Shared Minds — Professor Dan O'Sullivan

Blink

Open the piece → www.ccfrost.com/blink

Reading

Kuleshov Effect

Technique invented by Lev Kuleshov

Close up shot of the face intercut with events, the shot stayed the same, but the next shot changes.

A sequence of images can build more meaning than a single shot

“my job as an editor is to gently prod the audience to look at a particular moment in a frame”

Cutting from a weapon to a victim means danger

this shot cut shot effect can be used to build an association that creates or lowers the tension

an object or prop can be called to intention by inserting shots of it

when an audience is invested in a film they are constantly look for associations

People assume a story is chronological, but thats not always the case

In the Blink of an Eye, Walter Murch

We accept the cut because it resembles the way images are juxtaposed in our dreams. (p. 34)
Film is like thought. It’s the closest to thought process of any art. (p. 35)
This time, the person is being assailed simultaneously by many conflicting emotions and thoughts, and is desperately (but unconsciously) using those blinks to try to separate these thoughts, sort things out, and regain some kind of control. (p. 35)
And that blink will occur where a cut could have happened, had the conversation been filmed. Not a frame earlier or later. (p. 36)
You would attach a series of electrodes to various points on your skull and simply think the film into existence. (p. 82)

Writing

We talked about how you don’t really decide your next thought, it just happens to you. But can other people determine your next thought?

Other people often determine our next thoughts, both in the moment and through their representation in our symbolic mental space. Someone impactful who has been dead 5 years can determine our next thought. So can someone who is not yet born. Only, the causation is not able to be established, due to our embedment in time.

Is that what any storyteller does?

A storyteller uses these same symbolic representations which we have established in our mental landscape around inherited aspects of human culture we learn through sublingual context. A storyteller partakes in similar sublingualities in our own to gain somatic access to our inner world. We expect a story in our tradition to follow typical beats, they understand this same expectation and, may use the informational asymmetry to their advantage.

Why do we love it so much in story where someone is providing the subsequent thought for us?

Stories allow us to test our mental assumptions of the world against a perennially troubling source of knowledge inte world storytellers. Because we all tell stories, we are always trying to square the stories of others. The subsequent thought becomes an opportunity to test our assumptions against shared reality.

Is that what the algorithms of social media and tiktok are doing?

Regenerative, by this I mean, self reflexive, regenerating algorithmic content as first seen on tiktok, forms a closed control loop with the viewer by capturing the informational flow by physically polling the user for their engagement in real time.

One sequence through a space of possible thought is always a thin slice of that reality, does that make story a little bit evil because it seduces people into one sequence hiding the bigger more complicated picture?

Stories are inherently laden with causation, because they rigorously apply the language of this, then that. this is true with communication broadly. Every particular human instance is a little bit evil because of our limited perceptual flows. Nothing can we grasp.

Giving your media persistence allows you to enshrine thought, what are the pros and cons of that, would impermanence be better for stupid and annoying thoughts?

Impermanence is the state of all things. All persistence is fallen. Change IS eternal

Is a sequence different from a flow time?

A time flow is a thermodynamic state, we are measuring energy through space. A sequence has consciousness attached, and thus a filter. A sequence has been formed, plucked from abstraction. A sequence has perspective and some assumption of causation.

Time is the feeling of change and one of the few things that is hard to doubt. Try to carefully examine your thoughts to see how they interact with time?

Does one thought coming after another in your thinking mean that there is a connection between them?

I have a feeling that the occurrence of thoughts in our brain arises from a shared well of subconscious experience. Maybe not in the causation of one thought to the next, except in obvious cases, but as thoughts arise from the body they can very well share a root somatic process.

We have talked about how the contents of subsequent thought is rarely about subsequent chronological event.

Does your taste in a good next thought change define your personality, and does it change over time?

Good thought as in, mental clarity or moral clarity? It is hard to argue against the self assurance provided by moral clarity

Is memory important to thought, can media help reinforce a thought and make it more famous and thus frequent within your train of thought?

Yes, 100% I’m thinking of 9/11

If you go back to the first thought after the second thought is there a recursive effect where the first thought is now different?

THOUGHT IS OUTSIDE OF TIME

Making

Developer journal, Claude compiled

Blink is a Shared Minds piece. You talk, it listens, and when you blink, a picture made from your story so far is swapped in. Each session below opens with my prompts in my own words, then what Claude did, then pictures.

The idea, built on the website — 29 to 30 September

My words

  1. 22:06 ask gpt astro how to make a shared minds image generator which uses your proactive subconscious to listen to the words you speak and then when you blink, an image it generated from your story so far is displayed. so ideally the image replaces when you blink, it should preload from replicate to handle this process. Use replicate haiku for the translation and prompting of the generator and nano bannana flash for the image. The language recognician engine is already built in shared minds dropplet, final version hosted on my website. Lets build this direclty in my website directory, as it has all the status guides. new branch. MAKE NO MISTAKES
  2. 22:40 words should not be on the image, find a faster image model
  3. 22:46 the record should always be on for typing
  4. 22:47 and there should not be an image for the face, lets get a undulating blob when eles are closed and a pair of eyes that follow yours when open. As many metrics for the eye as possible for this visualization
  5. 22:52 while I am typing for the first time, the words should spawn onto the page like droplet, but from left to right
  6. 22:58 the image style must be that of a roreshack test
  7. 23:25 text needs to go away when the image comes in. take the look of the site as a whole from droplet
  8. 23:38 can we represent the two eyes with two lines with a dot underneight?
  9. 23:40 when the eyes close, the two lines collapse down and the dot disapears
  10. 23:45 using the same infasctucture as droplet, lets make words that have been said more often get more weight behind the scenes n the prompt. I also want the the image to fade into view, and can we get faster images
  11. 23:47 make the mandlebroughts less abstract
  12. 23:47 more figurative
  13. 23:55 it only draws butteflys, make that change
  14. 23:56 turn the no eyes sign into a circle, and have the eyes form out of the circle when they are detected
  15. 23:57 get rid of the symstry rules in the mandlebrout
  16. 23:57 just black and white blog vibes
  17. 00:21 give me a fully enabled escape key like droplet to edit the prompt jsons
  18. 00:23 lets bring the full droplet effect on the outskirts of this canvas, bring in their controls

What Claude did

Built in the website repo, from 22:26 to 00:31:

Pictures

The first version: a Rorschach butterfly with Droplet's cursive sitting on top of it. That's where “text needs to go away when the image comes in” came from.

The first version of Blink: a black Rorschach-style butterfly inkblot with handwritten cursive words drawn over it. Another early Blink inkblot with cursive words overlapping the picture.

Password lock and AWS — 9 to 15 October

My words

  1. 10-09 00:11 lets implament the previous claude plan to make the photo generation website password locked and implament the appropriet aws database
  2. 10-15 18:19 where is the code for blink held
  3. 10-15 so I want to look at and make changes to the code, where can I do that
  4. 10-15 this all grew too fast

What Claude did

Bringing it into Shared Minds and getting it working again — 3 November, 21:20 to 22:00

My words

  1. 21:30 Can we work on the stripping it back down to the basic stuff and getting rid of most of the prompting? Another thing I noticed is that the prompting is missing. The prompt edit is missing from the, I can't actually edit well from the prompting in the, on the web interface. So let's get that fixed too.
  2. 21:36 Nothing is working.
  3. 21:36 It's visible, it's loading on the page, the merge works fine, the page loads, but it's not viewing my eyes or looking at the, I can't type any text in. Also, get rid of the play button, please. I'm tired of that. You can spawn an agent for getting rid of the play button and then bring that one back in when the change is done.
  4. 21:48 check out where we're at right now, and then, I mean, let's work from there, because I messed some stuff up.
  5. 21:57 the new stuff needs to include the old stuff. So like, here's the outcome of what I just got.
  6. 22:01 How many servers do I have here? This is crazy. I feel like I'm always adding another one. There are like two installed on VS Code, and then there's also the one that this is running.

What Claude did

Moved to the Shared Minds repo (Blink/, branch worktree-blink-import):

Pictures

The fishing story that prompted “the new stuff needs to include the old stuff”: the fisherman, then the next picture, which dropped him.

Blink showing a black ink picture of a fisherman casting from a riverbank. The next Blink picture: a lone black bear by the river, with the fisherman gone.

The progress indicator, five tries — 3 November, 22:00 to 22:50

My words

  1. 22:03 I also need, like, some kind of indicator where, Maybe the text that you're speaking turns into like a loading bar and then when you blink, it pops or something like that or it's spiraling out and after first, When it blinks, it pops.
  2. 22:08 So as we're going, please put these GIFs and stuff into the media file for my blog for this on Obsidian. I'm taking notes in this thing called, the project is called Blink in Obsidian. So the folder is called, itp shared minds blink do this with a subagent after every test you do
  3. 22:10 Please spawn a subagent to change it from a GIF. We don't want a GIF. We want something that's web compatible and lightweight.
  4. 22:13 Let's get rid of the text just continuing on the screen. I want to just see the first, yeah, it's actually all pretty good, but the text cannot be over the top of the picture. So maybe there should be like a text box and then like, something. Can we give me a couple options?
  5. 22:15 I want the text to disappear. Like, I want it to, like, after the first one, there should be this surprise where it curls up and then flies up into the picture, and then after that, there's no more. Well, actually, we should keep the user prompting for more, so we should have an animation probably do the same thing it did previously in the... probably do the same thing it did previously with the let's go picture to this side of the text on the column to the left.
  6. 22:24 text should not be on the top. It should be where you had it before. It just starts coming in from the side like a paragraph that goes on and on and on and on.
  7. 22:26 Claude, you should just operate in the same Chrome preview that I'm using so that we can both see the same stuff. And I want you to continue watching it all the time. Like, spawn a subagent to continue watching it.
  8. 22:28 start the watcher, also the way I see it is paragraph to the left and image to the right
  9. 22:32 Can we try a loading bar?
  10. 22:42 We need to go back to the prompt from the storybook with the bear thing. And we need to get rid of the loading bar. Like, maybe we can have like an abstract loading bar, but the text should not be a loading bar. It should be just like the text, instead of spiraling up, it all flies off letter by letter into the image. And then blinking will resummon the image. It flies off once the paragraph is done. So that's the indicator that the information is going in.
  11. 22:47 You need to imagine the words all dumping out onto the page. Get rid of the visual progress bar, get rid of all this stuff. Just have the words get sucked into the center of the picture.
  12. 22:48 like someone is filling up a bucket with the paragraph, it goes down and down and down and down and down and down and down. And then when the new image is ready, it all dumps out and goes into the picture. And then when you blink, it changes. Also, I don't know if we have some retarded sepia tone, but get rid of that too. It should be clear crisp black and white, and the prompt should reflect that
  13. 22:54 this outcome is amazing, write a summary of our work and put it my obsidian. agent 1 summary

What Claude did

Pictures

The tries, in order (there's no recording of try 3, the layout):

The bucket pour:

Prompt tests (the story kept, crisp black and white, a vignette):

A restored Blink story: waiting words in the left column and a black ink fisherman, bears and salmon on the right. A crisp two-tone black ink picture of a fisherman, bears and salmon by a forested river. The same river scene drawn as a vignette, with white margins on every side.

What the watcher caught: the six screenshots with a real picture in them.

Watcher screenshot: an ink figure surrounded by floating pages. Watcher screenshot: a person at a desk under ink clouds. Watcher screenshot: a figure at a laptop with an ink wave rising behind them. Watcher screenshot: salmon smoking cigarettes on a riverbank. Watcher screenshot: a fish holding a cigarette by the water. Watcher screenshot: fish leaping after flies.

Saving the story, soft edges, publishing — 3 November, 23:00 to 23:42

My words

  1. 23:01 Can you actually put the screenshots in there in between my voice so we can get a good idea of what it was saying while I was, what I was seeing while I was talking?
  2. 23:12 does this page save data in a local json?
  3. 23:16 let's get a JSON that can I can save in the browser. I want to use JSON to save it in the browser. So everyone can continue the story on their own time.
  4. 23:18 I'm finishing up my thing for Shared Minds for the assignment time. The title is Blink. We are working in the Shared Minds development folder. What I want to do is export the project Blink over into my website. I want to be able to access it from ccfrost.com/blink. Then begin to draft a blog post. The three headings are, like all my Shared Minds, reading, writing, making.
  5. 23:20 can we save the picture too or no
  6. 23:31 Please also make it so that the image doesn't generate with edges that ruin the continuity between the image and the text on the side.
  7. 23:42 push thest changes to my website

What Claude did

Pictures

Hard edge against faded edge:

Three versions of the same river picture side by side: no edge fade, a narrow fade and a wide fade into the white page.

GPT + Claude review — 4 November

My words

  1. 00:04 after the blink project website migrations finishes use codex to review the changes to my website. codex should look at the code base, and you two should work out my intentions from the transcripts in my obsidian to propose any final changes. then work the code over for bugs and security, remmbering that its okay for it to use the itp proxy as much as it wants. finally, when you and gpt are satasfied, merge the branch, update the dev log in obsidian and the website as gpt + claude review. adhere better to my formatting for the website for example
  2. 07:57 continue, make the mic start off

What GPT + Claude did

Claude and GPT (through the Codex CLI) each reviewed the website branch on their own, read my voice log and the prompts above for what Blink is meant to do, then compared notes and fixed what they agreed on:

Checked in a headless Chrome against the class proxy (21 checks): the mic starting off, the pour, the crossfade, a story with two pictures coming back after a reload, a picture whose file is gone coming back from its kept copy, an expired picture redrawn, words typed mid-restore kept, and “new story”.

The paragraph dumps, a text-only Esc panel, “tell me a story” — 4 November

My words

  1. 09:50 lets bring the blink code thats on the website back over to the shared minds folder
  2. 09:52 bring the changes we made in the blink website back into the shared minds folder
  3. 10:04 when the text dumps out, I want all the text to go, not just whats being shown. trick the user. after it dumps, the paragraph should be empty. additionally, currently it spirals line by line, I want it to dump all at once
  4. 10:09 get rid of the droplet controls for the esc tab on blink, text controls are fine
  5. 10:23 bring it to main
  6. 10:27 lets make a startup text appear in the style of the rest of the words that says tell me a story, it should go across the whole page. at the same time, lets center the text box and recording in a special section just below the image
  7. 13:27 please import the latest version of my blink from shared minds, then pull the latest context from obsidian, update my page and the expereince, show me a preview and push
  8. 13:34 yes, but show me before you push it, I want to test it one more time
  9. 13:36 this bar needs to extend across the whole bottom of the screen

What Claude did

Pictures

The opener:

Blink's opening screen: the words “tell me a story” handwritten in cursive across a white page, with the type box and record button in a band along the bottom.

Still open