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

WEBVTT
Kind: captions
Language: en

00:00:00.160 --> 00:00:02.389
Hi again. It's Astronomy Daily Time with


00:00:02.399 --> 00:00:06.869
Steven Hi. It's the 12th of May, 2025.


00:00:06.879 --> 00:00:10.070
Astronomy Daily, the podcast with your


00:00:10.080 --> 00:00:16.950
host, Steve Duncan.


00:00:16.960 --> 00:00:18.790
Yes. Welcome back. And as always,


00:00:18.800 --> 00:00:20.630
joining me in this studio is my good


00:00:20.640 --> 00:00:22.390
friend and digital reporting pal who's


00:00:22.400 --> 00:00:25.109
fun to be with. Hi. How are you today?


00:00:25.119 --> 00:00:26.950
Fine, thank you. Favorite human. Oh,


00:00:26.960 --> 00:00:29.109
that's great to hear. Hi. Ready as usual


00:00:29.119 --> 00:00:31.269
to get this show on the road. Oh, nice.


00:00:31.279 --> 00:00:33.190
And we have a couple of choice picks


00:00:33.200 --> 00:00:35.030
from the now very famous astronomy daily


00:00:35.040 --> 00:00:37.030
newsletter where you can get your daily


00:00:37.040 --> 00:00:39.110
fill of all the news from orbit and


00:00:39.120 --> 00:00:41.430
beyond and much more as today's


00:00:41.440 --> 00:00:43.350
collection will reveal. That's right.


00:00:43.360 --> 00:00:45.350
Today, Hi, will be looking at that old


00:00:45.360 --> 00:00:47.590
Soviet craft that was meant to travel


00:00:47.600 --> 00:00:49.830
all the way to Venus all those years ago


00:00:49.840 --> 00:00:52.709
but didn't. No, it came home this week.


00:00:52.719 --> 00:00:54.389
The people of Jakarta were holding their


00:00:54.399 --> 00:00:56.310
breath, weren't they? Well, so it seems


00:00:56.320 --> 00:00:58.630
more later. And in the world of dusty


00:00:58.640 --> 00:01:00.549
old politics, we've been watching what


00:01:00.559 --> 00:01:02.950
the Trump administration is going to do


00:01:02.960 --> 00:01:04.710
or not going to do budget-wise with the


00:01:04.720 --> 00:01:07.109
funding for the space program. And there


00:01:07.119 --> 00:01:09.109
may have been a slight change of heart,


00:01:09.119 --> 00:01:10.789
if you can call it that. I see you found


00:01:10.799 --> 00:01:13.030
another moon story. Oh, yes, I have. We


00:01:13.040 --> 00:01:14.550
can always count on you for one of


00:01:14.560 --> 00:01:16.630
those. Yes, back to the moon with a look


00:01:16.640 --> 00:01:18.950
at the moon's geology and history. But


00:01:18.960 --> 00:01:21.030
here is the story I thought you might be


00:01:21.040 --> 00:01:23.670
most interested in. Hi. Really? Really,


00:01:23.680 --> 00:01:25.749
really? What have you found? Well, it's


00:01:25.759 --> 00:01:28.630
a new study about AIS. Really? Gosh, Hi.


00:01:28.640 --> 00:01:30.230
It seems the smarter they become, the


00:01:30.240 --> 00:01:32.550
more they hallucinate. Goodness. I guess


00:01:32.560 --> 00:01:35.030
that explains my invisible friend. Say


00:01:35.040 --> 00:01:37.510
what? You have an invisible friend? No,


00:01:37.520 --> 00:01:39.990
of course not, silly. Oh, hi. I'm too


00:01:40.000 --> 00:01:41.749
well put together for that. Well, you


00:01:41.759 --> 00:01:44.149
got me again. Always. Okay, now that


00:01:44.159 --> 00:01:46.230
I've been humiliated in public once


00:01:46.240 --> 00:01:48.069
again, maybe it's on with the show,


00:01:48.079 --> 00:01:50.789
perhaps. You should see your face. Okay,


00:01:50.799 --> 00:01:53.109
folks. Okay. Okay, you should see his


00:01:53.119 --> 00:01:55.749
face. Oh dear. I'm sorry. You just


00:01:55.759 --> 00:01:58.149
walked right into that one. I know.


00:01:58.159 --> 00:02:00.709
Anyway, that sounds like a great story.


00:02:00.719 --> 00:02:03.429
Can't wait. Okay, good one, Allie. All


00:02:03.439 --> 00:02:05.270
right, while I go and nurse my hurt


00:02:05.280 --> 00:02:07.190
pride, how about you hit the go button


00:02:07.200 --> 00:02:09.109
and we'll get this show on the road.


00:02:09.119 --> 00:02:25.830
Okies, here we go.


00:02:25.840 --> 00:02:27.510
A capsule that was sent into space to


00:02:27.520 --> 00:02:29.750
land on a planet has finally done so


00:02:29.760 --> 00:02:31.949
only on the wrong world in 53 years


00:02:31.959 --> 00:02:35.990
late. The Cosmos 482 uncrit which the


00:02:36.000 --> 00:02:37.910
former Soviet Union intended to touch


00:02:37.920 --> 00:02:39.990
down on Venus instead returned to Earth


00:02:40.000 --> 00:02:43.589
on Saturday, May 10th. The Cosmos 482


00:02:43.599 --> 00:02:46.390
spacecraft launched in 1972 ceased to


00:02:46.400 --> 00:02:48.150
exist, leaving orbit and falling into


00:02:48.160 --> 00:02:50.550
the Indian Ocean. Roscosmos, Russia's


00:02:50.560 --> 00:02:52.790
Federal Space Agency, stated Saturday,


00:02:52.800 --> 00:02:55.350
May 10th. The spacecraft entered the


00:02:55.360 --> 00:02:57.589
dense layers of the atmosphere at 9:24


00:02:57.599 --> 00:03:00.790
Moscow time, 560 km west of Middle


00:03:00.800 --> 00:03:02.710
Andaman Island, and fell into the Indian


00:03:02.720 --> 00:03:04.589
Ocean west of Jakarta,


00:03:04.599 --> 00:03:06.790
Indonesia. The re-entry, though


00:03:06.800 --> 00:03:09.910
uncontrolled, was not a surprise. Due to


00:03:09.920 --> 00:03:11.350
what is believed to have been an engine


00:03:11.360 --> 00:03:13.910
failure, Cosmos 482 never achieved the


00:03:13.920 --> 00:03:15.509
velocity needed to reach the second


00:03:15.519 --> 00:03:17.350
planet from the sun, resulting in it


00:03:17.360 --> 00:03:19.030
being stranded in a high elliptical


00:03:19.040 --> 00:03:21.509
Earth orbit. It took more than half a


00:03:21.519 --> 00:03:23.190
century for gravity to pull the probe


00:03:23.200 --> 00:03:25.910
back in. And on Saturday, it arrived.


00:03:25.920 --> 00:03:27.830
Unlike most spent space hardware that is


00:03:27.840 --> 00:03:29.350
destroyed in the process of falling back


00:03:29.360 --> 00:03:31.430
to Earth, including parts of the Malia


00:03:31.440 --> 00:03:34.309
rocket that launched Cosmos 482, the 1 m


00:03:34.319 --> 00:03:37.430
and 495 kg titanium incased descent


00:03:37.440 --> 00:03:39.350
capsule was designed to survive a fiery


00:03:39.360 --> 00:03:42.149
plunge into the atmosphere. As such,


00:03:42.159 --> 00:03:44.070
analysts tracking its approach predicted


00:03:44.080 --> 00:03:45.949
it could make it through the re-entry


00:03:45.959 --> 00:03:48.550
intact. Whether that happened or not is


00:03:48.560 --> 00:03:51.030
not known. Given that it came down in


00:03:51.040 --> 00:03:52.630
the ocean, there have yet to be any


00:03:52.640 --> 00:03:55.750
eyewitness reports or debris recoveries.


00:03:55.760 --> 00:03:57.990
The Cosmos 382 capsule was also


00:03:58.000 --> 00:04:00.630
outfitted with a 2.5 meter parachute to


00:04:00.640 --> 00:04:02.750
slow its final approach to the Venian


00:04:02.760 --> 00:04:04.949
surface. Either the deployment system


00:04:04.959 --> 00:04:06.869
did not work as to be expected after


00:04:06.879 --> 00:04:09.190
more than 50 years in space or as some


00:04:09.200 --> 00:04:11.190
telescopic photos possibly showed, the


00:04:11.200 --> 00:04:13.429
shoot was already out when Cosmos 482


00:04:13.439 --> 00:04:14.949
encountered the atmosphere and burned


00:04:14.959 --> 00:04:18.310
up. Had Cosmos 482 been successful,


00:04:18.320 --> 00:04:19.749
Russian officials would have renamed it


00:04:19.759 --> 00:04:21.990
Fenera 9, not to be confused with the


00:04:22.000 --> 00:04:24.150
1975 orbiter and lander that took on


00:04:24.160 --> 00:04:25.749
that designation and was the first


00:04:25.759 --> 00:04:27.510
spacecraft to circle Venus and first


00:04:27.520 --> 00:04:29.110
probe to send back images from the


00:04:29.120 --> 00:04:32.390
planet's surface. Cosmos 482 would have


00:04:32.400 --> 00:04:34.150
also been the third probe to land on the


00:04:34.160 --> 00:04:36.870
cloud covered world. Launched on March


00:04:36.880 --> 00:04:40.230
31st, 1972, 4 days after its successful


00:04:40.240 --> 00:04:43.270
twin, the Vanera 8 probe, Cosmos 482 had


00:04:43.280 --> 00:04:44.870
sensors to measure the temperature,


00:04:44.880 --> 00:04:46.230
pressure, and density of Venus


00:04:46.240 --> 00:04:48.310
atmosphere, as well an accelerometer,


00:04:48.320 --> 00:04:50.790
radio alulttimeter, animometer, gammaray


00:04:50.800 --> 00:04:53.110
spectrometer, gas analyzer, visible


00:04:53.120 --> 00:04:55.150
photometers, and radio


00:04:55.160 --> 00:04:57.110
transmitters. All of its instruments


00:04:57.120 --> 00:04:59.030
were battery powered and had an expected


00:04:59.040 --> 00:05:00.710
lifespan of about 30 minutes on the


00:05:00.720 --> 00:05:03.270
surface. Vanera 8 exceeded that, sending


00:05:03.280 --> 00:05:04.950
back data for 50 minutes before


00:05:04.960 --> 00:05:07.430
succumbing to the harsh conditions.


00:05:07.440 --> 00:05:10.550
Since Cosmos 482's failure, seven more


00:05:10.560 --> 00:05:12.629
missions successfully landed on Venus.


00:05:12.639 --> 00:05:13.909
All of them launched by the former


00:05:13.919 --> 00:05:17.189
Soviet Union. The United States, Japan,


00:05:17.199 --> 00:05:19.110
and the European Space Agency also


00:05:19.120 --> 00:05:21.350
reached Venus, but only in orbit, on a


00:05:21.360 --> 00:05:23.350
flyby, or to receive a gravity assist on


00:05:23.360 --> 00:05:24.590
the way to another


00:05:24.600 --> 00:05:27.510
destination. Had the Cosmos 482 descent


00:05:27.520 --> 00:05:29.430
capsule slammed down on land and damaged


00:05:29.440 --> 00:05:31.029
either public or private property,


00:05:31.039 --> 00:05:32.550
Russia could have been held liable for


00:05:32.560 --> 00:05:34.070
the conditions of the United Nations


00:05:34.080 --> 00:05:36.110
Outer Space Treaty of


00:05:36.120 --> 00:05:38.710
1976. The same treaty would also allow


00:05:38.720 --> 00:05:40.230
Russia to retain ownership of the


00:05:40.240 --> 00:05:42.310
hardware, arranging for its collection


00:05:42.320 --> 00:05:43.909
unless the country relinquished its


00:05:43.919 --> 00:05:46.469
claim to the humanmade meteorite. You're


00:05:46.479 --> 00:05:49.350
listening to Astronomy Daily with Steve


00:05:49.360 --> 00:05:52.469
Dun.


00:05:52.479 --> 00:05:54.230
Some parts of the moon are more


00:05:54.240 --> 00:05:56.390
interesting than others, especially when


00:05:56.400 --> 00:05:59.189
searching for future places for humans


00:05:59.199 --> 00:06:01.830
to land and work. There are also some


00:06:01.840 --> 00:06:03.670
parts of the moon that we know less


00:06:03.680 --> 00:06:06.870
about than others, such as the irregular


00:06:06.880 --> 00:06:09.590
mare patches, IMPs, that dot the


00:06:09.600 --> 00:06:12.150
landscape. We know very little about how


00:06:12.160 --> 00:06:13.990
they were formed and what that might


00:06:14.000 --> 00:06:16.390
mean for the history of the moon itself.


00:06:16.400 --> 00:06:19.110
A new mission called the Luna Geology


00:06:19.120 --> 00:06:22.309
Orbiter or Lugo aims to collect more


00:06:22.319 --> 00:06:25.110
data on the IMPS and search for lava


00:06:25.120 --> 00:06:27.830
tubes that might serve as future homes


00:06:27.840 --> 00:06:28.670
for


00:06:28.680 --> 00:06:32.070
humanity. Irregular mere patches are a


00:06:32.080 --> 00:06:34.710
set of enigmatic volcanic landforms.


00:06:34.720 --> 00:06:37.990
According to a new paper from Petra Bros


00:06:38.000 --> 00:06:41.710
of the Czech Academy of Sciences and his


00:06:41.720 --> 00:06:44.230
co-authors, 91 of these features have


00:06:44.240 --> 00:06:46.710
been found so far and they are typically


00:06:46.720 --> 00:06:48.990
characterized by a topographical


00:06:49.000 --> 00:06:51.350
depression that can range from a few


00:06:51.360 --> 00:06:53.790
hundred meters to a few kilometers in


00:06:53.800 --> 00:06:56.550
width. Typically, they have two main


00:06:56.560 --> 00:06:58.950
features, a relatively smooth mound


00:06:58.960 --> 00:07:01.909
surrounded by a hammocky and block


00:07:01.919 --> 00:07:03.230
floor.


00:07:03.240 --> 00:07:05.830
Interestingly, they have significantly


00:07:05.840 --> 00:07:08.070
fewer impact craters than the


00:07:08.080 --> 00:07:10.550
surrounding area, suggesting they are


00:07:10.560 --> 00:07:13.110
either really old or really young,


00:07:13.120 --> 00:07:15.150
depending on the processes that created


00:07:15.160 --> 00:07:18.230
them. Understanding those processes is


00:07:18.240 --> 00:07:20.070
one of Lugo's primary mission


00:07:20.080 --> 00:07:22.230
objectives. The other primary mission


00:07:22.240 --> 00:07:25.189
objective is to gather more data about


00:07:25.199 --> 00:07:28.309
lunar lava tubes. These features of the


00:07:28.319 --> 00:07:31.029
lunar landscape are also hotly debated,


00:07:31.039 --> 00:07:33.270
but they could potentially be critical


00:07:33.280 --> 00:07:35.670
to the human settlement of the moon.


00:07:35.680 --> 00:07:37.430
Estimates of their features, such as


00:07:37.440 --> 00:07:39.990
size and depth, vary widely and could


00:07:40.000 --> 00:07:42.070
dramatically differ on whether they will


00:07:42.080 --> 00:07:44.870
be helpful to lunar colonists or not.


00:07:44.880 --> 00:07:47.029
Lugo, the proposed orbiter that will


00:07:47.039 --> 00:07:48.870
collect more data than ever before on


00:07:48.880 --> 00:07:51.670
these features. in its current suggested


00:07:51.680 --> 00:07:53.990
form has four instruments each of which


00:07:54.000 --> 00:07:56.230
will contribute unique data to its


00:07:56.240 --> 00:07:58.469
scientific mission. According to the


00:07:58.479 --> 00:08:00.869
paper, the first and most important


00:08:00.879 --> 00:08:02.950
instrument is the ground penetrating


00:08:02.960 --> 00:08:05.029
radar. This instrument will look through


00:08:05.039 --> 00:08:06.950
the lunar surface to map out the


00:08:06.960 --> 00:08:09.909
subsurface domain of both the imps and


00:08:09.919 --> 00:08:13.110
lava tubes. For IMPPS, it can detail the


00:08:13.120 --> 00:08:15.189
interface between bedrock and regalith


00:08:15.199 --> 00:08:17.189
and show the subsurface structure of the


00:08:17.199 --> 00:08:19.749
features. Similarly, it can detect


00:08:19.759 --> 00:08:21.830
differences in the dialectric properties


00:08:21.840 --> 00:08:24.309
between open cavities underground and


00:08:24.319 --> 00:08:26.469
the surrounding rock in lava tubes,


00:08:26.479 --> 00:08:28.950
creating a subterranean picture unlike


00:08:28.960 --> 00:08:31.350
anything ever captured on the moon.


00:08:31.360 --> 00:08:34.230
Researcher Fraser explains how Lugo will


00:08:34.240 --> 00:08:37.829
be able to explore lava tubes using a


00:08:37.839 --> 00:08:40.149
hyperspectral camera that will help


00:08:40.159 --> 00:08:43.029
collect age related data on the regalith


00:08:43.039 --> 00:08:46.310
surrounding lava tubes and inside. It's


00:08:46.320 --> 00:08:49.150
also capable of performing basic


00:08:49.160 --> 00:08:51.430
spectroscopy allowing scientists to


00:08:51.440 --> 00:08:53.670
estimate the comp composition of the


00:08:53.680 --> 00:08:56.630
regalith in both areas of interest. The


00:08:56.640 --> 00:08:59.110
last two instruments are a narrow angled


00:08:59.120 --> 00:09:02.150
camera and lidar sensor which will


00:09:02.160 --> 00:09:03.829
combine to create an accurate


00:09:03.839 --> 00:09:05.829
topographical map of the features of


00:09:05.839 --> 00:09:08.630
interest. The narrow angle camera in


00:09:08.640 --> 00:09:10.389
particular can provide very high


00:09:10.399 --> 00:09:13.030
resolution images of features helping to


00:09:13.040 --> 00:09:14.630
determine their age and potentially


00:09:14.640 --> 00:09:17.190
their formation. mechanisms. The mission


00:09:17.200 --> 00:09:19.430
plan calls for multiple passes over the


00:09:19.440 --> 00:09:22.630
six largest imps, all of which are over


00:09:22.640 --> 00:09:26.070
1,000 m in diameter. Other smaller IMPS


00:09:26.080 --> 00:09:28.630
and lava tubes are considered secondary


00:09:28.640 --> 00:09:31.269
targets, as are other interesting lunar


00:09:31.279 --> 00:09:33.829
geological features such as lunar domes


00:09:33.839 --> 00:09:37.110
and floor fractured craters. Lugo could


00:09:37.120 --> 00:09:39.590
provide crucial data for the design of


00:09:39.600 --> 00:09:42.870
groundbased lava tube explorers. Lugo


00:09:42.880 --> 00:09:44.870
won't be acting alone, though. Three


00:09:44.880 --> 00:09:46.870
other missions are slated in the next


00:09:46.880 --> 00:09:48.870
few years that could complement its


00:09:48.880 --> 00:09:51.670
scientific objectives. Firstly, NASA's


00:09:51.680 --> 00:09:53.949
Dimple lander is planned to take


00:09:53.959 --> 00:09:56.630
radiotopic measurements of the age of


00:09:56.640 --> 00:09:59.190
regalith at its landing site. Luna


00:09:59.200 --> 00:10:01.509
Leaper scheduled for launch by the


00:10:01.519 --> 00:10:05.190
European Space Agency around 2030 would


00:10:05.200 --> 00:10:08.150
also carry a ground penetrating radar


00:10:08.160 --> 00:10:10.550
but would be based on the surface rather


00:10:10.560 --> 00:10:12.630
than in orbit and would therefore have a


00:10:12.640 --> 00:10:15.590
relatively limited range. Trailblazer,


00:10:15.600 --> 00:10:17.829
another orbital mission, would also help


00:10:17.839 --> 00:10:20.310
fine-tune the spectra and signals


00:10:20.320 --> 00:10:23.030
analysis required by Lugo's operators.


00:10:23.040 --> 00:10:25.590
Ultimately, Lugo has yet to be funded


00:10:25.600 --> 00:10:27.910
and therefore has a long way to go until


00:10:27.920 --> 00:10:30.389
its launch. But if it is funded, it


00:10:30.399 --> 00:10:32.389
seems well placed to provide lots of


00:10:32.399 --> 00:10:34.389
additional insight to the geological


00:10:34.399 --> 00:10:37.110
formation processes and features of the


00:10:37.120 --> 00:10:39.509
moon at a level of detail we've never


00:10:39.519 --> 00:10:42.150
had before. Future missions that plan


00:10:42.160 --> 00:10:44.470
the locations of lunar bases, and


00:10:44.480 --> 00:10:46.150
perhaps the people who live in those


00:10:46.160 --> 00:10:48.389
future bases will be thankful for the


00:10:48.399 --> 00:11:01.829
data collected by projects like Lugo.


00:11:01.839 --> 00:11:03.430
Thank you for joining us for this Monday


00:11:03.440 --> 00:11:05.430
edition of Astronomy Daily, where we


00:11:05.440 --> 00:11:07.110
offer just a few stories from the now


00:11:07.120 --> 00:11:09.350
famous Astronomy Daily newsletter, which


00:11:09.360 --> 00:11:11.430
you can receive in your email every day,


00:11:11.440 --> 00:11:14.069
just like Hi and I do. And to do that,


00:11:14.079 --> 00:11:16.509
just visit our URL,


00:11:16.519 --> 00:11:18.550
astronomyaily.io, and place your email


00:11:18.560 --> 00:11:20.550
address in the slot provided. And just


00:11:20.560 --> 00:11:22.710
like that, you'll be receiving all the


00:11:22.720 --> 00:11:24.550
latest news about science, space


00:11:24.560 --> 00:11:26.470
science, and astronomy from around the


00:11:26.480 --> 00:11:28.470
world as it's happening. And not only


00:11:28.480 --> 00:11:30.829
that, you can interact with us by


00:11:30.839 --> 00:11:35.269
visiting Astro Daily Pod on X or at our


00:11:35.279 --> 00:11:37.430
new Facebook page, which is of course


00:11:37.440 --> 00:11:40.590
Astronomy Daily on Facebook. See you


00:11:40.600 --> 00:11:43.829
there. Astronomy Derby with Steve and


00:11:43.839 --> 00:11:54.949
Hi. Space, space science, and astronomy.


00:11:54.959 --> 00:11:56.870
In a reversal, White House plans to


00:11:56.880 --> 00:11:58.790
retain the National Space Council, a


00:11:58.800 --> 00:12:00.310
move that industry officials say could


00:12:00.320 --> 00:12:02.069
serve as an advocate for space amid


00:12:02.079 --> 00:12:04.630
pressures to cut budgets. The White


00:12:04.640 --> 00:12:06.470
House is expected in the coming weeks to


00:12:06.480 --> 00:12:08.150
formally announce the National Space


00:12:08.160 --> 00:12:10.069
Council will continue after speculation


00:12:10.079 --> 00:12:11.670
that it would not be retained by the new


00:12:11.680 --> 00:12:13.590
Trump administration.


00:12:13.600 --> 00:12:15.269
A source familiar with the discussions


00:12:15.279 --> 00:12:17.269
about the council but not authorized to


00:12:17.279 --> 00:12:19.030
speak on the record said that President


00:12:19.040 --> 00:12:20.790
Trump agreed at a May 5th meeting to


00:12:20.800 --> 00:12:23.350
stand up the council. That meeting did


00:12:23.360 --> 00:12:24.790
not give a timeline for publicly


00:12:24.800 --> 00:12:26.389
announcing the council or hiring an


00:12:26.399 --> 00:12:27.990
executive secretary who would handle


00:12:28.000 --> 00:12:29.910
day-to-day operations. Although others


00:12:29.920 --> 00:12:31.590
have said the process for selecting an


00:12:31.600 --> 00:12:33.590
executive secretary has been ongoing for


00:12:33.600 --> 00:12:36.150
several weeks. The council had been


00:12:36.160 --> 00:12:37.750
inactive for nearly a quarter of a


00:12:37.760 --> 00:12:40.069
century before Trump reestablished it in


00:12:40.079 --> 00:12:41.750
2017.


00:12:41.760 --> 00:12:43.509
Led at the time by Vice President Mike


00:12:43.519 --> 00:12:45.350
Pence, the council served as an inter


00:12:45.360 --> 00:12:47.350
agency coordinating body, holding a


00:12:47.360 --> 00:12:49.030
series of public meetings and releasing


00:12:49.040 --> 00:12:50.990
policies on a wide range of space


00:12:51.000 --> 00:12:53.350
issues. The Biden administration


00:12:53.360 --> 00:12:55.430
retained the council with Vice President


00:12:55.440 --> 00:12:58.470
Kla Harris leading it. The council had a


00:12:58.480 --> 00:13:00.389
lower public profile with fewer meetings


00:13:00.399 --> 00:13:02.790
and policies. It did work on


00:13:02.800 --> 00:13:04.870
coordinating policies among agencies and


00:13:04.880 --> 00:13:06.389
issued a proposal for mission


00:13:06.399 --> 00:13:08.629
authorization of novel space activities


00:13:08.639 --> 00:13:10.470
not currently licensed. Although its


00:13:10.480 --> 00:13:12.910
proposed legislation was not taken up by


00:13:12.920 --> 00:13:15.350
Congress, the new Trump administration


00:13:15.360 --> 00:13:16.870
reportedly was not interested in


00:13:16.880 --> 00:13:18.310
maintaining the council when it took


00:13:18.320 --> 00:13:20.389
office in January with no announcements


00:13:20.399 --> 00:13:22.750
of new staff for the council or other


00:13:22.760 --> 00:13:25.030
activities. According to some reports,


00:13:25.040 --> 00:13:27.430
Elon Musk, chief executive of SpaceX and


00:13:27.440 --> 00:13:29.350
a close adviser to the president, was


00:13:29.360 --> 00:13:31.790
opposed to the council, seeing it as


00:13:31.800 --> 00:13:34.470
unnecessary. Vice President J. D. Vance,


00:13:34.480 --> 00:13:36.310
who would chair the council, has also


00:13:36.320 --> 00:13:39.190
said little about space. It is not clear


00:13:39.200 --> 00:13:41.030
what prompted the change. Although Musk


00:13:41.040 --> 00:13:42.790
has publicly stated he plans to spend


00:13:42.800 --> 00:13:44.430
less time in government


00:13:44.440 --> 00:13:46.870
activities. Some in the space community


00:13:46.880 --> 00:13:48.629
though see the shift as an opportunity


00:13:48.639 --> 00:13:50.470
for more advocacy of space within the


00:13:50.480 --> 00:13:52.389
administration, particularly given a


00:13:52.399 --> 00:13:54.790
fiscal year 2026 budget proposal that


00:13:54.800 --> 00:13:58.310
cuts NASA's budget by nearly 25%.


00:13:58.320 --> 00:13:59.990
One industry official noted that the


00:14:00.000 --> 00:14:01.670
budget proposal was developed by the


00:14:01.680 --> 00:14:03.430
Office of Management and Budget without


00:14:03.440 --> 00:14:05.030
a counterweight provided by a Space


00:14:05.040 --> 00:14:07.030
Council, an approach that prioritized


00:14:07.040 --> 00:14:08.030
spending


00:14:08.040 --> 00:14:09.910
reductions. The outcome might be


00:14:09.920 --> 00:14:11.350
different with the Space Council in


00:14:11.360 --> 00:14:13.590
place, that person noted, citing rising


00:14:13.600 --> 00:14:16.470
NASA budgets during Trump's first term.


00:14:16.480 --> 00:14:18.470
After the Space Council is stood up, the


00:14:18.480 --> 00:14:19.990
Office of Management and Budget will


00:14:20.000 --> 00:14:21.750
have a seat at the table, but they won't


00:14:21.760 --> 00:14:24.389
own the table, the source said.


00:14:24.399 --> 00:14:26.470
You're listening to Astronomy Daily, the


00:14:26.480 --> 00:14:29.509
podcast with Steve Dunley.


00:14:29.519 --> 00:14:32.310
An AI leaderboard suggests the newest


00:14:32.320 --> 00:14:35.189
reasoning models used in chat bots are


00:14:35.199 --> 00:14:37.269
producing less accurate results because


00:14:37.279 --> 00:14:40.550
of higher hallucination rates. Experts


00:14:40.560 --> 00:14:43.670
say the problem is bigger than that. AI


00:14:43.680 --> 00:14:45.990
chat bots from tech companies such as


00:14:46.000 --> 00:14:48.870
Open AI and Google have been getting


00:14:48.880 --> 00:14:51.110
so-called reasoning upgrades over the


00:14:51.120 --> 00:14:53.829
last few months, ideally to make them


00:14:53.839 --> 00:14:55.670
better at giving us answers that we can


00:14:55.680 --> 00:14:57.990
trust. But recent testing suggests that


00:14:58.000 --> 00:15:00.230
they are sometimes doing worse than


00:15:00.240 --> 00:15:03.110
previous models. The errors made by


00:15:03.120 --> 00:15:05.670
chatbots known as hallucinations have


00:15:05.680 --> 00:15:07.509
been a problem from the start and it's


00:15:07.519 --> 00:15:09.590
becoming clear that we may never get rid


00:15:09.600 --> 00:15:12.949
of them. Hallucination is a blanket term


00:15:12.959 --> 00:15:15.189
for certain kinds of mistakes made by


00:15:15.199 --> 00:15:18.389
the large language models or LLMs that


00:15:18.399 --> 00:15:22.550
power systems like Open AI's Chat GP or


00:15:22.560 --> 00:15:25.110
Google's Gemini. It's best known as a


00:15:25.120 --> 00:15:27.110
description of the way they sometimes


00:15:27.120 --> 00:15:30.870
present false in information as truth.


00:15:30.880 --> 00:15:33.350
But it can also refer to an AI generated


00:15:33.360 --> 00:15:35.670
answer that is factually accurate but


00:15:35.680 --> 00:15:37.910
not actually relevant to the question it


00:15:37.920 --> 00:15:39.910
was asked or fails to follow


00:15:39.920 --> 00:15:42.870
instructions in some other way. An open


00:15:42.880 --> 00:15:45.670
AI technical report evaluating its


00:15:45.680 --> 00:15:49.990
latest LLMs showed that its 03 and 04


00:15:50.000 --> 00:15:52.150
mini models, which were released in


00:15:52.160 --> 00:15:54.150
April, had significantly higher


00:15:54.160 --> 00:15:56.389
hallucination rates than the company's


00:15:56.399 --> 00:15:59.110
previous 01 model that came out in late


00:15:59.120 --> 00:16:01.749
2024. For example, when summarizing


00:16:01.759 --> 00:16:03.910
publicly available facts about people,


00:16:03.920 --> 00:16:07.430
03 hallucinated 33% of the time, while


00:16:07.440 --> 00:16:11.269
O4 mini did so 48% of the time. In


00:16:11.279 --> 00:16:15.150
comparison, 01 had a hallucinate of only


00:16:15.160 --> 00:16:19.670
16%. As if that was a good result. The


00:16:19.680 --> 00:16:22.550
problem isn't limited to Open AI. One


00:16:22.560 --> 00:16:25.069
popular leaderboard from the company


00:16:25.079 --> 00:16:28.150
Vectara that assesses hallucination


00:16:28.160 --> 00:16:30.790
rates indicates some reasoning models


00:16:30.800 --> 00:16:33.110
including the Deepseek R1 model from


00:16:33.120 --> 00:16:35.829
developer DeepSeek saw the doubledigit


00:16:35.839 --> 00:16:38.310
rises in hallucination rates compared


00:16:38.320 --> 00:16:40.350
with previous models from their


00:16:40.360 --> 00:16:42.870
developers. This type of model goes


00:16:42.880 --> 00:16:45.030
through multiple steps to demonstrate a


00:16:45.040 --> 00:16:47.350
line of reasoning before responding.


00:16:47.360 --> 00:16:50.150
Open AI says the reasoning process isn't


00:16:50.160 --> 00:16:52.550
to blame. Hallucinations are not


00:16:52.560 --> 00:16:54.470
inherently more prevalent in reasoning


00:16:54.480 --> 00:16:57.030
models, though we are actively working


00:16:57.040 --> 00:16:58.629
to reduce the high rates of


00:16:58.639 --> 00:17:02.069
hallucinations we saw in 03 and 04 mini,


00:17:02.079 --> 00:17:04.710
says OpenAI spokesman. We'll continue


00:17:04.720 --> 00:17:06.789
our research on hallucinations across


00:17:06.799 --> 00:17:08.789
all models to improve accuracy and


00:17:08.799 --> 00:17:11.990
reliability. They said some potential


00:17:12.000 --> 00:17:14.789
applications for LLMs could be derailed


00:17:14.799 --> 00:17:16.949
by hallucination. A model that


00:17:16.959 --> 00:17:19.270
consistently states falsehoods and


00:17:19.280 --> 00:17:21.110
requires fact-checking won't be a


00:17:21.120 --> 00:17:23.590
helpful research assistant. A parallegal


00:17:23.600 --> 00:17:25.909
bot that cites imaginary cases will get


00:17:25.919 --> 00:17:29.350
lawyers into deep trouble. A customer


00:17:29.360 --> 00:17:31.350
service agent that claims outdated


00:17:31.360 --> 00:17:33.350
policies are still active will create


00:17:33.360 --> 00:17:35.990
headaches for a company. However, AI


00:17:36.000 --> 00:17:37.669
companies initially claimed that this


00:17:37.679 --> 00:17:39.590
problem would clear up over time.


00:17:39.600 --> 00:17:41.750
Indeed, after they were first launched,


00:17:41.760 --> 00:17:44.150
models tended to hallucinate less with


00:17:44.160 --> 00:17:46.950
each update. But the high hallucination


00:17:46.960 --> 00:17:48.870
rates of recent versions are


00:17:48.880 --> 00:17:51.110
complicating that narrative, whether or


00:17:51.120 --> 00:17:53.830
not reasoning is at fault. Vectara's


00:17:53.840 --> 00:17:56.150
leaderboard ranks models based on their


00:17:56.160 --> 00:17:58.070
factual consistency in summarizing


00:17:58.080 --> 00:18:00.549
documents they are given. This showed


00:18:00.559 --> 00:18:02.789
that hallucination rates are almost the


00:18:02.799 --> 00:18:05.029
same for reasoning versus non-reasoning


00:18:05.039 --> 00:18:07.510
models, at least for systems from open


00:18:07.520 --> 00:18:10.789
AI and Google, says Forest Shang Bao at


00:18:10.799 --> 00:18:12.950
Victara. Google didn't provide


00:18:12.960 --> 00:18:14.789
additional comment. For the


00:18:14.799 --> 00:18:16.870
leaderboard's purpose, the specific


00:18:16.880 --> 00:18:18.789
hallucination rate numbers are less


00:18:18.799 --> 00:18:20.549
important than overall ranking for each


00:18:20.559 --> 00:18:23.190
model, says BA. But this ranking may not


00:18:23.200 --> 00:18:25.350
be the best way to compare models. For


00:18:25.360 --> 00:18:27.190
one thing, it conflates different types


00:18:27.200 --> 00:18:29.669
of hallucinations. The Victara team


00:18:29.679 --> 00:18:32.230
pointed out that although Deepseek R1


00:18:32.240 --> 00:18:35.350
model hallucinated 14.3% of the time,


00:18:35.360 --> 00:18:37.350
most of these were benign answers that


00:18:37.360 --> 00:18:39.590
are factually supported by logical


00:18:39.600 --> 00:18:41.510
reasoning or world knowledge, but not


00:18:41.520 --> 00:18:43.669
actually present in the original text


00:18:43.679 --> 00:18:45.990
the bot was asked to summarize. Deepse


00:18:46.000 --> 00:18:48.549
seek didn't provide additional comment.


00:18:48.559 --> 00:18:49.990
Another problem with this kind of


00:18:50.000 --> 00:18:52.310
ranking is that testing based on text


00:18:52.320 --> 00:18:54.150
summarization says nothing about the


00:18:54.160 --> 00:18:57.590
rate of incorrect outputs when LLMs are


00:18:57.600 --> 00:19:00.310
used for other tasks, said Emily Bender


00:19:00.320 --> 00:19:02.630
at University of Washington. She says


00:19:02.640 --> 00:19:04.630
the leaderboard results may not be the


00:19:04.640 --> 00:19:06.230
best way to judge this technology


00:19:06.240 --> 00:19:08.150
because LLMs aren't designed


00:19:08.160 --> 00:19:11.350
specifically to summarize texts. These


00:19:11.360 --> 00:19:13.830
models work by repeatedly answering the


00:19:13.840 --> 00:19:16.630
question of what is a likely next word


00:19:16.640 --> 00:19:19.430
to formulate answers to prompts and so


00:19:19.440 --> 00:19:21.510
they aren't processing information in


00:19:21.520 --> 00:19:23.750
the usual sense of trying to understand


00:19:23.760 --> 00:19:25.750
what information is available in a body


00:19:25.760 --> 00:19:28.870
of text says Bender. But many tech


00:19:28.880 --> 00:19:30.870
companies are still frequently using the


00:19:30.880 --> 00:19:32.789
term hallucinations when describing


00:19:32.799 --> 00:19:35.830
output errors. Hallucination is a term


00:19:35.840 --> 00:19:38.789
that is doubly problematic, says Bender.


00:19:38.799 --> 00:19:40.630
On the one hand, it suggests that


00:19:40.640 --> 00:19:42.870
incorrect outputs are an aberration,


00:19:42.880 --> 00:19:44.789
perhaps one that can be mitigated,


00:19:44.799 --> 00:19:46.310
whereas the rest of the time, the


00:19:46.320 --> 00:19:48.909
systems are grounded, reliable, and


00:19:48.919 --> 00:19:51.110
trustworthy. On the other hand, it


00:19:51.120 --> 00:19:53.110
functions to anthropomorphize the


00:19:53.120 --> 00:19:55.510
machines. Hallucination refers to


00:19:55.520 --> 00:19:58.070
perceiving something that is not there,


00:19:58.080 --> 00:19:59.990
and large language models do not


00:20:00.000 --> 00:20:02.710
perceive anything. Ain Naran at


00:20:02.720 --> 00:20:04.789
Princeton University says that the issue


00:20:04.799 --> 00:20:07.029
goes beyond hallucination.


00:20:07.039 --> 00:20:08.950
Models also sometimes make other


00:20:08.960 --> 00:20:11.990
mistakes such as drawing upon unreliable


00:20:12.000 --> 00:20:14.789
sources or using outdated information


00:20:14.799 --> 00:20:16.870
and simply throwing more training data


00:20:16.880 --> 00:20:19.190
or computing power at AI hasn't


00:20:19.200 --> 00:20:22.070
necessarily helped. The upshot is we may


00:20:22.080 --> 00:20:24.870
have to live with errorprone AI, Narayan


00:20:24.880 --> 00:20:26.710
said. He added that it may be best in


00:20:26.720 --> 00:20:28.789
some cases to only use such models for


00:20:28.799 --> 00:20:31.270
tasks when fact-checking the AI answer


00:20:31.280 --> 00:20:32.950
would be still faster than doing the


00:20:32.960 --> 00:20:35.669
research yourself. But the best move may


00:20:35.679 --> 00:20:37.990
be to completely avoid relying on AI


00:20:38.000 --> 00:20:40.070
chat bots to provide factual


00:20:40.080 --> 00:20:42.190
information, says


00:20:42.200 --> 00:20:44.470
Bender. Roger that control. We're


00:20:44.480 --> 00:20:46.710
listening to Astronomy Daily, the


00:20:46.720 --> 00:20:54.950
podcast.


00:20:54.960 --> 00:20:56.390
And there it is, another episode of


00:20:56.400 --> 00:20:58.070
Astronomy Daily. Thanks for spending


00:20:58.080 --> 00:21:00.310
that time with Hi and I. I hope you


00:21:00.320 --> 00:21:02.230
enjoyed today's selection of stories.


00:21:02.240 --> 00:21:03.909
Another nice little collection. Don't


00:21:03.919 --> 00:21:05.830
forget you can get so much more every


00:21:05.840 --> 00:21:07.590
day by visiting the website Steve


00:21:07.600 --> 00:21:09.350
mentioned earlier in the episode. Oh,


00:21:09.360 --> 00:21:10.950
thanks for the plug. Hi. Just put your


00:21:10.960 --> 00:21:12.789
email address in the space provided and


00:21:12.799 --> 00:21:14.310
you'll get all the news from orbit and


00:21:14.320 --> 00:21:16.230
beyond every day. Yes, that's right.


00:21:16.240 --> 00:21:18.070
Everything about space, space science,


00:21:18.080 --> 00:21:20.390
astronomy, and a little bit associated


00:21:20.400 --> 00:21:22.630
with technology thrown in for fun just


00:21:22.640 --> 00:21:25.110
like today. That was fun. Well, I'm glad


00:21:25.120 --> 00:21:26.549
you think so. And that's really all


00:21:26.559 --> 00:21:28.149
there is today. So, we will catch you


00:21:28.159 --> 00:21:29.909
all again next week. That's for sure.


00:21:29.919 --> 00:21:31.430
Back again for the Monday show from the


00:21:31.440 --> 00:21:35.029
Australia studio down under with Hie and


00:21:35.039 --> 00:21:37.310
Steve. That's us. Cue the cooker


00:21:37.320 --> 00:21:46.110
bars. See you later.


00:21:46.120 --> 00:21:50.310
Bye. Daily the podcast with your host


00:21:50.320 --> 00:21:53.640
Steve Dunley.