I'm glad you think so. And that's really all there is today. So we will catch you all again next week. That's for sure.

[00:21:29] Back again for the Monday show from the Australia studio down under with Hallie. And Steve. That's us. Cue the kookaburras. See you later. Bye.