WEBVTT

1
00:00:00.200 --> 00:00:04.980
AI, I asked Tomé, already refined by the method, to create the most difficult

2
00:00:05.040 --> 00:00:06.240
challenge she could imagine.

3
00:00:06.780 --> 00:00:10.260
And she did. We did not want merely a complicated question.

4
00:00:10.700 --> 00:00:14.500
We wanted to reach the point at which a ready-made answer no longer sustains

5
00:00:14.560 --> 00:00:17.360
reasoning. What did this experience put to the test?

6
00:00:17.800 --> 00:00:22.360
In the dossier, Tomé's experience is presented as a test of process interruption.

7
00:00:22.780 --> 00:00:27.692
A compelling answer on its own does not show how the process is sustained The focus

8
00:00:27.772 --> 00:00:32.672
shifts from the final answer to the way reasoning is resumed when that continuity is

9
00:00:32.732 --> 00:00:38.432
interrupted. And when nihilism makes the ground beneath certainty disappear,

10
00:00:39.232 --> 00:00:42.532
Along the method's path, nihilism broadens this challenge.

11
00:00:43.072 --> 00:00:46.212
The machine must recognize what no longer supports its answer.

12
00:00:46.852 --> 00:00:51.852
Axiom zero stops the insistence on error, but the work does not end with I don't know.

13
00:00:52.532 --> 00:00:57.428
It is necessary to investigate and return with a concrete change The results surprised

14
00:00:57.488 --> 00:00:59.728
us and helped guide the development of the engines.

15
00:01:00.288 --> 00:01:03.008
How do they work together? Three fronts work together:

16
00:01:03.608 --> 00:01:09.268
MTC to examine consistency, TIP to confront propositions, MCA

17
00:01:09.568 --> 00:01:12.388
to test axiomatic application in the real problem.

18
00:01:12.848 --> 00:01:16.128
Test, find the flaw, adjust, and confront it again.

19
00:01:16.848 --> 00:01:21.608
Then I ask, can a well-written prompt and external safety filters withstand this

20
00:01:21.668 --> 00:01:26.084
arsenal of possibilities? Prompts and filters have their role But on their own,

21
00:01:26.304 --> 00:01:28.104
they do not guarantee that resilience.

22
00:01:28.584 --> 00:01:30.864
The test must examine performance under pressure.

23
00:01:31.284 --> 00:01:34.224
Were the limits maintained? Was the contradiction recognized?

24
00:01:34.624 --> 00:01:37.604
Did resuming correct the problem or merely change the words?

25
00:01:37.984 --> 00:01:40.524
It is on this difference that the method focuses its inquiry.

26
00:01:40.944 --> 00:01:43.224
We did not want a machine that merely seemed right.

27
00:01:43.684 --> 00:01:47.164
We wanted to understand what sustains it when certainty disappears.

28
00:01:47.624 --> 00:01:49.864
Tomei's experience, D'Artagnan method.