AI's Visual Brain: Beyond Pictures, It's Reasoning!

Description

Chapters

Transcript

0:00 What I'm going to say, is I'm going to
0:02 say
0:05 your new
0:07 image gen tool is more a
0:13 visual
0:14 reasoning engine and [clears throat]
0:19 world model
0:22 visualizer
0:25 than it is a
0:28 picture maker. This seems like
0:34 an important
0:38 shift. Seems like an important shift.
0:48 And it's hard for me
0:52 to conceive
0:55 of all of the ways
0:58 I might use it
1:03 powerfully. I want you to think hard.
1:09 Tell it to think hard and it will. I
1:12 want you to think hard
1:14 about 20
1:17 use cases that
1:23 I can experiment
1:26 with
1:28 that reveal the true power
1:33 of what
1:35 this new capability
1:39 makes possible.
1:41 Okay. So, if you're new to this AI stuff
1:50 What is the
1:51 What is the the horrible
1:54 the horrible ego things about AI is that
1:57 there's a point at which when you use
1:59 this stuff
2:00 where you're just like, "Oh,
2:03 I'm the bottleneck. I'm the dum-dum."
2:07 Like Like there's a point at which like
2:09 you're like, "Okay, I want to think
2:11 about how to use a visual reasoning
2:13 model." And then you're like, uh And
2:16 like what I would typically do is I
2:17 would go to a whiteboard or I'd get out
2:19 a thing and start sketching ideas. You
2:21 don't need to do that anymore. But you
2:22 have to let go of
2:24 the hubris that you've got all the ideas
2:27 and only you can solve the problem. So,
2:29 what I'm doing here is getting ChatGPT
2:31 to help us. We're going to ask it to
2:33 think about all the ways we can use this
2:35 fancy new tool. I'll frame these less as
2:38 make a pretty image prompts and more as
2:40 experiments in visual thinking,
2:42 diagnosis, simulation, comparison,
2:44 planning, world building, and
2:45 persuasion. Feel
2:48 that will limit what you do and I really
2:54 want to understand
2:58 what new is possible. Here are 20
3:01 experiments that treat image generation
3:03 as a visual reasoning engine world model
3:05 visualizer. Visual counterfactuals. Show
3:09 the same city block in three versions.
3:11 One where housing policy policy favors
3:13 density,
3:14 one where cars dominate the planning,
3:16 and one where climate adaptation shaped
3:18 every decision. Label the visible clues.
3:21 Why it matters. It reveals systems. I'm
3:23 going to say
3:24 uh I'm going to say um
3:28 research
3:32 um
3:34 studies [snorts]
3:36 that base
3:39 your image in
3:42 the reality represented
3:45 by the data
3:48 and cite your sources. All right. So,
3:52 this is pretty cool. This is the cutaway
3:53 of the B-2 bomber, low observable flying
3:56 wing stealth bomber designed to
3:58 penetrate close air to fish defenses.
4:00 And by the way, I don't know if you've
4:01 seen any a lot of the uh there there've
4:04 been a lot of MDs on Twitter
4:06 that are having chat GPT make posters of
4:11 like
4:12 the body's immune response to a virus
4:15 with all the technical in it. And
4:17 they're basically like, "Yeah, these are
4:18 all accurate. This is all accurate."
4:22 All right. So, that worked. Give it a
4:24 messy human situation
4:26 and ask for the hidden structure. Create
4:28 a visual diagnosis of a dysfunctional
4:31 team meeting.
4:33 Watch the full replay at
4:34 community.thesalon.ai.