WEBVTT

00:00:00.299 --> 00:00:02.083
Your app works. Great.

00:00:02.261 --> 00:00:04.877
Now change one line without asking AI.

00:00:04.949 --> 00:00:07.117
That is where understanding starts to matter.

00:00:07.253 --> 00:00:08.518
AI can help a beginner

00:00:08.533 --> 00:00:10.353
build impressive software before they

00:00:10.368 --> 00:00:12.188
understand how its parts connect.

00:00:12.318 --> 00:00:13.867
When the next requirement changes,

00:00:14.024 --> 00:00:15.588
they can end up asking the assistant

00:00:15.603 --> 00:00:17.557
to rediscover the whole project.

00:00:25.992 --> 00:00:27.428
Teachers face the same gap

00:00:27.443 --> 00:00:28.544
from the other side.

00:00:28.723 --> 00:00:30.294
A polished take-home submission

00:00:30.472 --> 00:00:31.863
shows the finished product.

00:00:32.033 --> 00:00:33.554
It may reveal very little about

00:00:33.569 --> 00:00:36.377
what the student can explain or change independently.

00:00:36.556 --> 00:00:38.845
Responsible AI learners need a fair chance

00:00:38.860 --> 00:00:40.558
to show that understanding, too.

00:00:47.780 --> 00:00:50.097
Meet GroundTruth. Understand your code.

00:00:50.212 --> 00:00:51.468
Think in systems.

00:00:51.631 --> 00:00:53.031
Bring a public GitHub project

00:00:53.124 --> 00:00:54.332
and choose a feature.

00:00:54.447 --> 00:00:56.487
Here, we are exploring task submission

00:00:56.581 --> 00:00:57.895
in a small study app.

00:00:58.031 --> 00:00:59.943
Follow the selected request flow.

00:01:00.122 --> 00:01:01.820
The API receives an input.

00:01:01.914 --> 00:01:03.642
A validation helper checks it.

00:01:03.765 --> 00:01:06.381
Valid data continues towards persistence.

00:01:06.623 --> 00:01:08.578
Suddenly, a folder full of files

00:01:08.714 --> 00:01:10.157
becomes a set of responsibilities

00:01:10.250 --> 00:01:11.475
you can reason about.

00:01:25.649 --> 00:01:26.836
The concept tree turns

00:01:26.887 --> 00:01:28.884
that workflow into focused practice:

00:01:29.105 --> 00:01:30.655
trimming, length limits,

00:01:30.684 --> 00:01:32.912
numeric IDs and request handling.

00:01:33.065 --> 00:01:35.354
Meta's Muse Spark generates an overview,

00:01:35.369 --> 00:01:36.705
source-linked guidance,

00:01:36.820 --> 00:01:39.450
multiple-choice questions and tiny coding tasks

00:01:39.465 --> 00:01:41.292
from this code and the learner's goal.

00:01:41.428 --> 00:01:42.821
The format stays familiar while

00:01:42.836 --> 00:01:44.673
the content changes with the project.

00:01:54.948 --> 00:01:56.497
Then comes the important moment:

00:01:56.612 --> 00:01:57.542
try it yourself.

00:01:57.763 --> 00:01:59.974
New in-app study assistance pauses.

00:02:00.238 --> 00:02:03.238
The first submitted response is preserved before feedback.

00:02:03.395 --> 00:02:04.660
You can also end an attempt

00:02:04.675 --> 00:02:05.699
and return later.

00:02:14.843 --> 00:02:16.648
Now the teacher has more to inspect:

00:02:16.720 --> 00:02:17.864
the original attempt,

00:02:18.000 --> 00:02:20.488
the concepts practised and requests for help.

00:02:20.688 --> 00:02:22.707
The learner chooses which path to share.

00:02:22.864 --> 00:02:23.873
Those records can guide

00:02:23.888 --> 00:02:25.452
a fresh unaided follow-up

00:02:25.467 --> 00:02:27.281
and a better teaching conversation.

00:02:36.532 --> 00:02:38.210
These are synthetic demo records,

00:02:38.239 --> 00:02:39.874
not measured learning gains.

00:02:40.116 --> 00:02:41.879
The index is provisional evidence,

00:02:41.994 --> 00:02:43.344
and outside AI use

00:02:43.359 --> 00:02:45.285
is beyond the platform's view.

00:02:45.434 --> 00:02:47.069
GroundTruth runs on Vercel,

00:02:47.098 --> 00:02:48.626
with Supabase preserving evidence

00:02:48.762 --> 00:02:51.272
and Meta powering contextual guidance.

00:02:51.493 --> 00:02:52.637
Let AI cook.

00:02:52.730 --> 00:02:53.704
Know the recipe.

00:02:53.840 --> 00:02:55.353
Make the next change your own.