WEBVTT

1
00:00:02.230 --> 00:00:05.030
STFC-RAL-CR03  R61: Good morning. Thank you for being here.

2
00:00:05.640 --> 00:00:10.340
STFC-RAL-CR03  R61: But they have been dominant everywhere. There is no shortage of talk.

3
00:00:10.540 --> 00:00:12.850
STFC-RAL-CR03  R61: Regarding its crucial potential.

4
00:00:13.120 --> 00:00:16.529
STFC-RAL-CR03  R61: Today, I'm excited to move beyond that height.

5
00:00:16.980 --> 00:00:19.799
STFC-RAL-CR03  R61: And do what we have actually done here.

6
00:00:20.410 --> 00:00:28.399
STFC-RAL-CR03  R61: We have created a system That is designed to support our project managers, PIs in real-time decision making.

7
00:00:28.710 --> 00:00:33.319
STFC-RAL-CR03  R61: My goal today is to focus on the theory and practical solution.

8
00:00:33.810 --> 00:00:38.200
STFC-RAL-CR03  R61: So, I want to keep this high level and accessible.

9
00:00:38.460 --> 00:00:40.719
STFC-RAL-CR03  R61: How do these tools actually work?

10
00:00:40.930 --> 00:00:42.780
STFC-RAL-CR03  R61: Rather than going to the code.

11
00:00:43.410 --> 00:00:46.429
STFC-RAL-CR03  R61: To get the heads up, Got it.

12
00:00:47.260 --> 00:00:58.980
STFC-RAL-CR03  R61: This presentation, throughout the presentation, you'll see some lot of sci-fi material to help demonstrate the data journey and the future we have built in.

13
00:01:00.120 --> 00:01:01.830
STFC-RAL-CR03  R61: During the today's talk.

14
00:01:08.560 --> 00:01:10.470
STFC-RAL-CR03  R61: Remaining good life.

15
00:01:10.850 --> 00:01:15.390
STFC-RAL-CR03  R61: We will look at the specific problem, the technical problem.

16
00:01:15.570 --> 00:01:19.240
STFC-RAL-CR03  R61: Dr. Pierre is…

17
00:01:19.730 --> 00:01:25.720
STFC-RAL-CR03  R61: And in the field up for the monument. And the methodology will tell you what tool

18
00:01:26.040 --> 00:01:32.210
STFC-RAL-CR03  R61: On how we actually transform the raw data into actionable insights.

19
00:01:32.460 --> 00:01:34.339
STFC-RAL-CR03  R61: Then we will look at the…

20
00:01:34.940 --> 00:01:41.860
STFC-RAL-CR03  R61: under the hold of the AI engine, the IT infrastructure, and how we optimize the model.

21
00:01:42.330 --> 00:01:43.680
STFC-RAL-CR03  R61: And finally.

22
00:01:44.760 --> 00:01:55.480
STFC-RAL-CR03  R61: We will share some of our performance data metrics of our models and framework we use to ensure our data is fair and unbatched.

23
00:01:55.740 --> 00:02:04.720
STFC-RAL-CR03  R61: Before I start, I want to say that this implementation, is a direct result of the operating system.

24
00:02:04.890 --> 00:02:07.300
STFC-RAL-CR03  R61: It has been a great opportunity.

25
00:02:07.420 --> 00:02:12.169
STFC-RAL-CR03  R61: To apply the training to our department's specific challenges.

26
00:02:17.840 --> 00:02:24.959
STFC-RAL-CR03  R61: So, about the business problems, we are looking at one of the specific project management models.

27
00:02:25.200 --> 00:02:28.029
STFC-RAL-CR03  R61: Known as the East Coast Management.

28
00:02:29.500 --> 00:02:35.179
STFC-RAL-CR03  R61: Since 2020, we have maintained Detailed the truth, broad.

29
00:02:35.540 --> 00:02:41.740
STFC-RAL-CR03  R61: And today, we are tracking over 200 active issues.

30
00:02:42.030 --> 00:02:46.480
STFC-RAL-CR03  R61: That is, over 20,000 words of technical updates.

31
00:02:47.420 --> 00:02:57.389
STFC-RAL-CR03  R61: Currently, our project managers, PIs, work package managers, are maintaining this log, as well as driving the project delivery.

32
00:02:57.690 --> 00:02:58.959
STFC-RAL-CR03  R61: I've just given time.

33
00:02:59.360 --> 00:03:04.480
STFC-RAL-CR03  R61: So, when a leader had to manually Go through all of this.

34
00:03:04.740 --> 00:03:07.890
STFC-RAL-CR03  R61: And decide what needs immediate attention.

35
00:03:08.040 --> 00:03:11.610
STFC-RAL-CR03  R61: It creates what I call as the paragraph gap.

36
00:03:12.630 --> 00:03:18.770
STFC-RAL-CR03  R61: And this gap has got two major risks, too. So, one is the black entry.

37
00:03:19.220 --> 00:03:21.360
STFC-RAL-CR03  R61: This is a difficult delay.

38
00:03:21.910 --> 00:03:26.660
STFC-RAL-CR03  R61: Who… This is typically, like, about,

39
00:03:26.980 --> 00:03:32.070
STFC-RAL-CR03  R61: about the history being updated and declared at the GED.

40
00:03:32.370 --> 00:03:37.749
STFC-RAL-CR03  R61: In a multi-million project, time is a luxury which will always don't have.

41
00:03:38.000 --> 00:03:41.860
STFC-RAL-CR03  R61: And the second minister is paid, subduede.

42
00:03:42.120 --> 00:03:45.350
STFC-RAL-CR03  R61: We are often seeing a class of perspective.

43
00:03:45.480 --> 00:03:52.629
STFC-RAL-CR03  R61: For example, a work package manager may naturally see an issue in their domain a top priority.

44
00:03:52.780 --> 00:03:57.090
STFC-RAL-CR03  R61: Well, the project manager, looking at the project launch day.

45
00:03:57.230 --> 00:04:04.210
STFC-RAL-CR03  R61: I mean, let's say that is true as a medium for locality. So, without the standards.

46
00:04:04.710 --> 00:04:21.329
STFC-RAL-CR03  R61: We do not have the ranking become more inconsistent. Therefore, our PIs and project managers need to see the big picture instantly, without spending the green… I mean, without spending too much of time, and…

47
00:04:21.550 --> 00:04:25.050
STFC-RAL-CR03  R61: update.

48
00:04:25.810 --> 00:04:30.350
STFC-RAL-CR03  R61: So, that challenge, we are solving between that, yeah.

49
00:04:30.870 --> 00:04:40.430
STFC-RAL-CR03  R61: So, transforming all these documented, technical, manual logos into an immediate, actionable, Interesting.

50
00:04:46.910 --> 00:04:48.720
STFC-RAL-CR03  R61: This is the entire map.

51
00:04:49.090 --> 00:04:50.609
STFC-RAL-CR03  R61: I took joked.

52
00:04:51.220 --> 00:05:08.000
STFC-RAL-CR03  R61: The model we use, the AI model, it is 12, 2.5 total. This isn't a simple question and answer setup. It's a multi-stage hybrid pipeline where Python and AI work together.

53
00:05:08.530 --> 00:05:18.479
STFC-RAL-CR03  R61: The computer workflow is split into two stages. One, we use the T1 linear extraction, and the analyzed test.

54
00:05:18.710 --> 00:05:20.970
STFC-RAL-CR03  R61: I'll talk about this unit in a moment.

55
00:05:21.380 --> 00:05:24.980
STFC-RAL-CR03  R61: But what makes this workplace pressure?

56
00:05:25.150 --> 00:05:29.049
STFC-RAL-CR03  R61: It's how different girls work together.

57
00:05:29.540 --> 00:05:31.270
STFC-RAL-CR03  R61: We have a Python project.

58
00:05:31.620 --> 00:05:36.049
STFC-RAL-CR03  R61: Managing the high precision tasks, like pulling the data.

59
00:05:36.310 --> 00:05:39.739
STFC-RAL-CR03  R61: Type question, and got a privacy marketing.

60
00:05:40.350 --> 00:05:49.829
STFC-RAL-CR03  R61: Meanwhile, the AI handles the language, that is, the attraction, the reality, and explanation, and systemarization.

61
00:05:50.280 --> 00:05:53.609
STFC-RAL-CR03  R61: all cornered with each other.

62
00:05:54.060 --> 00:06:00.420
STFC-RAL-CR03  R61: The entire practice is controlled by the strong motivator.

63
00:06:00.770 --> 00:06:12.739
STFC-RAL-CR03  R61: So these are specific problems that define the rules of engagement, telling the AI exactly how to test to ensure it meets our specific challenges.

64
00:06:13.050 --> 00:06:19.060
STFC-RAL-CR03  R61: Our example is data privacy might be filled later on, but in the next few slides.

65
00:06:19.300 --> 00:06:21.900
STFC-RAL-CR03  R61: This is a rock, rooted from.

66
00:06:22.010 --> 00:06:25.769
STFC-RAL-CR03  R61: And the air engine, and the detector.

67
00:06:31.800 --> 00:06:33.720
STFC-RAL-CR03  R61: Now we are looking at a plum.

68
00:06:34.060 --> 00:06:34.900
STFC-RAL-CR03  R61: More good.

69
00:06:36.060 --> 00:06:40.400
STFC-RAL-CR03  R61: So, if you look at the map, what we just saw on last slide, yeah.

70
00:06:41.170 --> 00:06:44.709
STFC-RAL-CR03  R61: So, this is a blueprint of our entire system.

71
00:06:45.120 --> 00:06:51.099
STFC-RAL-CR03  R61: We keep our code clean by putting all our rules or prompts into this area.

72
00:06:51.290 --> 00:06:53.949
STFC-RAL-CR03  R61: Didn't it? I was just turned very modular.

73
00:06:54.300 --> 00:06:57.099
STFC-RAL-CR03  R61: So, if we need to order anything.

74
00:06:57.390 --> 00:07:09.319
STFC-RAL-CR03  R61: See, individual prompt or anything related to, you know, the instruction, and we only update in the prompt area, total rest remaining the same.

75
00:07:12.270 --> 00:07:14.420
STFC-RAL-CR03  R61: But Dr. Bron.

76
00:07:15.310 --> 00:07:17.989
STFC-RAL-CR03  R61: We need the example to teach the AI.

77
00:07:18.270 --> 00:07:20.530
STFC-RAL-CR03  R61: Also known now to fine-tune.

78
00:07:20.820 --> 00:07:26.259
STFC-RAL-CR03  R61: And this… This tells the AI exactly what item to look for.

79
00:07:27.160 --> 00:07:31.180
STFC-RAL-CR03  R61: But it's reality analysis, that is the model.

80
00:07:31.370 --> 00:07:36.490
STFC-RAL-CR03  R61: We provide a clear… so that the AI knows what to do.

81
00:07:37.400 --> 00:07:41.059
STFC-RAL-CR03  R61: And the Trump is… the Trump is split into four paths.

82
00:07:41.230 --> 00:07:45.680
STFC-RAL-CR03  R61: the system prompt, and the user prompt, which I'll show you in the next slide.

83
00:07:45.810 --> 00:07:57.319
STFC-RAL-CR03  R61: The system prompt defines the AI's problem and how it needs to act. And the user prompt is a set of tasks that tells the AI what to do.

84
00:07:58.230 --> 00:08:03.209
STFC-RAL-CR03  R61: So, what makes this area unique is the helper function.

85
00:08:03.590 --> 00:08:05.069
STFC-RAL-CR03  R61: You have built into it.

86
00:08:05.250 --> 00:08:07.150
STFC-RAL-CR03  R61: They work like a kitten.

87
00:08:07.370 --> 00:08:11.180
STFC-RAL-CR03  R61: Before the AI looked into the world, these tools

88
00:08:11.370 --> 00:08:22.850
STFC-RAL-CR03  R61: the destruction, cleaning, and other tasks like privacy matching. And once the AI is done, these students load the process data back into our database.

89
00:08:23.280 --> 00:08:32.880
STFC-RAL-CR03  R61: By doing the heavy work of formatting here, we allow the AI to focus only on the regional plane and dimples.

90
00:08:38.690 --> 00:08:45.169
STFC-RAL-CR03  R61: So, to make model work, We balance the AI with our local hardware and our data sites.

91
00:08:45.280 --> 00:08:53.710
STFC-RAL-CR03  R61: First, we use a method called quantitations, or optimization. This runs the model at a reduced size.

92
00:08:54.130 --> 00:08:58.809
STFC-RAL-CR03  R61: It's helped the system to run faster on our local cluster.

93
00:08:59.240 --> 00:09:08.530
STFC-RAL-CR03  R61: It's the mathematical way of making the model smaller, so it uses the less memory while giving the high quality.

94
00:09:09.590 --> 00:09:11.900
STFC-RAL-CR03  R61: Our raw data comes from the toasters.

95
00:09:12.290 --> 00:09:15.949
STFC-RAL-CR03  R61: And more will be helper function.

96
00:09:16.290 --> 00:09:20.170
STFC-RAL-CR03  R61: And so, if you are on the other side.

97
00:09:20.490 --> 00:09:25.679
STFC-RAL-CR03  R61: We have the prompt, which I didn't talked about earlier, victim handling the prompt.

98
00:09:26.650 --> 00:09:28.729
STFC-RAL-CR03  R61: So, to do that, to get that.

99
00:09:29.590 --> 00:09:33.770
STFC-RAL-CR03  R61: We have the coordination layer that does the heavy lifting.

100
00:09:33.990 --> 00:09:38.419
STFC-RAL-CR03  R61: And this layer, I have to attach it to generate a report.

101
00:09:39.440 --> 00:09:48.989
STFC-RAL-CR03  R61: So, by keeping the rules separate and the coordination, we have created a system that is far, fair, and deeper.

102
00:09:53.040 --> 00:09:58.000
STFC-RAL-CR03  R61: So, we will look at the stage one in your direction.

103
00:09:58.370 --> 00:10:04.640
STFC-RAL-CR03  R61: So… What is the English term for is known entity recognition?

104
00:10:06.000 --> 00:10:15.900
STFC-RAL-CR03  R61: So this is the area the AI will automatically find the label or key information from the raw technical gloves.

105
00:10:16.000 --> 00:10:21.529
STFC-RAL-CR03  R61: into a defined package, such as a non-alitinian entities.

106
00:10:22.310 --> 00:10:24.730
STFC-RAL-CR03  R61: So, under using our tool.

107
00:10:25.470 --> 00:10:29.960
STFC-RAL-CR03  R61: We move to the middle area, where we do not look only at the workers.

108
00:10:30.590 --> 00:10:33.899
STFC-RAL-CR03  R61: But we are finding 11 categories of data.

109
00:10:34.030 --> 00:10:40.439
STFC-RAL-CR03  R61: I agree with them, We turn and line up tech into an order database.

110
00:10:41.270 --> 00:10:43.410
STFC-RAL-CR03  R61: So, why are we doing this?

111
00:10:43.520 --> 00:10:45.549
STFC-RAL-CR03  R61: The reason is Candel Database.

112
00:10:46.090 --> 00:10:48.320
STFC-RAL-CR03  R61: Which we don't really have.

113
00:10:48.550 --> 00:10:56.000
STFC-RAL-CR03  R61: I mean, we didn't… we have a lot of documents, we have reports in the past, but they are not easily searchable.

114
00:10:56.370 --> 00:11:00.250
STFC-RAL-CR03  R61: Therefore, we are building the base for the crucial knowledge graph.

115
00:11:00.680 --> 00:11:05.670
STFC-RAL-CR03  R61: And, AI targeting… drunk.

116
00:11:07.290 --> 00:11:19.490
STFC-RAL-CR03  R61: So, this makes sure that data we find in 2020, or previous data, says this will filter PI and PMs in 2020 tests and beyond.

117
00:11:25.210 --> 00:11:27.870
STFC-RAL-CR03  R61: Moving on to the death tool.

118
00:11:28.280 --> 00:11:33.000
STFC-RAL-CR03  R61: Once we have the structural data, we must solve the project management time.

119
00:11:33.750 --> 00:11:36.010
STFC-RAL-CR03  R61: This issue is a priority.

120
00:11:36.580 --> 00:11:40.610
STFC-RAL-CR03  R61: To confirm the problem, we have a defined setup and structure.

121
00:11:40.790 --> 00:11:46.940
STFC-RAL-CR03  R61: Telling the AI work to trade between 1 to 10, depending on the nature of the issue.

122
00:11:47.710 --> 00:11:49.680
STFC-RAL-CR03  R61: And this is happening fast.

123
00:11:49.780 --> 00:11:52.810
STFC-RAL-CR03  R61: Until we need to harvest, and harvest that manually work.

124
00:11:53.140 --> 00:11:56.960
STFC-RAL-CR03  R61: In the past, ranking them was a matter of opinion.

125
00:11:57.180 --> 00:12:06.100
STFC-RAL-CR03  R61: So, one person is a high risk, while the other says that shop is a medium risk. So, the other AI chopped up by us.

126
00:12:06.250 --> 00:12:13.259
STFC-RAL-CR03  R61: Using those set of rules, you can look at every issue on a clear 1 to 10 scale.

127
00:12:14.500 --> 00:12:17.100
STFC-RAL-CR03  R61: But we do not drop the AI applying it.

128
00:12:17.540 --> 00:12:20.710
STFC-RAL-CR03  R61: We also will be at the guardrails.

129
00:12:21.100 --> 00:12:25.440
STFC-RAL-CR03  R61: Within the… our road, we have mandatory minimum.

130
00:12:25.560 --> 00:12:32.769
STFC-RAL-CR03  R61: For example, if the AI finds a risk to a milestone or a lockup, Lots of guitar.

131
00:12:33.030 --> 00:12:35.539
STFC-RAL-CR03  R61: It cannot drop below that deadline.

132
00:12:35.760 --> 00:12:40.950
STFC-RAL-CR03  R61: This is how we put our expertise directly around AI and logic.

133
00:12:41.550 --> 00:12:45.779
STFC-RAL-CR03  R61: So, one thing really clear, is the AI will look

134
00:12:46.370 --> 00:12:49.859
STFC-RAL-CR03  R61: Only, about the grading and originally.

135
00:12:50.250 --> 00:12:52.250
STFC-RAL-CR03  R61: And the Python would do the math.

136
00:12:53.050 --> 00:13:02.859
STFC-RAL-CR03  R61: So, thanks, making the entire project faster with minimum memory, and of any component requirements.

137
00:13:03.200 --> 00:13:07.099
STFC-RAL-CR03  R61: So, from this, we have 3 distinct regions.

138
00:13:07.360 --> 00:13:08.870
STFC-RAL-CR03  R61: One is the boat.

139
00:13:09.240 --> 00:13:15.639
STFC-RAL-CR03  R61: of the story. The other one with the project manager will look at the limited summary.

140
00:13:16.180 --> 00:13:19.140
STFC-RAL-CR03  R61: of YAI.co.

141
00:13:19.340 --> 00:13:24.830
STFC-RAL-CR03  R61: And for the reporting purpose, We will be, doing the summarization.

142
00:13:29.990 --> 00:13:35.019
STFC-RAL-CR03  R61: Now we are on to the next pillar, the IT infrastructure and optimization.

143
00:13:35.850 --> 00:13:38.470
STFC-RAL-CR03  R61: So, our current infrastructure failures.

144
00:13:39.140 --> 00:13:48.459
STFC-RAL-CR03  R61: It runs on our local cluster. It uses the powerful signals, like MPGRA100, or Tesla 300.

145
00:13:48.840 --> 00:13:52.390
STFC-RAL-CR03  R61: So we are still using condo system.

146
00:13:52.610 --> 00:13:57.569
STFC-RAL-CR03  R61: Which manages the work across and, network in a common book.

147
00:13:58.340 --> 00:14:04.120
STFC-RAL-CR03  R61: The feet popped up this… Workflow with data leakage.

148
00:14:04.590 --> 00:14:16.970
STFC-RAL-CR03  R61: Unlike the consumer AI tools, like ChatGPT or Google Gemini, that send the data to the cloud, our system is entirely local, because our workers are sensitive

149
00:14:17.620 --> 00:14:20.009
STFC-RAL-CR03  R61: We make sure no data brew.

150
00:14:20.240 --> 00:14:21.940
STFC-RAL-CR03  R61: Internet network.

151
00:14:22.670 --> 00:14:23.590
STFC-RAL-CR03  R61: to help

152
00:14:23.720 --> 00:14:32.909
STFC-RAL-CR03  R61: Models run fast, we have reduced ties over the methods, which we talked earlier, and we also project the data in batches.

153
00:14:33.370 --> 00:14:40.759
STFC-RAL-CR03  R61: So, instead of making the model being too heavy, you can't clean and part intake.

154
00:14:43.910 --> 00:14:45.930
STFC-RAL-CR03  R61: To defend the booster.

155
00:14:46.070 --> 00:14:52.529
STFC-RAL-CR03  R61: We aim to build an AI application that runs simultaneously through the cloud for tech model.

156
00:14:52.950 --> 00:14:59.779
STFC-RAL-CR03  R61: by integrating Python here to manage the logic framework, to coordinate complex workflows.

157
00:14:59.940 --> 00:15:03.940
STFC-RAL-CR03  R61: We can leverage the material known as vis-a-prem service.

158
00:15:04.130 --> 00:15:16.210
STFC-RAL-CR03  R61: So, we have the hosted 11 active, you know, that is managed by the group of AI for 50 days. That is run by Jen, and then on the team.

159
00:15:16.500 --> 00:15:27.710
STFC-RAL-CR03  R61: So, we have to… the advantage of this is the rapid development. So, we can scale without managing the physical hardware and reduce complexity.

160
00:15:28.120 --> 00:15:32.609
STFC-RAL-CR03  R61: And allowing the developers to focus only on application logic.

161
00:15:33.010 --> 00:15:34.969
STFC-RAL-CR03  R61: Rather than people.

162
00:15:35.600 --> 00:15:38.819
STFC-RAL-CR03  R61: So… Do we have the Kooling aids yet?

163
00:15:39.170 --> 00:15:42.930
STFC-RAL-CR03  R61: alpaca, and the tuna.

164
00:15:43.610 --> 00:15:53.570
STFC-RAL-CR03  R61: The first one is… works like a strategy during the night. In conclusion, so you can pick whichever models for…

165
00:15:53.690 --> 00:15:56.299
STFC-RAL-CR03  R61: Well, thanks for you.

166
00:15:56.530 --> 00:16:03.620
STFC-RAL-CR03  R61: So, the only thing is, it is private of free, and it needs to be top of your login, and you can switch it.

167
00:16:04.150 --> 00:16:04.950
STFC-RAL-CR03  R61: Oh.

168
00:16:05.140 --> 00:16:10.809
STFC-RAL-CR03  R61: For the second leg, If you want to run your own AI application on the fly.

169
00:16:10.920 --> 00:16:13.759
STFC-RAL-CR03  R61: You can use the check-in link that is equal now.

170
00:16:14.380 --> 00:16:16.270
STFC-RAL-CR03  R61: So, you can join our group.

171
00:16:16.570 --> 00:16:19.960
STFC-RAL-CR03  R61: Just, yeah, to find out more.

172
00:16:20.720 --> 00:16:27.779
STFC-RAL-CR03  R61: Throughout… Just want to highlight that please do… please do link first on the CPD network.

173
00:16:34.090 --> 00:16:38.239
STFC-RAL-CR03  R61: Now we are at the final layer of the top, model validator.

174
00:16:38.800 --> 00:16:45.730
STFC-RAL-CR03  R61: During the initial stages, we have to decide which AI model to use for NER extraction.

175
00:16:46.130 --> 00:16:50.770
STFC-RAL-CR03  R61: So, our first key indicator was the token rulebook.

176
00:16:52.600 --> 00:17:02.329
STFC-RAL-CR03  R61: So, you will soon report the token or the basic unit of test, or think about the delivery, or fragment of words that AI used to process.

177
00:17:02.480 --> 00:17:10.789
STFC-RAL-CR03  R61: and generate the language, well, mathematically. So, we are actually in your custom structure on the tree model.

178
00:17:10.940 --> 00:17:13.159
STFC-RAL-CR03  R61: So, what is the Gemma to…

179
00:17:13.540 --> 00:17:24.329
STFC-RAL-CR03  R61: That is the 9 Boolean parameters. The other one is, mutual parameters, and the last thing is 22.5, that is about 14 Boolean.

180
00:17:24.470 --> 00:17:25.569
STFC-RAL-CR03  R61: primary deaths.

181
00:17:27.190 --> 00:17:30.089
STFC-RAL-CR03  R61: So, activity here at the total output.

182
00:17:34.420 --> 00:17:35.150
STFC-RAL-CR03  R61: Goodbye.

183
00:17:36.160 --> 00:17:39.959
STFC-RAL-CR03  R61: Alright. So just total output of that.

184
00:17:40.280 --> 00:17:42.969
STFC-RAL-CR03  R61: It's much higher than the rate.

185
00:17:44.480 --> 00:17:46.940
STFC-RAL-CR03  R61: And it's important to clarify that.

186
00:17:47.430 --> 00:18:07.300
STFC-RAL-CR03  R61: This higher token output forms, when comparing it with other models, doesn't mean the model is hallucinating for rumbling. We do the heavy lifting, which is actually taking time to attract all relevant categories and provide the detailed reasoning we'll project in our trunk.

187
00:18:08.040 --> 00:18:14.019
STFC-RAL-CR03  R61: And this is what we are looking for, the AI to understand both and how we work.

188
00:18:15.520 --> 00:18:20.219
STFC-RAL-CR03  R61: And this is where I was expecting Commissioner Nemo.

189
00:18:20.440 --> 00:18:25.049
STFC-RAL-CR03  R61: Despite being a 12 millionth parameter, is the 10…

190
00:18:25.980 --> 00:18:30.609
STFC-RAL-CR03  R61: 10, 11 hours, you get October, which is 9, 3 and 4 weeks.

191
00:18:31.390 --> 00:18:36.399
STFC-RAL-CR03  R61: So, this suggests that we find the model is maybe taking the total.

192
00:18:36.530 --> 00:18:42.889
STFC-RAL-CR03  R61: Or likely taking complex entities, or maybe failing to provide some anonymity.

193
00:18:43.790 --> 00:18:49.369
STFC-RAL-CR03  R61: And, however, the smaller models often fail if there are a lot of complexity.

194
00:18:51.720 --> 00:18:57.480
STFC-RAL-CR03  R61: And we will look at the model performed on different categories in the next slide.

195
00:19:02.400 --> 00:19:17.950
STFC-RAL-CR03  R61: So, when we look at the total number of categories, that is the 11 categories, so the attractions seem to be very high. So, whether it's the organization or financial, etc.

196
00:19:18.370 --> 00:19:22.730
STFC-RAL-CR03  R61: And it is only about… 14 other economies come.

197
00:19:25.260 --> 00:19:27.260
STFC-RAL-CR03  R61: And this is really detailed.

198
00:19:27.910 --> 00:19:31.620
STFC-RAL-CR03  R61: So, what makes this special, if you look at the second factor.

199
00:19:31.790 --> 00:19:35.590
STFC-RAL-CR03  R61: So, when I talk about the technical gap between

200
00:19:36.020 --> 00:19:42.819
STFC-RAL-CR03  R61: like FPCA, or PCB, the specific particle physics done.

201
00:19:42.940 --> 00:19:53.199
STFC-RAL-CR03  R61: So, whereas the other model could not identify. So, whereas the plan is able to check all of the… should be able to follow up what we want.

202
00:19:53.780 --> 00:19:57.190
STFC-RAL-CR03  R61: We couldn't get them twice, even for the rent.

203
00:19:57.670 --> 00:20:00.650
STFC-RAL-CR03  R61: 2.5, 13 billion bandwidth model.

204
00:20:02.720 --> 00:20:11.150
STFC-RAL-CR03  R61: This is the general report, etc. How much time it has taken for, Go ahead, pardon.

205
00:20:11.670 --> 00:20:15.729
STFC-RAL-CR03  R61: And the total strong is about $500,000.

206
00:20:16.000 --> 00:20:27.640
STFC-RAL-CR03  R61: So without talking, that would be something to touch on. And do they tend to talk about fund management, and the… and we don't have any problem

207
00:20:27.760 --> 00:20:29.350
STFC-RAL-CR03  R61: Okay, depart.

208
00:20:29.550 --> 00:20:40.860
STFC-RAL-CR03  R61: The GPU infrastructure, we had a maximum output utilization, and the average About, 200 or 400.

209
00:20:45.530 --> 00:20:46.460
STFC-RAL-CR03  R61: Shit.

210
00:20:46.860 --> 00:20:51.310
STFC-RAL-CR03  R61: So, here, we are talking about this test tool, the GUT story.

211
00:20:52.260 --> 00:20:55.760
STFC-RAL-CR03  R61: The literal actual output of our analyzer.

212
00:20:56.170 --> 00:21:08.399
STFC-RAL-CR03  R61: So, out of 211 individual issues, the AI has applied this logic and given its objective distribution. So, 37% is about the hiring

213
00:21:08.790 --> 00:21:22.729
STFC-RAL-CR03  R61: priority, so that means the project managers and the guys actually look at this, and then they can get to decide whether to look about this now or later, and these are the very low issues.

214
00:21:23.080 --> 00:21:24.220
STFC-RAL-CR03  R61: I'm…

215
00:21:24.920 --> 00:21:37.480
STFC-RAL-CR03  R61: But there's something interesting happening here. So, we have something called unknown. So, in the prompt, we have mentioned that if you don't find anything, so please let us know. So, this is what has happened here.

216
00:21:37.810 --> 00:21:44.029
STFC-RAL-CR03  R61: So this is where do I put the woman in the blue? It's, like, for the woman over time.

217
00:21:44.400 --> 00:21:51.559
STFC-RAL-CR03  R61: And this is how we manage from the flat test to the high-grade workflow.

218
00:21:55.680 --> 00:21:58.070
STFC-RAL-CR03  R61: So, talking about these methods.

219
00:22:01.670 --> 00:22:07.019
STFC-RAL-CR03  R61: And it's the past array of, to confirm. That is because

220
00:22:07.110 --> 00:22:25.569
STFC-RAL-CR03  R61: There is a problem with the document from, we didn't understand. So, you need to go back to the prompt, and check it again, and then come back to it. And I understand this, we do have some stickers to it, I'm very high, but it is captured, so we need to check what is actually happening.

221
00:22:25.960 --> 00:22:36.330
STFC-RAL-CR03  R61: Strategic, we have about 650 tokens, and these are the, already dropping 5 seconds, so the whole thing will run about 8 minutes.

222
00:22:36.540 --> 00:22:42.049
STFC-RAL-CR03  R61: So, it's all the… It's about being projected.

223
00:22:45.650 --> 00:22:55.310
STFC-RAL-CR03  R61: So, in the… Second stage, we have the AI running the two CRT expression, and the customization.

224
00:22:55.470 --> 00:22:59.910
STFC-RAL-CR03  R61: So, this is your total amount of what has happened at the… at every level.

225
00:23:00.830 --> 00:23:04.679
STFC-RAL-CR03  R61: So, for the explanatory step, we have also on the different

226
00:23:04.790 --> 00:23:12.310
STFC-RAL-CR03  R61: And the 34, and it's just summarized it, and for the reporting purpose, it,

227
00:23:15.830 --> 00:23:18.120
STFC-RAL-CR03  R61: And the other, tick me.

228
00:23:18.360 --> 00:23:21.810
STFC-RAL-CR03  R61: And it has taken very detailed explanation.

229
00:23:27.900 --> 00:23:30.120
STFC-RAL-CR03  R61: Now, let's talk about the AI edition.

230
00:23:30.680 --> 00:23:36.759
STFC-RAL-CR03  R61: We do not want the fast AI. We also want the AI to be fair and unfair.

231
00:23:37.500 --> 00:23:41.520
STFC-RAL-CR03  R61: So, the air, the whole system is stressed.

232
00:23:41.680 --> 00:23:44.289
STFC-RAL-CR03  R61: It's… the entire system is paragraph.

233
00:23:44.860 --> 00:23:51.290
STFC-RAL-CR03  R61: So, how do we know that AI is fair in one project or another during the analysis?

234
00:23:51.630 --> 00:23:54.800
STFC-RAL-CR03  R61: How do you know if the AI itself is domestic?

235
00:23:55.330 --> 00:23:58.510
STFC-RAL-CR03  R61: So, whether the AI didn't come out.

236
00:23:58.770 --> 00:24:09.780
STFC-RAL-CR03  R61: We meet all these models that are fair, transparent, empower, and accountable as other privacy or other training partner strategy.

237
00:24:12.520 --> 00:24:15.919
STFC-RAL-CR03  R61: So, due to the area which we are currently infected.

238
00:24:16.120 --> 00:24:24.899
STFC-RAL-CR03  R61: The first thing, there is someone in the room. Like, we have seen that. We have detected some issues, so we need to find out. So this is the first step.

239
00:24:25.030 --> 00:24:35.919
STFC-RAL-CR03  R61: One of the things that's to make sure, the repair business, and then giving security, making the proper results.

240
00:24:37.400 --> 00:24:42.190
STFC-RAL-CR03  R61: We do not let the AI go our company, and we have to make sure

241
00:24:42.470 --> 00:24:48.030
STFC-RAL-CR03  R61: There are some tech involved. As you can still need to become a post before.

242
00:24:48.510 --> 00:24:52.429
STFC-RAL-CR03  R61: And to make AI safe and fair, you have not,

243
00:24:53.070 --> 00:25:00.550
STFC-RAL-CR03  R61: tool for the first one. We did the open source Python library, and adjusting the internet of AI and models.

244
00:25:01.470 --> 00:25:21.050
STFC-RAL-CR03  R61: It is very powerful. It allows us to see the balance between the print and fairness. We can also generate reports and pick the best ways to stop any bias we find. This ensures our system is staying smooth and twists every project with the same level of scale.

245
00:25:21.570 --> 00:25:24.530
STFC-RAL-CR03  R61: And then we have a trailback by the government.

246
00:25:25.140 --> 00:25:29.569
STFC-RAL-CR03  R61: And we have got a set guidelines, and what, what, and whatnot.

247
00:25:29.780 --> 00:25:31.809
STFC-RAL-CR03  R61: You have to go to the animal.

248
00:25:32.010 --> 00:25:35.230
STFC-RAL-CR03  R61: Like, and to understand how

249
00:25:35.410 --> 00:25:42.760
STFC-RAL-CR03  R61: we follow the rules and the regulations to meet all the free complement. And you have to download a new

250
00:25:43.130 --> 00:25:50.180
STFC-RAL-CR03  R61: We have the, the questionnaire, where do we have the… about 100 to 200 questions?

251
00:25:50.310 --> 00:25:54.730
STFC-RAL-CR03  R61: We'll need to go through them with our stakeholders of the project.

252
00:25:54.850 --> 00:26:02.409
STFC-RAL-CR03  R61: And we get to decide what are the breakups to be considered in making sure all expenditures are met.

253
00:26:05.370 --> 00:26:09.939
STFC-RAL-CR03  R61: And this is one of the reports that we generated from the survey lab.

254
00:26:10.090 --> 00:26:16.240
STFC-RAL-CR03  R61: So the first one we call it, the fast success rate. So, one of the projects

255
00:26:16.590 --> 00:26:22.720
STFC-RAL-CR03  R61: tour. I thought, It's only about 93% that we are changing about 2%.

256
00:26:22.950 --> 00:26:27.140
STFC-RAL-CR03  R61: And the… the other one is we wanted to check whether

257
00:26:27.450 --> 00:26:39.619
STFC-RAL-CR03  R61: that are being computed or not, and whether there are many issues or not. So, addition to that idea of what's happening using Excel program module, Python modules.

258
00:26:43.140 --> 00:26:45.700
STFC-RAL-CR03  R61: And this is about the hallucination risk.

259
00:26:46.060 --> 00:26:53.020
STFC-RAL-CR03  R61: So, if you want to check all the things, like the fast success rate, and the error invalidation.

260
00:26:53.250 --> 00:27:05.549
STFC-RAL-CR03  R61: And then we do the total what is happening, and then finally we get some sort of, running, which is running. So that means we have… there is some problem with the…

261
00:27:05.800 --> 00:27:15.260
STFC-RAL-CR03  R61: structure of the prompt. So we need to check what's happening, and maybe there's a problem with detection. So that is what we're checking on.

262
00:27:15.870 --> 00:27:23.940
STFC-RAL-CR03  R61: And you do have the data government report, so it's due to, update of what's happening up in that project.

263
00:27:28.540 --> 00:27:30.390
STFC-RAL-CR03  R61: So finally, the conclusion.

264
00:27:31.210 --> 00:27:32.590
STFC-RAL-CR03  R61: This is the roadmap.

265
00:27:33.890 --> 00:27:38.470
STFC-RAL-CR03  R61: What AI applications are in the pipeline in the project of this?

266
00:27:38.830 --> 00:27:43.480
STFC-RAL-CR03  R61: In the next human, you have something called a task categorization.

267
00:27:43.790 --> 00:27:49.700
STFC-RAL-CR03  R61: So, where we categorize the financial data, Provided by the universities.

268
00:27:49.850 --> 00:27:54.140
STFC-RAL-CR03  R61: Is universities have dot their own format or template?

269
00:27:54.360 --> 00:28:08.399
STFC-RAL-CR03  R61: So we need to tackle items like money, or travel, etc. So, this will help to tell project managers, project managers have a sharp manual work.

270
00:28:08.810 --> 00:28:14.499
STFC-RAL-CR03  R61: In the near future, we plan to be able to listen to learn.

271
00:28:15.070 --> 00:28:17.510
STFC-RAL-CR03  R61: From the architect we are building up.

272
00:28:18.030 --> 00:28:20.390
STFC-RAL-CR03  R61: So, this will help us

273
00:28:20.760 --> 00:28:32.010
STFC-RAL-CR03  R61: to… not only for the two-step project, but also to understand what, if anything, had been missed, so we can use it for the

274
00:28:33.150 --> 00:28:39.199
STFC-RAL-CR03  R61: But even if the headquarting development, that's what… Lastly, in the long term.

275
00:28:40.130 --> 00:28:49.009
STFC-RAL-CR03  R61: So we have the predictive forecasting, and the opportunities are in life, like, the generation, and this is what we're trying to build up.

276
00:28:50.020 --> 00:28:56.579
STFC-RAL-CR03  R61: So, this ends my selected dog. I'm happy to say any proximity.

277
00:29:00.150 --> 00:29:01.929
STFC-RAL-CR03  R61: Yes, please.

278
00:29:02.060 --> 00:29:06.150
STFC-RAL-CR03  R61: It's… Have some questions.

279
00:29:07.790 --> 00:29:11.500
STFC-RAL-CR03  R61: Hi. So, any questions about,

280
00:29:11.620 --> 00:29:30.260
STFC-RAL-CR03  R61: should be token outputs for the different models, and with more tokens, and made a comment about, making a decision on the quality of that. How do you determine that more is better, as far as the tokens are concerned?

281
00:29:30.260 --> 00:29:35.569
STFC-RAL-CR03  R61: like you mentioned rambling, how were you able to check that it wasn't just… Excuse me.

282
00:29:35.680 --> 00:29:38.229
STFC-RAL-CR03  R61: So, open my mind, real fast.

283
00:29:38.430 --> 00:29:40.450
STFC-RAL-CR03  R61: Run the model.

284
00:29:41.110 --> 00:29:45.069
STFC-RAL-CR03  R61: So we see the harmony,

285
00:29:45.600 --> 00:29:47.599
STFC-RAL-CR03  R61: The output that is given up.

286
00:29:48.290 --> 00:29:55.549
STFC-RAL-CR03  R61: So, then we will look at the output of every… what we want to the region.

287
00:29:55.690 --> 00:29:59.609
STFC-RAL-CR03  R61: So, based on that, we'll check with the other model.

288
00:30:00.000 --> 00:30:09.170
STFC-RAL-CR03  R61: And then, finally, This is great that maybe we can have the conversation later on, like, about the tokenization.

289
00:30:09.310 --> 00:30:13.610
STFC-RAL-CR03  R61: To think about How we can use it in the future.

290
00:30:13.720 --> 00:30:14.540
STFC-RAL-CR03  R61: That's it.

291
00:30:16.860 --> 00:30:23.200
STFC-RAL-CR03  R61: Okay, because I was interested, the same reason why it's got these different models, and…

292
00:30:23.800 --> 00:30:29.200
STFC-RAL-CR03  R61: You see, in some cases, they behave in a biased way, at least check.

293
00:30:29.330 --> 00:30:43.860
STFC-RAL-CR03  R61: Yeah, to be testing these models for that. So, I see you have this furnace check, software. Did you use that only just with point 3, or were you able to look at the others? Because I think you'd be interesting. Right, again, you didn't see any…

294
00:30:44.310 --> 00:30:51.390
STFC-RAL-CR03  R61: And did they all perform similarly, or it was really out loud? Again, it depends. If you look only from the…

295
00:30:51.930 --> 00:30:55.829
STFC-RAL-CR03  R61: Let's say, mistrial, and it gives the new permanent check.

296
00:30:56.400 --> 00:31:07.880
STFC-RAL-CR03  R61: But in the firm run, you have different terminology type of functions, where you can check, like, halogen within the model, only the 10 model.

297
00:31:08.040 --> 00:31:23.249
STFC-RAL-CR03  R61: So, there are various types that we are looking at. So, we can see, like, the type of token that is being used, or if there is any hallucination, or if there's any, other things, like, you know, the…

298
00:31:23.380 --> 00:31:24.609
STFC-RAL-CR03  R61: Let's jump in.

299
00:31:25.170 --> 00:31:35.330
STFC-RAL-CR03  R61: We will collect the name, and we have the measuring the how the AI is thinking about it. So that, then we can do all that kind of thing.

300
00:31:35.670 --> 00:31:39.719
STFC-RAL-CR03  R61: And you didn't see… do you see any significant differences between the models, or…

301
00:31:40.170 --> 00:31:42.720
STFC-RAL-CR03  R61: But, it's unknown.

302
00:31:42.990 --> 00:31:47.969
STFC-RAL-CR03  R61: Joe, when we talk about the… it's not much different, unless…

303
00:31:48.090 --> 00:31:52.990
STFC-RAL-CR03  R61: In terms of the NBA attraction, Indeed!

304
00:31:53.120 --> 00:31:57.799
STFC-RAL-CR03  R61: That, the second chase, and we should… we started a little bit of,

305
00:31:58.110 --> 00:32:11.950
STFC-RAL-CR03  R61: different. Like, actually, gamma model is having a good quality, because it would be the type of data set that is being trained on. But when you compare with the 12 mod,

306
00:32:12.160 --> 00:32:15.220
STFC-RAL-CR03  R61: It will cover, like, slightly better.

307
00:32:16.140 --> 00:32:21.590
STFC-RAL-CR03  R61: Yes, yes. I mean, we have different type of,

308
00:32:22.030 --> 00:32:31.200
STFC-RAL-CR03  R61: one of the pair, but we have another thing, like, AI4, so we can filter some texting on stage as well.

309
00:32:31.870 --> 00:32:32.640
STFC-RAL-CR03  R61: Oops.

310
00:32:38.510 --> 00:32:43.220
STFC-RAL-CR03  R61: I didn't quite understand how the validation works, where you're…

311
00:32:43.470 --> 00:32:46.019
STFC-RAL-CR03  R61: From the point of view of the…

312
00:32:46.520 --> 00:32:53.740
STFC-RAL-CR03  R61: Human in the loop, in that, is the, other… do you have, then, people

313
00:32:54.070 --> 00:33:00.339
STFC-RAL-CR03  R61: Checking, or experts checking, that the output of the model

314
00:33:00.500 --> 00:33:09.900
STFC-RAL-CR03  R61: for a subset, or, I mean, how do they… because I think if the people can do everything, then you don't need the AI, so how do you…

315
00:33:11.290 --> 00:33:12.040
STFC-RAL-CR03  R61: Good luck.

316
00:33:13.100 --> 00:33:14.799
STFC-RAL-CR03  R61: This is a great documentation.

317
00:33:16.490 --> 00:33:18.210
STFC-RAL-CR03  R61: So, we have been…

318
00:33:18.510 --> 00:33:30.919
STFC-RAL-CR03  R61: explanation summary. So, for every score that the AI will test for, the PM or the BIs will read that explanation summary. So, that is a woman in the loop.

319
00:33:31.550 --> 00:33:46.690
STFC-RAL-CR03  R61: So, for example, I mean, they don't go through all of the issues, so our idea, we want to look at the top 10 issues, or top 20 issues. So, and that, we filter them out, and then we will look at the expense internally.

320
00:33:46.860 --> 00:34:00.580
STFC-RAL-CR03  R61: So this is where the woman in the loop's happening, so the PI will check that summary. Yeah, okay. If it's marked something as not critical as a phone.

321
00:34:00.780 --> 00:34:20.389
STFC-RAL-CR03  R61: Yes. How do you check? Do you have some fraction of those also checked that they are, that they really are low priority? Yeah. I mean, we had that incident happen, because one of the actual issues is actually important, but it has gone to a very low.

322
00:34:20.449 --> 00:34:26.959
STFC-RAL-CR03  R61: So, that is where the Fairland helps us to find out. I mean, there's some… there's some issues with the structure.

323
00:34:27.270 --> 00:34:33.549
STFC-RAL-CR03  R61: The… there are two ways happening. One is the problem with the JSL format of the front.

324
00:34:33.710 --> 00:34:38.919
STFC-RAL-CR03  R61: And the second one, if you remember the product module, we need to define the

325
00:34:39.090 --> 00:34:58.149
STFC-RAL-CR03  R61: We need to… to highlight what are the problems that we are facing, and we need to define clear. So, then, for example, we have a setup example to… to highlight, okay, for this issue, it needs to be, the high risk, or higher price, or medium. So.

326
00:34:58.330 --> 00:35:00.229
STFC-RAL-CR03  R61: It's a PI or a PM.

327
00:35:00.460 --> 00:35:08.780
STFC-RAL-CR03  R61: say that, okay, this is… the calculation is wrong. Then we go back to the wrong, and we add an example back to the

328
00:35:09.300 --> 00:35:15.640
STFC-RAL-CR03  R61: Back to the wrong system. So then, again, we run a retexture model again.

329
00:35:16.260 --> 00:35:17.970
STFC-RAL-CR03  R61: Is that how it works?

330
00:35:19.250 --> 00:35:26.959
STFC-RAL-CR03  R61: Yeah, what fraction… Of your… of all the things are checked by human.

331
00:35:27.840 --> 00:35:29.790
STFC-RAL-CR03  R61: At the moment,

332
00:35:30.410 --> 00:35:39.149
STFC-RAL-CR03  R61: And so far, the previous project managers are doing one of them, because, like… So they have their own checks.

333
00:35:39.950 --> 00:35:46.329
STFC-RAL-CR03  R61: It's, yeah, I mean… Because since we are implementing this for the first time.

334
00:35:46.850 --> 00:35:49.060
STFC-RAL-CR03  R61: So, they go through all of them.

335
00:35:49.310 --> 00:35:54.470
STFC-RAL-CR03  R61: Actually. But when the next time, when the new issues get added up.

336
00:35:55.060 --> 00:36:02.900
STFC-RAL-CR03  R61: It doesn't run the whole thing once again, but it takes a new issue, and it used to report on that particular new issue.

337
00:36:03.160 --> 00:36:09.459
STFC-RAL-CR03  R61: So, for example, the students have added up, like, maybe 2 or 3 times in a week.

338
00:36:10.160 --> 00:36:11.120
STFC-RAL-CR03  R61: So…

339
00:36:11.470 --> 00:36:19.390
STFC-RAL-CR03  R61: Sometimes it can be a lower issue or a higher issue, so it keeps running whenever the issues are up, and you choose the prompt in between.

340
00:36:19.840 --> 00:36:24.499
STFC-RAL-CR03  R61: To, at least We set up a tiny type of…

341
00:36:24.980 --> 00:36:38.580
STFC-RAL-CR03  R61: every Friday to run the issues, and give us very short report, saying that, okay, for this issue, what has happened this week, these are the problems, these are the scores, and then we compare with the entire data chain.

342
00:36:41.310 --> 00:36:41.990
STFC-RAL-CR03  R61: Thank you.

343
00:36:43.200 --> 00:36:44.330
STFC-RAL-CR03  R61: Yes, sure.

344
00:36:44.670 --> 00:36:46.219
STFC-RAL-CR03  R61: Oh yeah, sorry,

345
00:36:46.440 --> 00:36:51.030
STFC-RAL-CR03  R61: Do you get… do you get re… okay, I don't get so too popular. Do you get reproducible?

346
00:36:51.240 --> 00:36:52.960
STFC-RAL-CR03  R61: Results, if you run this.

347
00:36:53.420 --> 00:36:54.340
STFC-RAL-CR03  R61: Twice.

348
00:36:55.820 --> 00:36:56.490
STFC-RAL-CR03  R61: on.

349
00:36:57.660 --> 00:37:08.949
STFC-RAL-CR03  R61: refer to the… I mean, do you mean that, without being… So you run this loop, and then you get your text summarization and your explanation summary. You do that the same twice on…

350
00:37:09.200 --> 00:37:11.260
STFC-RAL-CR03  R61: the same input? Do you get the same…

351
00:37:12.030 --> 00:37:17.190
STFC-RAL-CR03  R61: I mean, planting air planted. I mean, it runs, like,

352
00:37:17.420 --> 00:37:20.349
STFC-RAL-CR03  R61: First, it's run on the parental terminal.

353
00:37:20.760 --> 00:37:25.320
STFC-RAL-CR03  R61: After you need mark calculation with quantity explanation summary.

354
00:37:25.540 --> 00:37:29.349
STFC-RAL-CR03  R61: After finishing this, we come back to the test summarization.

355
00:37:29.730 --> 00:37:35.549
STFC-RAL-CR03  R61: You have some set of things you're doing. Do you get the same result if you do it twice?

356
00:37:36.050 --> 00:37:45.009
STFC-RAL-CR03  R61: Oh, okay, got it. I mean, when you run it, yes, yeah. And because that is where we define the prompt. We made sure that we

357
00:37:45.180 --> 00:37:50.769
STFC-RAL-CR03  R61: That we highlighted that mixture doesn't change every time.

358
00:37:51.170 --> 00:37:56.079
STFC-RAL-CR03  R61: So, we did that test before. When we gave very generalized prompts.

359
00:37:56.230 --> 00:37:59.530
STFC-RAL-CR03  R61: And, we tend to exchange it maybe two or three times.

360
00:37:59.610 --> 00:38:10.060
STFC-RAL-CR03  R61: But then we defined… we made a series of check-up prompts, and then when we ran for the second or third time, it remained continuous. Unless…

361
00:38:10.060 --> 00:38:22.920
STFC-RAL-CR03  R61: when the project managers, because the issue doesn't remain changed for the entire project, so it keeps changing. So, maybe after 2 or 3 weeks, they update the issue. So, again, you want it.

362
00:38:22.960 --> 00:38:31.950
STFC-RAL-CR03  R61: And then it also highlights, I mean, we wrote the prompt in such a way that it also highlights what is the previous score, and this is the current score.

363
00:38:32.020 --> 00:38:34.090
STFC-RAL-CR03  R61: So, that is one of the tech…

364
00:38:34.450 --> 00:38:44.950
STFC-RAL-CR03  R61: Thank you for that idea. That is one of the steps that we are going to put. What was the previous, and if there is any changes that happened. Okay. And then a kind of follow-up question, so…

365
00:38:45.140 --> 00:38:49.639
STFC-RAL-CR03  R61: If you run this on an experiment like G, So June is a publicly…

366
00:38:50.320 --> 00:38:53.219
STFC-RAL-CR03  R61: There is public information out there, right?

367
00:38:53.370 --> 00:39:04.320
STFC-RAL-CR03  R61: So there are articles in New Scientists saying June is a deeply troubled project from, you know, 3 or 4 years ago. So then that sentiment analysis goes into the LLM, which then thinks.

368
00:39:04.890 --> 00:39:22.620
STFC-RAL-CR03  R61: things differently about that compared to, say, HyperK or, you know, Atlas or CMS, depending on the publicly available information that it has. So if you try changing the experiment name, so you could change it to a generic one that doesn't exist, you could start calling it good experiment or bad experiment, and see if it actually makes a difference to your output.

369
00:39:23.730 --> 00:39:24.610
STFC-RAL-CR03  R61: Don't.

370
00:39:26.380 --> 00:39:36.829
STFC-RAL-CR03  R61: to try to avoid that, this is one of the fair Python modules, It has one feature that…

371
00:39:37.900 --> 00:39:42.110
STFC-RAL-CR03  R61: We can find out whether the AI is by us or not.

372
00:39:42.650 --> 00:39:52.470
STFC-RAL-CR03  R61: That's one… one of the things that I'm touching on. But the second one, again, back to the prompt, that we need to make sure, again, we define the

373
00:39:52.610 --> 00:39:57.280
STFC-RAL-CR03  R61: Particular, prompting that, okay, we need to follow this.

374
00:39:57.510 --> 00:40:02.730
STFC-RAL-CR03  R61: And do not, you know, they do not provide any other

375
00:40:02.940 --> 00:40:05.559
STFC-RAL-CR03  R61: idea, apart from this on Lake.

376
00:40:05.780 --> 00:40:11.839
STFC-RAL-CR03  R61: There's a difference between the prompt and knowing that the AI is, is actually doing it.

377
00:40:13.130 --> 00:40:15.959
STFC-RAL-CR03  R61: I mean… Aren't they always…

378
00:40:16.930 --> 00:40:23.890
STFC-RAL-CR03  R61: You can say in the prompts, be fair, but I don't know if you can verify that it's being fair without trying to trick it somewhere.

379
00:40:25.830 --> 00:40:33.840
STFC-RAL-CR03  R61: So, in that case, we… when we look at the results from the, you know, the PI or the project manager.

380
00:40:34.290 --> 00:40:37.920
STFC-RAL-CR03  R61: That's where we look at the text summarization.

381
00:40:38.130 --> 00:40:43.870
STFC-RAL-CR03  R61: So we try and see if there's any change in the behavior, or something like that.

382
00:40:44.200 --> 00:40:47.129
STFC-RAL-CR03  R61: So that is one of the checks that we're doing, look.

383
00:40:47.560 --> 00:40:50.859
STFC-RAL-CR03  R61: Alright. That's good.

384
00:40:51.560 --> 00:40:53.970
STFC-RAL-CR03  R61: Yes. That was sort of overlap.

385
00:40:56.180 --> 00:40:56.890
STFC-RAL-CR03  R61: Okay.

386
00:40:57.450 --> 00:41:02.599
STFC-RAL-CR03  R61: Can I… sorry, this is a very specific question.

387
00:41:02.910 --> 00:41:13.329
STFC-RAL-CR03  R61: if you're… you're running on the PPD, AI machine that Jens has set up, I understand,

388
00:41:13.650 --> 00:41:18.669
STFC-RAL-CR03  R61: what sort of times do you run this? Because I'm also using this one.

389
00:41:18.910 --> 00:41:19.690
STFC-RAL-CR03  R61: Stop.

390
00:41:19.820 --> 00:41:21.840
STFC-RAL-CR03  R61: It's a different segment.

391
00:41:22.460 --> 00:41:37.569
STFC-RAL-CR03  R61: Sandeep's running these as condo jobs, right? Yeah. So they get submitted to the same type of machines, but because this machine is taken out of the queue, they're not competing. They're competing with other people submitting GPU jobs.

392
00:41:38.410 --> 00:41:43.690
STFC-RAL-CR03  R61: Okay, maybe I should do this, I think this is…

393
00:41:44.450 --> 00:41:47.999
STFC-RAL-CR03  R61: I was just wondering whether we'd be competing.

394
00:41:48.120 --> 00:41:50.840
STFC-RAL-CR03  R61: Sounds good.

395
00:41:51.260 --> 00:42:06.589
STFC-RAL-CR03  R61: So, that's quite interesting that these are run as… They can run on many machines, can they? Yes. I mean, we have only set up for 5 GPUs. So, what I have done is, in the condo, and with LipJo.

396
00:42:07.120 --> 00:42:09.980
STFC-RAL-CR03  R61: I have checked the time, like, every Friday.

397
00:42:10.300 --> 00:42:14.349
STFC-RAL-CR03  R61: 6 o'clock in the evening, after the end of the day.

398
00:42:14.480 --> 00:42:18.430
STFC-RAL-CR03  R61: So… And I'll try to see if there is any other…

399
00:42:18.640 --> 00:42:22.509
STFC-RAL-CR03  R61: whether the… whether the GPUs on both time.

400
00:42:22.890 --> 00:42:27.410
STFC-RAL-CR03  R61: I try not to interrupt or dictate related to

401
00:42:27.600 --> 00:42:40.149
STFC-RAL-CR03  R61: lab, if I remember correctly. It's a 12-minute job, I don't think it matters. Yeah, if it's a 12-minute job once a week, I think.

402
00:42:40.730 --> 00:42:47.020
STFC-RAL-CR03  R61: But I think, because of the condo job, it won't, on… based on their priority.

403
00:42:47.820 --> 00:42:48.660
STFC-RAL-CR03  R61: Maybe.

404
00:42:52.330 --> 00:42:56.950
STFC-RAL-CR03  R61: Sorry, they were just wondering whether, that seems to be quite…

405
00:42:57.170 --> 00:43:02.229
STFC-RAL-CR03  R61: If you're always running it on a Friday evening, is that the best time to get

406
00:43:02.940 --> 00:43:11.229
STFC-RAL-CR03  R61: results, because people then don't have time to look at them until Monday. Normally, the…

407
00:43:12.250 --> 00:43:15.770
STFC-RAL-CR03  R61: But the last week of August, we will look on Monday.

408
00:43:15.900 --> 00:43:17.779
STFC-RAL-CR03  R61: So we have the system set up.

409
00:43:17.990 --> 00:43:26.310
STFC-RAL-CR03  R61: So, for the previous week, last week of the update, like, in the chat will be updated on Monday. So, we collect all the data.

410
00:43:26.490 --> 00:43:28.660
STFC-RAL-CR03  R61: that this happened until due Friday.

411
00:43:29.020 --> 00:43:37.839
STFC-RAL-CR03  R61: So, at the moment, nobody's looking on Fridays, but the report that is generated, it goes to the email, and they're looking at whenever

412
00:43:38.290 --> 00:43:40.179
STFC-RAL-CR03  R61: pandemic.

413
00:43:44.170 --> 00:43:51.470
STFC-RAL-CR03  R61: Yeah, so coming back to the, the issue severity analysis stuff…

414
00:43:51.850 --> 00:44:01.509
STFC-RAL-CR03  R61: So, the main purpose of that is to generate this summary document, is it, that would then be read by PIs and PMs, or is that for…

415
00:44:01.830 --> 00:44:07.679
STFC-RAL-CR03  R61: what do you call it, Rose, that thing? Task managers or workstream managers as well.

416
00:44:08.780 --> 00:44:17.740
STFC-RAL-CR03  R61: It's different. I mean, normally it's just we call… we call each other project stakeholders, so who are involved in that.

417
00:44:18.110 --> 00:44:33.749
STFC-RAL-CR03  R61: particular work package, or something like that. So, normally, the project manager will work along with the PIs, or work package managers, and they coordinate with each other. So, whatever the project manager said, that information will be shared

418
00:44:34.100 --> 00:44:47.799
STFC-RAL-CR03  R61: Yeah, so, because in that case, yeah, so I work on, one of the, you know, one of the workstream managers for the CMS upgrade side, for the hardware work package, and,

419
00:44:48.000 --> 00:44:55.629
STFC-RAL-CR03  R61: So there, we often… we have chats with Lei every… every month or two to, update on the issues.

420
00:44:55.800 --> 00:45:01.120
STFC-RAL-CR03  R61: And then we also update the risk register as a result of that, but then…

421
00:45:01.360 --> 00:45:10.839
STFC-RAL-CR03  R61: what I was thinking was that at that stage, then, we're effectively ranking the severity of each of the issues in the risk register itself, right?

422
00:45:13.650 --> 00:45:24.269
STFC-RAL-CR03  R61: So then, yeah, we've kind of fed in all of the information, we know all of the information for this, so the main purpose is then for the DI to be able to read this as a summary, or…

423
00:45:25.720 --> 00:45:29.550
STFC-RAL-CR03  R61: Yeah, I'm trying to think out of this to the… into the workflow.

424
00:45:30.280 --> 00:45:34.620
STFC-RAL-CR03  R61: Yeah, I mean, that is what we planned. At the moment.

425
00:45:34.790 --> 00:45:38.319
STFC-RAL-CR03  R61: This is particularly trademarked me, because

426
00:45:38.510 --> 00:45:56.679
STFC-RAL-CR03  R61: the project managers are focused more on the project… on the risk register, so they are handling it by themselves. At some point, we will work… we will do the AI on the risk register as well. But we have, in our primary software, we have the… our own Monto Pablo analogy.

427
00:45:56.760 --> 00:46:03.170
STFC-RAL-CR03  R61: And where they feed into the information, and they calculate, and they come back to it.

428
00:46:03.350 --> 00:46:08.330
STFC-RAL-CR03  R61: So, because… Sometimes the issues are too long, or so…

429
00:46:08.470 --> 00:46:14.650
STFC-RAL-CR03  R61: It changes all the time, very frequently, so there is some sort of

430
00:46:14.710 --> 00:46:29.450
STFC-RAL-CR03  R61: summary of what's happening, and then they decide what happens, with their risk register or something. So normally, I think some days, too, we update the risk register, and, password. So, it's like…

431
00:46:29.910 --> 00:46:44.290
STFC-RAL-CR03  R61: Okay, but then the future change that you mentioned around the AI, the risk register, would that be the… one of these, AI setups taking the risk register, the project.

432
00:46:45.000 --> 00:46:48.060
STFC-RAL-CR03  R61: Workstream… the workstream managers have…

433
00:46:48.270 --> 00:46:59.760
STFC-RAL-CR03  R61: Updated with the… with the project manager, and using that as input, and then, using that to do some kind of analysis of likely future costs.

434
00:47:00.800 --> 00:47:06.960
STFC-RAL-CR03  R61: It is, I'm… When we talk about the future, I mean, I would expect not.

435
00:47:07.760 --> 00:47:10.669
STFC-RAL-CR03  R61: And near future, because…

436
00:47:10.980 --> 00:47:24.879
STFC-RAL-CR03  R61: which is very sensitive management, so we need to understand a bit more. Okay. So once our… once we get comfortable about it, and then we'll start working on the concept of the AI model.

437
00:47:26.220 --> 00:47:27.910
STFC-RAL-CR03  R61: Okay, okay.

438
00:47:28.070 --> 00:47:29.000
STFC-RAL-CR03  R61: I'm good, but…

439
00:47:29.260 --> 00:47:38.559
STFC-RAL-CR03  R61: as soon as we start, we will come to know about it, and this is one of the pipelines that is happening in the future of now. So, at the moment, we are looking at the gap.

440
00:47:38.880 --> 00:47:48.039
STFC-RAL-CR03  R61: That's a good idea. It takes a bit of time, because you need to analyze the data, the financial data. Right.

441
00:47:48.250 --> 00:47:54.219
STFC-RAL-CR03  R61: So hopefully, maybe after this, maybe we can update this job, hopefully.

442
00:47:55.410 --> 00:47:57.880
STFC-RAL-CR03  R61: Okay, and that would be to… to…

443
00:47:58.010 --> 00:48:09.299
STFC-RAL-CR03  R61: rank items based on the risk register? Oh, is that? We do have the system, I mean, already we have this top term of a software that is automatically pipeline.

444
00:48:09.520 --> 00:48:14.069
STFC-RAL-CR03  R61: But at the moment, the project managers, they're feeding the numbers.

445
00:48:14.190 --> 00:48:18.950
STFC-RAL-CR03  R61: Yeah, so… Again, when we collect that information.

446
00:48:19.240 --> 00:48:30.309
STFC-RAL-CR03  R61: at some point, we will be going to find a prompt and then make it work another way. Okay. But you need to be a little bit careful.

447
00:48:30.680 --> 00:48:36.310
STFC-RAL-CR03  R61: That is about the risk, and because there's other factors, you know, like the cost.

448
00:48:36.530 --> 00:48:40.279
STFC-RAL-CR03  R61: And the tile, and the chip wall. Yeah. So…

449
00:48:40.480 --> 00:48:48.510
STFC-RAL-CR03  R61: You need to find out, like, how they work together. I mean, to say that that becomes a big pipeline.

450
00:48:48.540 --> 00:49:08.180
STFC-RAL-CR03  R61: Right. So, we need to… we need a bit of time to look on that. Okay. Well, yeah, I mean, in a sense, I would have thought the risk register is easier to analyze because it's effectively just a set… actually, what I remember is that it's a set of numbers of the probabilities and the cost ranges, associated with risks.

451
00:49:09.000 --> 00:49:11.900
STFC-RAL-CR03  R61: And so that's already quantitative, whereas…

452
00:49:12.130 --> 00:49:17.970
STFC-RAL-CR03  R61: So it could just be ranked based on that without AI, whereas the issue is the text, so…

453
00:49:18.810 --> 00:49:25.609
STFC-RAL-CR03  R61: maybe lend themselves more to a tech sub. Yes, and one of the factors is, the Monte Carlo.

454
00:49:25.800 --> 00:49:28.649
STFC-RAL-CR03  R61: Which we haven't started working on that yet.

455
00:49:28.850 --> 00:49:35.719
STFC-RAL-CR03  R61: So, we need to understand a bit about the Monte Carlo, and then we'll start looking at the AIS.

456
00:49:36.640 --> 00:49:37.940
STFC-RAL-CR03  R61: Because…

457
00:49:38.790 --> 00:49:53.219
STFC-RAL-CR03  R61: we need to run the model, for the Monte Carlo iteration, like, thousand to a thousand times and 3, and so this is where you don't really understand that Monte Carlo analogy from the AI point of view.

458
00:49:53.810 --> 00:50:01.339
STFC-RAL-CR03  R61: till I'm doing… but… But would… would you consider replacing that Montecolor analysis with an AI one instead?

459
00:50:01.820 --> 00:50:08.490
STFC-RAL-CR03  R61: No, I mean… If you are talking about the normal population, that can be done.

460
00:50:08.880 --> 00:50:13.689
STFC-RAL-CR03  R61: But, we want the Monte Cablo as well, because… Okay.

461
00:50:13.880 --> 00:50:16.029
STFC-RAL-CR03  R61: So, because we want to…

462
00:50:16.280 --> 00:50:26.730
STFC-RAL-CR03  R61: let's say, unlike the delay in path, unlicatur, and what permit, takes probably the, 50% or 90%. So.

463
00:50:26.870 --> 00:50:31.599
STFC-RAL-CR03  R61: the running beta calculation is what the one you're

464
00:50:31.780 --> 00:50:35.030
STFC-RAL-CR03  R61: to application of Quanto Carlo, is where…

465
00:50:35.150 --> 00:50:50.379
STFC-RAL-CR03  R61: It requires a little bit of detailed understanding. So, which is why we are relying on our primary software to do that sort of analysis for us at the moment. So, at the moment, that is why we are manually creating

466
00:50:50.500 --> 00:50:52.130
STFC-RAL-CR03  R61: We are trying not.

467
00:50:53.410 --> 00:50:54.270
STFC-RAL-CR03  R61: Okay.

468
00:50:55.650 --> 00:50:58.870
STFC-RAL-CR03  R61: We have a raised hand on Zoom. Chris.

469
00:50:58.870 --> 00:50:59.469
Christopher Townsley - STFC UKRI: Yes, I think…

470
00:50:59.470 --> 00:51:01.150
STFC-RAL-CR03  R61: So, would you like to speak?

471
00:51:01.150 --> 00:51:18.250
Christopher Townsley - STFC UKRI: Yeah, it was just about the question before about potential bias on the sentiment analysis, say if a project has got into the news. I think my understanding is that this is a completely closed system, so wouldn't actually have access to the news, so this wouldn't creep in. Is that correct, Sandeep?

472
00:51:22.620 --> 00:51:23.530
STFC-RAL-CR03  R61: So…

473
00:51:24.960 --> 00:51:34.280
STFC-RAL-CR03  R61: So, the question was, depending on when the model was trained, that knowledge is part of the model, because it might have been based on text that is out in the open.

474
00:51:38.500 --> 00:51:42.650
STFC-RAL-CR03  R61: So once it has no access to external systems and can't query the web.

475
00:51:42.820 --> 00:51:52.259
STFC-RAL-CR03  R61: It might have been trained in that way beforehand, and it's a bias that is inherent to the model that then needs to be… you need to figure out whether it's in there or not.

476
00:51:57.140 --> 00:52:01.550
STFC-RAL-CR03  R61: Sorry, paraphrasing, Stuart.

477
00:52:03.030 --> 00:52:06.680
STFC-RAL-CR03  R61: Yes, Chris, if I sat me.

478
00:52:07.090 --> 00:52:08.120
STFC-RAL-CR03  R61: Okay.

479
00:52:08.230 --> 00:52:10.260
STFC-RAL-CR03  R61: Oh, I appreciate you.

480
00:52:10.370 --> 00:52:16.590
STFC-RAL-CR03  R61: Okay, so, are there other questions?

481
00:52:21.560 --> 00:52:26.480
STFC-RAL-CR03  R61: I have one last question, maybe. I was curious how to measure the

482
00:52:26.600 --> 00:52:33.389
STFC-RAL-CR03  R61: how you quantify the input data, and I… yeah, that's… we lost this line, but…

483
00:52:33.670 --> 00:52:37.910
STFC-RAL-CR03  R61: I think on the first slide, or second one, you've had,

484
00:52:38.400 --> 00:52:44.900
STFC-RAL-CR03  R61: The number of words. So, could you explain a bit what the meaning of that… that is?

485
00:52:46.400 --> 00:52:54.200
STFC-RAL-CR03  R61: It's looked like a little number of words, but maybe it has another meaning, which… Yes.

486
00:52:56.300 --> 00:53:02.319
STFC-RAL-CR03  R61: So what is this amount of data that you, your users?

487
00:53:03.520 --> 00:53:09.750
STFC-RAL-CR03  R61: What that's available at the moment. At the moment, we have the exclusive log.

488
00:53:10.470 --> 00:53:15.839
STFC-RAL-CR03  R61: Yes, the… when we worked on the project, we entered all that information.

489
00:53:16.290 --> 00:53:24.719
STFC-RAL-CR03  R61: So, like, for example, there is a crystal with an history, or a CPA, or any sort of car.

490
00:53:24.830 --> 00:53:34.859
STFC-RAL-CR03  R61: So, it extracts all of that information, and it counts all of it, and we trust it for the 1700.

491
00:53:35.230 --> 00:53:39.270
STFC-RAL-CR03  R61: So that is the input data from the juice log.

492
00:53:39.850 --> 00:53:54.490
STFC-RAL-CR03  R61: So you only use the issues log? At the moment, because we are focusing on the issues management, so this is one of the specific model of the project management, so we are focusing on the issues we are. So this is the data that we've got, it's out there.

493
00:53:54.490 --> 00:54:00.559
STFC-RAL-CR03  R61: And this is only for the open districts family, so we have closed issues, it's more than…

494
00:54:00.590 --> 00:54:08.959
STFC-RAL-CR03  R61: But this is, already open to about, 200 pictures for the entire project regular manager.

495
00:54:09.000 --> 00:54:10.120
STFC-RAL-CR03  R61: Oh, my seat.

496
00:54:15.900 --> 00:54:22.570
STFC-RAL-CR03  R61: Okay, thank you very much. So, let's thank Sandeep again, and we close it.

497
00:54:26.060 --> 00:54:31.479
STFC-RAL-CR03  R61: Yeah, please, like, some cookies, and, lunch. And lunch.

498
00:54:34.780 --> 00:54:44.090
STFC-RAL-CR03  R61: Lots of applause from Zoom, so it was very nice.