WEBVTT

1
00:00:00.500 --> 00:00:05.951
From The Assistant’s signed-in dashboard, open Digital Twin workspace. The following steps

2
00:00:05.951 --> 00:00:08.676
review the completed local training run.

3
00:00:09.876 --> 00:00:14.625
Use Professional documents to choose your source files. The completed run used

4
00:00:14.625 --> 00:00:19.373
Brian’s profile, AI software work, and engineering projects. The saved source documents

5
00:00:19.373 --> 00:00:20.560
are shown here.

6
00:00:21.760 --> 00:00:26.323
New datasets default to an eighty percent training and twenty percent held-out

7
00:00:26.323 --> 00:00:30.886
target, with whole source documents kept together. Document sizes can make the

8
00:00:30.886 --> 00:00:32.026
actual ratio differ.

9
00:00:33.226 --> 00:00:37.236
Advanced settings under the uploads let you choose a different train-test split

10
00:00:37.236 --> 00:00:41.246
when needed. The earlier run keeps its original dataset and recorded split.

11
00:00:42.446 --> 00:00:47.462
Prepare training examples builds a dataset from source paragraphs. This completed run

12
00:00:47.462 --> 00:00:50.805
contained twelve training examples and four held-out examples.

13
00:00:52.005 --> 00:00:56.228
The base model is Qwen three, zero point six billion parameters. This

14
00:00:56.228 --> 00:01:00.451
run trained a small LoRA adapter, with about one point one five

15
00:01:00.451 --> 00:01:04.675
million trainable parameters. It did not train the entire model from scratch.

16
00:01:05.875 --> 00:01:10.203
Train Digital Twin starts the local training run after the dataset is

17
00:01:10.203 --> 00:01:14.531
prepared. The recorded run completed three optimizer updates. The video skips the

18
00:01:14.531 --> 00:01:15.253
compute wait.

19
00:01:16.453 --> 00:01:21.415
Training took eighty-three point eight seconds on an HP EliteBook eight forty-five

20
00:01:21.415 --> 00:01:26.377
G eight laptop, with an AMD Ryzen five PRO fifty-six fifty U

21
00:01:26.377 --> 00:01:31.339
processor and thirty-two gigabytes of RAM. The run used four CPU threads,

22
00:01:31.339 --> 00:01:33.406
with an eight-gibibyte memory cap.

23
00:01:34.606 --> 00:01:39.449
Compare with base model evaluates the result after training. It is not

24
00:01:39.449 --> 00:01:44.292
another training step. This comparison took seventy-nine point five seconds. Both models

25
00:01:44.292 --> 00:01:49.135
scored zero out of four exact matches, so training success alone does

26
00:01:49.135 --> 00:01:50.750
not establish answer quality.

27
00:01:51.950 --> 00:01:56.769
Cancel active operation stops a running training or comparison job. It is

28
00:01:56.769 --> 00:02:01.589
optional and is not required to finish training. The controls shown here

29
00:02:01.589 --> 00:02:04.802
are disabled because the recorded run has finished.

30
00:02:06.002 --> 00:02:10.112
Use trained Digital Twin for chat selects the new adapter. This completed

31
00:02:10.112 --> 00:02:12.167
run already has that adapter selected.

32
00:02:13.367 --> 00:02:17.348
Enter the background question, then choose Ask your Digital Twin. The loading

33
00:02:17.348 --> 00:02:20.002
and generation wait is shortened in this edit.

34
00:02:21.202 --> 00:02:27.490
The answer describes Brian’s independent applied AI and research consulting, with engineering

35
00:02:27.490 --> 00:02:33.778
and computational neuroscience experience. The recorded reply took thirty-seven point seven seconds,

36
00:02:33.778 --> 00:02:39.017
including model loading. Earlier imperfect replies remain in the history.

37
00:02:40.217 --> 00:02:44.016
The source excerpts let you inspect the evidence behind the answer. They

38
00:02:44.016 --> 00:02:45.599
come from the uploaded documents.

39
00:02:46.799 --> 00:02:51.526
For a website example, enter B-Theory’s address and choose Fetch website preview.

40
00:02:51.526 --> 00:02:56.253
The app records the public page, final URL, capture time, and source

41
00:02:56.253 --> 00:02:56.647
hash.

42
00:02:57.847 --> 00:03:02.695
The example asks what B-Theory is about. Its proposed answer is an

43
00:03:02.695 --> 00:03:07.543
exact website excerpt. B-Theory is prepared separately; it has not been added

44
00:03:07.543 --> 00:03:09.968
to the selected adapter or trained.

45
00:03:11.168 --> 00:03:15.773
Download the website draft to save the source and example. Training this

46
00:03:15.773 --> 00:03:20.379
example later needs a new dataset, held-out examples, and a separately approved

47
00:03:20.379 --> 00:03:21.146
compute budget.

48
00:03:22.346 --> 00:03:27.842
B-Theory develops mathematical ways to describe states, observations, and change. It examines

49
00:03:27.842 --> 00:03:33.339
how observations separate possible states and help update models. Connections to minds

50
00:03:33.339 --> 00:03:36.087
and self-awareness are exploratory research questions.

51
00:03:37.287 --> 00:03:42.302
The workflow is upload, prepare, train, compare, select the trained adapter, and

52
00:03:42.302 --> 00:03:47.318
ask. Keep the measured results and supporting sources visible when judging what

53
00:03:47.318 --> 00:03:48.990
Digital Twin can do.