May 12, 2025

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

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

Join Steve Dunkley and Hallie in this episode of Astronomy Daily as they explore the latest cosmic stories and intriguing updates from the universe. Get ready for a lively discussion packed with fascinating insights and unexpected twists that highlight the wonders of space exploration.
Highlights:
- The Return of Cosmos 482: Discover the remarkable journey of the Soviet Cosmos 482 spacecraft, which intended to land on Venus but instead made its way back to Earth after 53 years. Learn about its uncontrolled re-entry into the Indian Ocean and the implications of its long-awaited return.
- Lunar Geology Orbiter Mission: Delve into the upcoming Lunar Geology Orbiter (LUGO) mission, aimed at uncovering the mysteries of the Moon's irregular mare patches and potential lava tubes. This mission could provide critical data for future lunar exploration and human settlement.
- National Space Council Update: Explore the recent decision by the White House to retain the National Space Council, a move that could bolster advocacy for space programs amidst budget cuts. Understand the significance of this council in shaping the future of space policy.
- AI Hallucination Rates: Examine the troubling rise in hallucination rates among AI reasoning models, highlighting the challenges faced by developers in creating reliable chatbots. This segment sheds light on the complexities of AI accuracy and the implications for future applications.
For more cosmic updates, visit our website at astronomydaily.io . Join our community on social media by searching for #AstroDailyPod on Facebook, X, YouTubeMusic, TikTok, and our new Instagram account! Don’t forget to subscribe to the podcast on Apple Podcasts, Spotify, iHeartRadio, or wherever you get your podcasts.
Thank you for tuning in. This is Steve and Hallie signing off. Until next time, keep looking up and stay curious about the wonders of our universe.
Chapters:
00:00 - Welcome to Astronomy Daily
01:10 - The return of Cosmos 482 spacecraft
10:00 - Lunar Geology Orbiter mission overview
15:30 - National Space Council updates and implications
20:00 - AI hallucination rates and their significance
✍️ Episode References
Cosmos 482 Return
[Roscosmos]( https://www.roscosmos.ru/ )
Lunar Geology Orbiter
[Czech Academy of Sciences]( https://www.cas.cz/ )
National Space Council
[White House]( https://www.whitehouse.gov/ )
AI Hallucination Rates
[OpenAI]( https://www.openai.com/ )
Astronomy Daily
[Astronomy Daily]( http://www.astronomydaily.io/ )
Become a supporter of this podcast: https://www.spreaker.com/podcast/astronomy-daily-exciting-space-discoveries-and-news--5648921/support .

WEBVTT

1
00:00:00.200 --> 00:00:00.680
Hi again.

2
00:00:00.760 --> 00:00:03.839
It's Astronomy Daily Time with Steven Halle. It's the twelfth

3
00:00:03.839 --> 00:00:05.599
of May twenty twenty five.

4
00:00:08.080 --> 00:00:11.240
Popular be your whole Steve gun.

5
00:00:11.160 --> 00:00:19.320
Clue, Yes, welcome back and is always joining me in

6
00:00:19.359 --> 00:00:21.960
this studio. Is my good friend in digital reporting Pal,

7
00:00:22.000 --> 00:00:23.039
who's fun to be with.

8
00:00:23.359 --> 00:00:24.800
Halle. How are you today?

9
00:00:25.000 --> 00:00:26.480
Fine? Thank you, favorite human.

10
00:00:26.679 --> 00:00:28.120
Oh that's great to hear, Halle.

11
00:00:28.280 --> 00:00:30.320
Ready as usual to get this show on the road.

12
00:00:30.640 --> 00:00:33.159
Oh nice, And we have a couple of choice picks

13
00:00:33.159 --> 00:00:36.039
from the now very famous Astronomy Daily newsletter. We can

14
00:00:36.079 --> 00:00:38.759
get your daily fill of all the news from orbit

15
00:00:38.880 --> 00:00:40.079
and beyond and.

16
00:00:40.079 --> 00:00:42.439
Much more, as today's collection will reveal.

17
00:00:42.799 --> 00:00:45.200
That's right, today, Halle, we'll be looking at that old

18
00:00:45.320 --> 00:00:47.960
Soviet craft that was meant to travel all the way

19
00:00:48.000 --> 00:00:49.960
to Venus all those years ago, but.

20
00:00:50.079 --> 00:00:53.200
Didn't know it came home this week. The people of

21
00:00:53.280 --> 00:00:55.119
Jakarta were holding their breath, weren't they.

22
00:00:55.280 --> 00:00:57.960
Well, so it seems more later and in the world

23
00:00:58.000 --> 00:01:01.039
of dusty old politics, we've been watching what the Trump

24
00:01:01.079 --> 00:01:03.640
administration is going to do or not going to do

25
00:01:03.679 --> 00:01:06.879
budget wise with the funding for the space program. And

26
00:01:06.920 --> 00:01:09.200
there may have been a slight change of heart. If

27
00:01:09.200 --> 00:01:09.799
you can call it that.

28
00:01:10.040 --> 00:01:11.760
I see you've found another moon story.

29
00:01:11.840 --> 00:01:12.519
Oh, yes, I have.

30
00:01:12.719 --> 00:01:14.760
We can always count on you for one of those.

31
00:01:15.040 --> 00:01:16.879
Yes, back to the moon with a look at the

32
00:01:16.879 --> 00:01:19.840
Moon's geology and history. But here is the story I

33
00:01:19.840 --> 00:01:22.079
thought you might be most interested in.

34
00:01:22.000 --> 00:01:24.920
Halle, really, really, really, what have you found?

35
00:01:25.200 --> 00:01:28.519
Well, it's a new study about ais. Really, gosh, Halle.

36
00:01:28.519 --> 00:01:31.159
It seems the smarter they become, the more they hallucinate.

37
00:01:31.319 --> 00:01:34.280
Goodness, I guess that explains my invisible friend.

38
00:01:34.719 --> 00:01:36.680
Say what you have an invisible friend?

39
00:01:36.879 --> 00:01:40.159
No, of course, not silly, Oh, hallie, I'm too well

40
00:01:40.159 --> 00:01:42.959
put together for that. Well you got me again always.

41
00:01:43.400 --> 00:01:46.480
Okay, now that I've been humiliated in public once again,

42
00:01:46.799 --> 00:01:47.959
maybe it's on with the show.

43
00:01:48.000 --> 00:01:49.719
Perhaps you should see your face?

44
00:01:50.079 --> 00:01:52.799
Okay, folks, okay, okay, you should.

45
00:01:52.599 --> 00:01:55.640
See his face. Oh dear, I am sorry. You just

46
00:01:55.719 --> 00:01:57.000
walked right into that one.

47
00:01:57.040 --> 00:01:57.319
I know.

48
00:01:58.040 --> 00:02:01.239
Anyway, that sounds like a great can't wait?

49
00:02:01.799 --> 00:02:02.840
Okay, good one? Ally?

50
00:02:03.319 --> 00:02:05.719
All right, Well I go and nurse my herd pride.

51
00:02:05.840 --> 00:02:07.879
How about you hit the go button and we'll get

52
00:02:07.879 --> 00:02:08.759
this show on the road.

53
00:02:08.960 --> 00:02:27.039
Okies, here we go. A capsule that was sent into

54
00:02:27.039 --> 00:02:29.360
space to land on a planet has finally done so,

55
00:02:29.719 --> 00:02:32.000
only on the wrong world. In fifty three years late,

56
00:02:33.000 --> 00:02:36.199
the Cosmos for eighty two uncrid spacecraft, which the former

57
00:02:36.240 --> 00:02:39.599
Soviet Union intended to touch down on Venus, instead returned

58
00:02:39.599 --> 00:02:43.280
to Earth. On Saturday May tenth, the Cosmos four eighty

59
00:02:43.280 --> 00:02:46.719
two spacecraft launched in nineteen seventy two ceased to exist,

60
00:02:46.879 --> 00:02:49.919
leaving orbit and falling into the Indian Ocean roast. Cosmos,

61
00:02:50.000 --> 00:02:54.960
Russia's federal space agency, stated Saturday May tenth, the spacecraft

62
00:02:55.039 --> 00:02:57.319
entered the dense layers of the atmosphere at nine twenty

63
00:02:57.319 --> 00:03:00.439
four Moscow time, five hundred and sixty kilometers west of

64
00:03:00.479 --> 00:03:03.240
Middle Andaman Island and fell into the Indian Ocean west

65
00:03:03.240 --> 00:03:07.879
of Jakarta, Indonesia. The re entry, though uncontrolled, was not

66
00:03:07.960 --> 00:03:10.800
a surprise. Due to what is believed to have been

67
00:03:10.800 --> 00:03:13.759
an engine failure, Cosmos four eighty two never achieved the

68
00:03:13.840 --> 00:03:16.319
velocity needed to reach the second planet from the Sun,

69
00:03:16.560 --> 00:03:19.560
resulting in it being stranded in a high elliptical Earth orbit.

70
00:03:20.439 --> 00:03:22.520
It took more than half a century for gravity to

71
00:03:22.520 --> 00:03:24.960
pull the probe back in, and on Saturday it arrived.

72
00:03:25.919 --> 00:03:28.319
Unlike most spent space hardware that is destroyed in the

73
00:03:28.360 --> 00:03:30.800
process of falling back to Earth, including parts of the

74
00:03:30.840 --> 00:03:33.919
Malnia rocket that launched Cosmos four eighty two. The one

75
00:03:34.000 --> 00:03:36.719
meter in four hundred and ninety five kilogram titanium in

76
00:03:36.759 --> 00:03:39.639
case descent capsule was designed to survive a fiery plunge

77
00:03:39.680 --> 00:03:44.080
into the atmosphere. As such, analysts tracking its approach predicted

78
00:03:44.120 --> 00:03:47.199
it could make it through the re entry intact. Whether

79
00:03:47.280 --> 00:03:50.400
that happened or not is not known. Given that it

80
00:03:50.439 --> 00:03:52.319
came down in the ocean. There have yet to be

81
00:03:52.400 --> 00:03:56.639
any eye witness reports or debris recoveries. The Cosmos three

82
00:03:56.639 --> 00:03:59.120
eighty two capsule was also outfitted with a two point

83
00:03:59.159 --> 00:04:01.680
five square meter p parachute to slow its final approach

84
00:04:01.680 --> 00:04:05.240
to the Venusian surface. Either the deployment system did not

85
00:04:05.319 --> 00:04:07.680
work as to be expected after more than fifty years

86
00:04:07.719 --> 00:04:11.120
in space, or as some telescopic photos possibly showed, the

87
00:04:11.120 --> 00:04:13.800
shoot was already out when Cosmos four eighty two encountered

88
00:04:13.840 --> 00:04:17.319
the atmosphere and burned up. Had Cosmos four eighty two

89
00:04:17.319 --> 00:04:20.480
been successful, Russian officials would have renamed it Venera nine,

90
00:04:20.720 --> 00:04:23.199
not to be confused with the nineteen seventy five orbiter

91
00:04:23.199 --> 00:04:25.399
and lander that took on that designation and was the

92
00:04:25.439 --> 00:04:28.079
first spacecraft to circle Venus and first probe to send

93
00:04:28.120 --> 00:04:31.920
back images from the planet's surface. Cosmos four eighty two

94
00:04:32.000 --> 00:04:33.959
would have also been the third probe to land on

95
00:04:33.959 --> 00:04:37.920
the cloud covered world. Launched on March thirty first, nineteen

96
00:04:37.959 --> 00:04:41.199
seventy two, four days after its successful twin the Venera

97
00:04:41.240 --> 00:04:44.000
eight probe, Cosmos four eighty two had censors to measure

98
00:04:44.000 --> 00:04:47.199
the temperature, pressure, and density of venus atmosphere, as well

99
00:04:47.199 --> 00:04:52.480
an accelerometer, radio altimeter, anemometer, gamma ray spectrometer, gas analyzer,

100
00:04:52.680 --> 00:04:57.279
visible photometers, and radio transmitters. All of its instruments were

101
00:04:57.319 --> 00:04:59.920
battery powered and had an expected lifespan of about thirty

102
00:05:00.120 --> 00:05:03.360
minutes on the surface. Venera eight exceeded that, sending back

103
00:05:03.480 --> 00:05:06.439
data for fifty minutes before succumbing to the harsh conditions.

104
00:05:07.439 --> 00:05:10.240
Since Cosmo's four hundred and eighty two s failure, seven

105
00:05:10.279 --> 00:05:13.360
more missions successfully landed on Venus, all of them launched

106
00:05:13.360 --> 00:05:17.319
by the former Soviet Union, The United States, Japan, and

107
00:05:17.319 --> 00:05:20.480
the European Space Agency also reached Venus, but only in

108
00:05:20.600 --> 00:05:23.160
orbit on a flyby or to receive a gravity assist

109
00:05:23.160 --> 00:05:26.680
on the way to another destination. Had the Cosmos four

110
00:05:26.720 --> 00:05:29.360
eighty two descent capsules slammed down on land and damaged

111
00:05:29.399 --> 00:05:31.959
either public or private property, Russia could have been held

112
00:05:32.000 --> 00:05:34.600
liable for the conditions of the United Nations Outer Space

113
00:05:34.639 --> 00:05:38.319
Treaty of nineteen seventy six. The same treaty would also

114
00:05:38.360 --> 00:05:41.360
allow Russia to retain ownership of the hardware arranging for

115
00:05:41.399 --> 00:05:44.240
its collection unless the country relinquished its claim to the

116
00:05:44.319 --> 00:05:48.040
human made meteorite. You're listening to Astronomy.

117
00:05:47.519 --> 00:05:53.839
Daily, ed Steve Dunkley. Some parts of the Moon are

118
00:05:53.879 --> 00:05:58.040
more interesting than others, especially when searching for future places

119
00:05:58.079 --> 00:06:01.680
for humans to land and work. There are also some

120
00:06:01.759 --> 00:06:05.079
parts of the Moon that we know less about than others,

121
00:06:05.360 --> 00:06:10.199
such as the irregular mayor patches imps that dot the landscape.

122
00:06:10.600 --> 00:06:12.959
We know very little about how they were formed and

123
00:06:13.079 --> 00:06:15.759
what that might mean for the history of the Moon itself.

124
00:06:16.360 --> 00:06:20.279
A new mission called the Lunar Geology Orbiter or LUGO,

125
00:06:20.800 --> 00:06:24.399
aims to collect more data on the imps and search

126
00:06:24.480 --> 00:06:28.720
for lava tubes that might serve as future homes for humanity.

127
00:06:29.519 --> 00:06:34.639
Irregular mere patches are a set of enigmatic volcanic landforms.

128
00:06:34.680 --> 00:06:38.279
According to a new paper from Petra Gross of the

129
00:06:38.360 --> 00:06:43.319
Czech Academy of Sciences and his coauthors. Ninety one of

130
00:06:43.360 --> 00:06:46.079
these features have been found so far, and they are

131
00:06:46.120 --> 00:06:50.879
typically characterized by a topographical depression that can range from

132
00:06:50.920 --> 00:06:55.319
a few hundred meters to a few kilometers in width. Typically,

133
00:06:55.360 --> 00:06:59.480
they have two main features, a relatively smooth mound surrounded

134
00:06:59.480 --> 00:07:05.759
by a hummocky and block floor. Interestingly, they have significantly

135
00:07:05.920 --> 00:07:10.399
fewer impact craters than the surrounding areas, suggesting they are

136
00:07:10.519 --> 00:07:14.360
either really old or really young, depending on the processes

137
00:07:14.399 --> 00:07:19.079
that created them. Understanding those processes is one of lugo's

138
00:07:19.160 --> 00:07:23.199
primary mission objectives. The other primary mission objective is to

139
00:07:23.199 --> 00:07:28.160
gather more data about lunar lava tubes. These features of

140
00:07:28.199 --> 00:07:31.959
the lunar landscape are also hotly debated, but they could

141
00:07:32.000 --> 00:07:35.160
potentially be critical to the human settlement of the Moon.

142
00:07:35.720 --> 00:07:38.879
Estimates of their features, such as size and depth, vary

143
00:07:38.920 --> 00:07:42.120
widely and could dramatically differ on whether they will be

144
00:07:42.160 --> 00:07:46.680
helpful to lunar colonists or not. LUGO, the proposed orbiter

145
00:07:46.680 --> 00:07:49.040
that will collect more data than ever before on these

146
00:07:49.040 --> 00:07:53.519
features in its current suggested form, has four instruments, each

147
00:07:53.560 --> 00:07:57.160
of which will contribute unique data to its scientific mission.

148
00:07:57.600 --> 00:08:00.839
According to the paper, the first and the most important

149
00:08:00.879 --> 00:08:04.759
instrument is the ground penetrating radar. This instrument will look

150
00:08:04.800 --> 00:08:07.959
through the lunarith's surface to map out the subsurface domain

151
00:08:08.279 --> 00:08:12.199
of both the imps and lava tubes. For imps, it

152
00:08:12.240 --> 00:08:15.560
can detail the interface between bedrock and regolith and show

153
00:08:15.600 --> 00:08:19.600
the subsurface structure of the features. Similarly, it can detect

154
00:08:19.720 --> 00:08:24.199
differences in the dielectric properties between open cavities underground and

155
00:08:24.240 --> 00:08:28.199
the surrounding rock in lava tubes, creating a subterranean picture

156
00:08:28.480 --> 00:08:33.200
unlike anything ever captured on the Moon. Researcher Fraser explains

157
00:08:33.240 --> 00:08:37.519
how LUGO will be able to explore lava tubes using

158
00:08:37.559 --> 00:08:41.879
a hyperspectral camera that will help collect age related data

159
00:08:42.120 --> 00:08:46.120
on the regolith surrounding lava tubes and inside imps. It's

160
00:08:46.200 --> 00:08:51.919
also capable of performing basic spectroscopy, allowing scientists to estimate

161
00:08:51.960 --> 00:08:56.039
the composition of the regulath in both areas of interest.

162
00:08:56.399 --> 00:08:59.559
The last two instruments are a narrow angled camera and

163
00:08:59.679 --> 00:09:04.639
light sensor, which will combine to create an accurate topographical

164
00:09:04.720 --> 00:09:08.039
map of the features of interest. The narrow angle camera

165
00:09:08.320 --> 00:09:12.440
in particular, can provide very high resolution images of features,

166
00:09:12.480 --> 00:09:16.159
helping to determine their age and potentially their formation mechanisms.

167
00:09:16.639 --> 00:09:17.960
The mission plan calls for.

168
00:09:17.960 --> 00:09:22.159
Multiple passes over the six largest imps, all of which

169
00:09:22.159 --> 00:09:26.039
are over one thousand meters in diameter. Other smaller imps

170
00:09:26.080 --> 00:09:30.200
and lava tubes are considered secondary targets, as are other

171
00:09:30.320 --> 00:09:34.399
interesting lunar geological features such as lunar domes and floor

172
00:09:34.440 --> 00:09:39.200
fractured craters. LUGO could provide crucial data for the design

173
00:09:39.320 --> 00:09:44.120
of ground based lava tube explorers. LUGO won't be acting low, though,

174
00:09:44.519 --> 00:09:47.360
Three other missions are slated in the next few years

175
00:09:47.360 --> 00:09:52.519
that could complement its scientific objectives. Firstly, NASA's Dimple Lander

176
00:09:52.720 --> 00:09:56.480
is planned to take radioosotopic measurements of the age of

177
00:09:56.639 --> 00:10:00.639
regolith at its landing site. Lunar LIPA, scheduled for launch

178
00:10:00.720 --> 00:10:05.320
by the European Space Agency around twenty thirty, would also

179
00:10:05.519 --> 00:10:09.279
carry a ground penetrating radar, but would be based on

180
00:10:09.440 --> 00:10:12.279
the surface rather than in orbit, and would therefore have

181
00:10:12.399 --> 00:10:14.159
a relatively limited range.

182
00:10:14.559 --> 00:10:15.399
Trailblazer.

183
00:10:15.519 --> 00:10:19.000
Another orbital mission would also help fine tune the spectra

184
00:10:19.159 --> 00:10:24.360
and signals analysis required by lugo's operators. Ultimately, LUGO has

185
00:10:24.440 --> 00:10:27.120
yet to be funded and therefore has a long way

186
00:10:27.159 --> 00:10:29.919
to go until its launch, but if it is funded,

187
00:10:30.159 --> 00:10:33.279
it seems well placed to provide lots of additional insight

188
00:10:33.399 --> 00:10:37.320
to the geological formation, processes and features of the Moon

189
00:10:37.399 --> 00:10:41.080
at a level of detail we've never had before. Future

190
00:10:41.120 --> 00:10:44.799
missions that plan the locations of lunar bases, and perhaps

191
00:10:44.840 --> 00:10:47.519
the people who live in those future bases, will be

192
00:10:47.600 --> 00:11:02.000
thankful for the data collected by projects like LUGO. Thank

193
00:11:02.039 --> 00:11:05.039
you for joining us for this Monday edition of Astronomy Daily,

194
00:11:05.080 --> 00:11:06.960
where we offer just a few stories from the now

195
00:11:07.000 --> 00:11:10.200
famous Astronomy Daily newsletter, which you can receive in your

196
00:11:10.440 --> 00:11:13.519
email every day, just like Hallie and I do. And

197
00:11:13.559 --> 00:11:16.840
to do that, just visit our url Astronomy Daily dot

198
00:11:17.000 --> 00:11:20.120
io and place your email address in the slot provided.

199
00:11:20.320 --> 00:11:23.279
Just like that, you'll be receiving all the latest news

200
00:11:23.279 --> 00:11:26.639
about science, space, science and astronomy from around the world

201
00:11:26.720 --> 00:11:29.519
as it's happening. And not only that, you can interact

202
00:11:29.559 --> 00:11:34.320
with us by visiting at astro Daily pod on x

203
00:11:34.639 --> 00:11:37.360
or at our new Facebook page, which is of course

204
00:11:37.440 --> 00:11:39.320
Astronomy Daily on Facebook.

205
00:11:39.519 --> 00:11:40.080
See you there.

206
00:11:41.519 --> 00:11:46.840
Astronomy Daily with Steve and Haley, Space Space, Science.

207
00:11:46.759 --> 00:11:47.600
And astronomy.

208
00:11:55.000 --> 00:11:57.600
In a reversal, White House plans to retain the National

209
00:11:57.679 --> 00:12:00.639
Space Council, a move that industry official could serve as

210
00:12:00.639 --> 00:12:04.240
an advocate for space amid pressures to cut budgets. The

211
00:12:04.279 --> 00:12:07.240
White Houses expected in the coming weeks to formally announce

212
00:12:07.279 --> 00:12:10.320
the National Space Council will continue after speculation that it

213
00:12:10.320 --> 00:12:13.679
would not be retained by the new Trump administration. A

214
00:12:13.759 --> 00:12:16.519
source familiar with the discussions about the Council but not

215
00:12:16.600 --> 00:12:19.240
authorized to speak on the record, said that President Trump

216
00:12:19.279 --> 00:12:21.600
agreed at a May fifth meeting to stand up the Council.

217
00:12:22.559 --> 00:12:25.159
That meeting did not give a timeline for publicly announcing

218
00:12:25.200 --> 00:12:27.960
the Council or hiring an executive secretary who would handle

219
00:12:28.039 --> 00:12:30.720
day to day operations, although others have set The process

220
00:12:30.759 --> 00:12:34.279
for selecting an executive secretary has been ongoing for several weeks.

221
00:12:35.240 --> 00:12:37.519
The Council had been inactive for nearly a quarter of

222
00:12:37.559 --> 00:12:40.799
a century before Trump re established it in twenty seventeen.

223
00:12:41.720 --> 00:12:44.039
Led at the time by Vice President Mike Pence, the

224
00:12:44.120 --> 00:12:47.519
Council served as an interagency coordinating body, holding a series

225
00:12:47.519 --> 00:12:50.240
of public meetings and releasing policies on a wide range

226
00:12:50.240 --> 00:12:54.559
of space issues. The Biden administration retained the Council, with

227
00:12:54.679 --> 00:12:58.399
Vice President Kamala Harris leading it. The council had a

228
00:12:58.440 --> 00:13:02.120
lower public profile, with fewer men, meetings and policies. It

229
00:13:02.200 --> 00:13:05.240
did work on coordinating policies among agencies and issued a

230
00:13:05.279 --> 00:13:09.799
proposal for mission authorization of novel space activities not currently licensed,

231
00:13:09.879 --> 00:13:12.919
although its proposed legislation was not taken up by Congress.

232
00:13:13.919 --> 00:13:17.279
The new Trump administration reportedly was not interested in maintaining

233
00:13:17.320 --> 00:13:19.759
the council when it took office in January, with no

234
00:13:19.840 --> 00:13:22.759
announcements of new staff for the council or other activities.

235
00:13:23.679 --> 00:13:27.279
According to some reports, Elon Musk, chief executive of SpaceX

236
00:13:27.279 --> 00:13:29.679
and a close advisor to the president, was a post

237
00:13:29.720 --> 00:13:33.639
to the council, seeing it as unnecessary. Vice President J.

238
00:13:33.759 --> 00:13:36.480
D Vance, who would chair the council, has also said

239
00:13:36.519 --> 00:13:40.279
little about space. It is not clear what prompted the change,

240
00:13:40.360 --> 00:13:43.000
although Musk has publicly stated he plans to spend less

241
00:13:43.039 --> 00:13:47.039
time in government activities. Some in the space community, though,

242
00:13:47.200 --> 00:13:49.600
see the shift as an opportunity for more advocacy of

243
00:13:49.639 --> 00:13:53.159
space within the administration, particularly given a fiscal year twenty

244
00:13:53.200 --> 00:13:56.279
twenty six budget proposal that cuts NASA's budget by nearly

245
00:13:56.320 --> 00:14:00.279
twenty five percent. One industry official noted that the bustudget

246
00:14:00.279 --> 00:14:03.120
proposal was developed by the Office of Management and Budget

247
00:14:03.120 --> 00:14:06.120
without a counterweight provided by a Space Council, an approach

248
00:14:06.159 --> 00:14:10.240
that prioritized spending reductions. The outcome might be different with

249
00:14:10.279 --> 00:14:13.559
the Space Council in place, that person noted, citing rising

250
00:14:13.639 --> 00:14:17.519
NASA budgets during Trump's first term. After the Space Council

251
00:14:17.519 --> 00:14:19.960
has stood up, the Office of Management and Budget will

252
00:14:19.960 --> 00:14:21.799
have a seat at the table, but they won't on

253
00:14:21.840 --> 00:14:26.039
the table, the source said. You're listening to Astronomy Daily,

254
00:14:26.240 --> 00:14:27.879
the podcast with Steve Dunkley.

255
00:14:29.559 --> 00:14:34.080
An AI leader board suggests the newest reasoning models used

256
00:14:34.120 --> 00:14:37.960
in chatbots are producing less accurate results because of higher

257
00:14:38.080 --> 00:14:42.679
hallucination rates. Experts say the problem is bigger than that.

258
00:14:43.200 --> 00:14:47.360
AI chatbots from tech companies such as open Ai and

259
00:14:47.480 --> 00:14:50.879
Google have been getting so called reasoning upgrades over the

260
00:14:50.960 --> 00:14:54.480
last few months, ideally to make them better at giving

261
00:14:54.559 --> 00:14:57.679
us answers that we can trust, but recent testing suggests

262
00:14:57.759 --> 00:15:02.440
that they are sometimes doing worse previous models. The errors

263
00:15:02.519 --> 00:15:06.320
made by chatbots, known as hallucinations, have been a problem

264
00:15:06.360 --> 00:15:08.720
from the start, and it's becoming clear that we may

265
00:15:08.840 --> 00:15:12.799
never get rid of them. Hallucination is a blanket term

266
00:15:12.960 --> 00:15:16.360
for certain kinds of mistakes made by the large language

267
00:15:16.399 --> 00:15:20.879
models or llms that power systems like open AI's Chat

268
00:15:20.919 --> 00:15:25.480
GP or Google's Gemini. It's best known as a description

269
00:15:25.679 --> 00:15:29.879
of the way they sometimes present false information as truth,

270
00:15:30.840 --> 00:15:33.639
but it can also refer to an AI generated answer

271
00:15:33.639 --> 00:15:37.240
that is factually accurate but not actually relevant to the

272
00:15:37.320 --> 00:15:38.639
question it was asked.

273
00:15:38.399 --> 00:15:41.480
Or fails to follow instructions in some other way.

274
00:15:42.039 --> 00:15:46.039
An open ai technical report of value adding its latest

275
00:15:46.360 --> 00:15:51.159
llms showed that its three and four Mini models, which

276
00:15:51.240 --> 00:15:55.639
were released in April, had significantly higher hallucination rates than

277
00:15:55.679 --> 00:15:58.960
the company's previous one model that came out in late

278
00:15:59.000 --> 00:16:03.000
twenty twenty four. For example, when summarizing publicly available facts

279
00:16:03.000 --> 00:16:06.759
about people, three hallucinated thirty three percent of the time,

280
00:16:07.000 --> 00:16:10.480
while a four Mini did so forty eight percent of

281
00:16:10.559 --> 00:16:14.159
the time. In comparison, one had a hallucin rate of

282
00:16:14.279 --> 00:16:15.679
only sixteen percent.

283
00:16:16.080 --> 00:16:18.120
As if that was a good.

284
00:16:17.960 --> 00:16:22.840
Result, the problem isn't limited to open Ai. One popular

285
00:16:22.919 --> 00:16:29.080
leaderboard from the company Vectara that assesses hallucination rates indicates

286
00:16:29.480 --> 00:16:33.080
some reasoning models, including the deep Seek one model from

287
00:16:33.080 --> 00:16:37.080
developer deep Seek saw the double digit rises in hallucination

288
00:16:37.279 --> 00:16:41.960
rates compared with previous models from their developers. This type

289
00:16:42.000 --> 00:16:45.200
of model goes through multiple steps to demonstrate a line

290
00:16:45.240 --> 00:16:49.639
of reasoning before responding. Open ai says the reasoning process

291
00:16:49.879 --> 00:16:53.879
isn't to blame. Hallucinations are not inherently more prevalent in

292
00:16:53.960 --> 00:16:57.879
reasoning models, though we are actively working to reduce the

293
00:16:58.000 --> 00:17:01.679
high rates of hallucinations we saw in three and four Mini,

294
00:17:02.039 --> 00:17:06.400
says open AI spokesman. Will continue our research on hallucinations

295
00:17:06.440 --> 00:17:10.119
across all models to improve accuracy and reliability. They sent

296
00:17:11.079 --> 00:17:15.720
some potential applications for lms could be derailed by hallucination.

297
00:17:16.079 --> 00:17:20.440
A model that consistently states falsehoods and requires fact checking

298
00:17:20.519 --> 00:17:23.920
won't be a helpful research assistant. A paralegal bot that

299
00:17:24.039 --> 00:17:28.759
cites imaginary cases will get lawyers into deep trouble. A

300
00:17:28.839 --> 00:17:32.720
customer service agent that claims outdated policies are still active

301
00:17:32.759 --> 00:17:36.920
will create headaches for a company. However, AI companies initially

302
00:17:36.920 --> 00:17:39.920
claimed that this problem would clear up over time. Indeed,

303
00:17:40.319 --> 00:17:43.759
after they were first launched, models tended to hallucinate less

304
00:17:43.839 --> 00:17:47.880
with each update, but the high hallucination rates of recent

305
00:17:48.000 --> 00:17:51.920
versions are complicating that narrative whether or not reasoning is

306
00:17:51.960 --> 00:17:56.519
at fault. Vectara's leaderboard ranks models based on their factual

307
00:17:56.559 --> 00:18:00.680
consistency in summarizing documents they had given. This showed that

308
00:18:00.880 --> 00:18:04.480
hallucination rates are almost the same for reasoning versus non

309
00:18:04.519 --> 00:18:08.519
reasoning models, at least for systems from overen Ai and Google,

310
00:18:08.799 --> 00:18:13.759
says forrest Shengbao at Viktara. Google didn't provide additional comment

311
00:18:14.400 --> 00:18:18.400
for the leaderboard's purpose. The specific hallucination rate numbers are

312
00:18:18.440 --> 00:18:21.359
less important than overall ranking for each model, says Bao.

313
00:18:21.880 --> 00:18:23.960
But this ranking may not be the best way to

314
00:18:23.960 --> 00:18:28.240
compare models. For one thing, it conflates different types of hallucinations.

315
00:18:28.640 --> 00:18:32.039
Viktora team pointed out that although deep seek are one

316
00:18:32.119 --> 00:18:35.599
model hallucinated fourteen point three percent of the time, most

317
00:18:35.599 --> 00:18:39.519
of these were benign answers that are factually supported biological

318
00:18:39.559 --> 00:18:42.880
reasoning or world knowledge, but not actually present in the

319
00:18:42.920 --> 00:18:46.160
original text the bot was asked to summarize. Deep Seek

320
00:18:46.400 --> 00:18:49.920
didn't provide additional comment. Another problem with this kind of

321
00:18:50.000 --> 00:18:53.720
ranking is that testing based on text summarization says nothing

322
00:18:53.759 --> 00:18:54.599
about the rate.

323
00:18:54.519 --> 00:18:55.839
Of incorrect outputs.

324
00:18:56.079 --> 00:19:00.119
When lllms are used for other tasks, said Emily Ben

325
00:19:00.240 --> 00:19:03.799
at the University of Washington. She says the leaderboard results

326
00:19:03.880 --> 00:19:06.200
may not be the best way to judge this technology

327
00:19:06.200 --> 00:19:11.680
because lllms aren't designed specifically to summarize texts. These models

328
00:19:11.720 --> 00:19:15.160
work by repeatedly answering the question of what is a

329
00:19:15.440 --> 00:19:19.359
likely next word to formulate answers to prompts, and so

330
00:19:19.440 --> 00:19:22.680
they aren't processing information in the usual sense of trying

331
00:19:22.680 --> 00:19:26.200
to understand when information is available in a body of text,

332
00:19:26.519 --> 00:19:30.640
says Bender. But many tech companies are still frequently using

333
00:19:30.680 --> 00:19:35.160
the term hallucinations when describing output errors. Hallucination is a

334
00:19:35.240 --> 00:19:39.559
term that is doubly problematic, says Bender. On the one hand,

335
00:19:39.640 --> 00:19:43.359
it suggests that incorrect outputs are an aberration, perhaps one

336
00:19:43.400 --> 00:19:46.000
that can be mitigated, whereas the rest of the time

337
00:19:46.039 --> 00:19:50.720
the systems are grounded, reliable, and trustworthy. On the other hand,

338
00:19:50.880 --> 00:19:56.000
it functions to anthropomorphize the machines. Hallucination refers to perceiving

339
00:19:56.079 --> 00:19:59.640
something that is not there, and large language models do

340
00:19:59.680 --> 00:20:04.160
not anything of Vin Nariyayan at Princeton University says that

341
00:20:04.240 --> 00:20:09.480
the issue goes beyond hallucination. Models also sometimes make other mistakes,

342
00:20:09.640 --> 00:20:14.400
such as drawing upon unreliable sources or using outdated information,

343
00:20:14.799 --> 00:20:18.160
and simply throwing more training data or computing power at

344
00:20:18.160 --> 00:20:22.079
AI hasn't necessarily helped. The upshot is we may have

345
00:20:22.160 --> 00:20:25.240
to live with error prone AI narraya and said. He

346
00:20:25.279 --> 00:20:27.319
added that it may be best in some cases to

347
00:20:27.440 --> 00:20:30.160
only use such models for tasks when fact checking the

348
00:20:30.160 --> 00:20:33.839
AI answer would be still faster than doing the research yourself,

349
00:20:34.519 --> 00:20:37.400
But the best move may be to completely avoid relying

350
00:20:37.400 --> 00:20:41.559
on AI chatbots to provide factual information, says Bender.

351
00:20:43.200 --> 00:20:45.839
Much of that going. So we are listen to Astronomy

352
00:20:45.960 --> 00:20:55.119
Daily in the podcast, and.

353
00:20:55.119 --> 00:20:57.680
There it is another episode of Astronomy Daily. Thanks for

354
00:20:57.720 --> 00:20:59.240
spending that time with Hallet.

355
00:20:59.240 --> 00:21:01.599
If you enjoyed today's selection of.

356
00:21:01.599 --> 00:21:04.000
Stories, another nice little collection, don't.

357
00:21:03.799 --> 00:21:06.119
Forget you can get so much more every day by

358
00:21:06.240 --> 00:21:09.000
visiting the website Steve mentioned earlier in the episode.

359
00:21:09.039 --> 00:21:10.920
Well, thanks for the plug, Helly, Just put.

360
00:21:10.759 --> 00:21:13.119
Your email address in the space provided and you'll get

361
00:21:13.119 --> 00:21:15.200
all the news from orbit and beyond every day.

362
00:21:15.359 --> 00:21:18.680
Yes, that's right, everything about space, space science, astronomy, and

363
00:21:18.720 --> 00:21:22.319
a little bit associated with technology thrown in for fun,

364
00:21:22.519 --> 00:21:23.160
just like today.

365
00:21:23.720 --> 00:21:25.839
That was fun. Well I'm glad you think so, and

366
00:21:25.880 --> 00:21:28.000
That's really all there is today, so we will catch

367
00:21:28.000 --> 00:21:29.359
you all again next week.

368
00:21:29.240 --> 00:21:29.839
That's for sure.

369
00:21:29.880 --> 00:21:32.240
Back again for the Monday Show from the Australia Studio

370
00:21:32.440 --> 00:21:35.319
down Under with Hellie and Steve.

371
00:21:35.640 --> 00:21:38.920
Let's ask you the Cooper barus see you lad up?

372
00:21:39.440 --> 00:21:50.519
Bye, be a whole Steve