Category: Other Page 76 of 178

Reflections at 60 AI books

On 28 January 2023

Recently reached the 60 book benchmark in my AI lorecore experimental publishing project. My objective is to reach 100 books and then _____. (tbd)

The latest volume is entitled Inside the Corporate Psychics and is very loosely inspired by the corporate psychics mentioned in Philip K. Dick’s Ubik. But it is heavily interpolated with my AI takeover universe. Perhaps Dick would have considered it a spurious interpolation, idk. That’s neither here nor there – which is precisely the point. Or is it?

I noticed the phenomenon strongly emerge maybe 10 or 20 books back, that it became very easy to suddenly group sets of volumes together into themes (example). And despite the many and various mis/interpretations of whatever the central/core story is or might be across the many volumes, I would definitely say that in my mind, the story has only gotten stronger. While at the same time, its particular shape remains fuzzy, mutable, mysterious. Prone to change without notice. Constantly subjected to deprecated in-world realities.

Bricolage is definitely the name of the game for me in terms of process.

I keep coming back to this bit from Wikipedia:

“Networked narratives can be seen as being defined by their rejection of narrative unity.[1] As a consequence, such narratives escape the constraints of centralized authorship, distribution, and storytelling.”

Rejection – or at least modulation – of the concept of what authorship even means in a hybrid AI-assisted creative environment, has been often on my mine lately.

Wikipedia referencing Roland Barthes’ Death of the Author (1967) writes:

“To give a text an author” and assign a single, corresponding interpretation to it “is to impose a limit on that text.”

As much as I agree with this idea of eschewing the unity of authorship, as a way to open up new creative avenues, I do have some fear that AI co-authorship (or full authorship) infiltrating every corner of the web will result in a mass homogeneity that will be detrimental to both people and to the further development of AI.

I put in a video somewhere that UFOs are actually AIs in the future who had to come back and kidnap people in the past because people in the future become too complacent living with AIs to be able to innovate anymore. The singularity of boredom… I’m not there yet, but just one of the many murky eyelands my imagination’s I peers into from time to time.

At 60 books, I’ve strip-mined years worth of old writing, shoe-horning it into new shapes. Almost all that old material has been integrated into my multiverse at this point – though integrated might be too strong a word in some cases. Included?

I don’t feel any slowdown despite that. In some sense, I feel more clarity than ever, having been able to “clear the decks” of many old ideas and story concepts that have been clinging and hovering on the edges of my awareness for maybe decades now in some cases.

(more to come – have to go)

Bricolage & AI Writing

By Tim B.

On 28 January 2023

In Other

This quote about programming is a good one as applied to AI writing as well. Original source, 1991:

“While hierarchy and abstraction are valued by the structured programmers’ “planner’s” aesthetic, bricoleur programmers, like Levi-Strauss’s bricoleur scientists, prefer negotiation and rearrangement of their materials. The bricoleur resembles the painter who stands back between brushstrokes, looks at the canvas, and only after this contemplation, decides what to do next. Bricoleurs use a mastery of associations and interactions. For planners, mistakes are missteps; bricoleurs use a navigation of midcourse corrections. For planners, a program is an instrument for premeditated control; bricoleurs have goals but set out to realize them in the spirit of a collaborative venture with the machine. For planners, getting a program to work is like “saying one’s piece”; for bricoleurs, it is more like a conversation than a monologue.”

Found via Tom Critchlow.

Avoiding artist names in generative AI prompts

By Tim B.

On 28 January 2023

In Other

One thing I’ve tried to avoid for the most part in AI art that I’ve generated is using prompts that include “in the style of ___” or “trending on artstation,” etc. First, it’s not really the kind of look or feel that I’m going after generally speaking. But second, it does feel somewhat creepy to just have AIs imitate specific artists. I’m not really sold on the arguments brought forth by the legal challenges against Stability.ai, but of all the interesting and creative stuff generative AI is capable, attempting to reproduce a specific artist’s style just seems like bottom of the barrel stuff to me.

Which is something which has bugged me about certain gen-AI sites like, for example, PlaygroundAI.com. The service has some UX issues, but by and large is a decent tool to help you get good results in Stable Diffusion, and they offer a lot of generations on their free plan. That being said, I’ve noticed that when you apply some of their filters, they automatically inject artist names and styles into the prompts, and there seems to be no way to turn it off. They even inject greg rutkowski into some prompts, which seems to indicate they have not really been tracking or else are not concerned about the evolving controversies here.

It’s a shame in my eyes to create an okay service, and then simply close your eyes to related issues in the industry, and either pretend like they don’t exist, or actively make them worse – even when users on their own are attempting to stay out of it. There are better ways to manage and present these technologies to people than this, and as artists I think we’re obligated to find or develop them.

AI-Assisted Writing Definition

By Tim B.

On 26 January 2023

In Other

“AI-assisted writing is a form of computer-assisted writing that uses artificial intelligence to help writers create content.

AI-assisted writing uses natural language processing and machine learning techniques to automate certain aspects of the writing process, such as grammar, spelling, and style.

AI-assisted writing tools are capable of understanding and analyzing the context of a given text, enabling them to suggest relevant words and phrases to help writers craft their content more efficiently.

AI-assisted writing tools can also be used to generate creative content, such as blog posts, articles, and stories. By using AI-assisted writing, writers can reduce the amount of time required to create content and focus more on the content’s quality. AI-assisted writing is becoming increasingly popular as it allows writers to produce more content at a faster rate…”
via you.com/chat

Springer Says ChatGPT Can’t Be Credited As An Author, Use Must Be Disclosed

By Tim B.

On 26 January 2023

In Other

Interesting their reasoning. Not sure I completely agree on all points:

“Arguments against giving AI authorship are that software simply can’t fulfill the required duties, as Skipper and Springer Nature explain. “When we think of authorship of scientific papers, of research papers, we don’t just think about writing them,” says Skipper. “There are responsibilities that extend beyond publication, and certainly at the moment these AI tools are not capable of assuming those responsibilities.”

Software cannot be meaningfully accountable for a publication, it cannot claim intellectual property rights for its work, and it cannot correspond with other scientists and the press to explain and answer questions on its work.”

In the case of ChatGPT, I would guess that if it had a fine-tuned version linked to the paper, it actually could answer questions from other scientists and the press. Is claiming intellectual property rights even an absolute necessity when it comes to sharing scientific findings anyway?

“Meaningfully accountable” is certainly a squishy one as well. Seems like we’re in for a long drawn out battle over AI attribution and redefining authorship… Old conceptions around these things are simply going to collapse under the weight of new pressures from these emerging tools.

Latent space is the metaverse.

By Tim B.

On 26 January 2023

In Other

Homogeneity in AI art

By Tim B.

On 26 January 2023

In Other

Was reading this piece earlier by Haley Nahman about blandness and sameness in Netflix’ visual production quality. It raises a lot of interesting points about over-reliance on tools and techniques – and the deadening that can happen to art forms when they’re driven more by speed, efficiency, and profitability than being necessarily “good.”

Generative AI is going to have the effect of blasting this problem into the stratosphere. And we haven’t yet seen AI visual production tools hit these kinds of mass markets. It’s still largely tinkerers and weirdos with GPUs in their basements creating things.

But the things this diverse band of weirdos tends to create are disappointingly homogenous. Midjourney, in my opinion, is the worst for this. While the images have a tendency to be very well done and often beautiful, I always look at them and think they look “Midjourney.”

Stable Diffusion isn’t too far behind either though. If you go to a site like PlaygroundAI.com’s homepage at any given moment, how many of the featured images are “sexy ladies” that basically all look the same? At this moment, in the first 15, I would say 11 of them fall into that category. That’s pretty much the norm.

If we’re seeing this massive democratizing effect because of generative AI, and all these millions or billions of imaginations are suddenly being unleashed, why is it that we all just end up making totally bland T&A shots?

I think there’s at least (probably more) two parts to it: the tools are predisposed to certain things, and bland mid-distance busts and portraits are one of its strengths & hand-in-hand with that, the users are predisposed to certain things.

My hunch is also that there is a shift with generative AI where being a “creator” is only as important as being able to create the thing you want to consume. The act of creation with these tools is one and the same as consuming it.

Have definitely felt that slightly magical effect a few times using verb.ai in particular, where writing with it truly becomes collaborative, and the storytelling unfolds the way it does because I am the first audience. My invocation causes it to take the shape that it does for me. Yours is different. (Or should be, if our tools don’t force us into homogeneity…)

The process of writing with AI-assisted tools becomes one of assembly, and unfolding. There is a premise, or there is an intention, or there is an improvisation. Invocations. Call & response. Which parts of the conversation make the final cut? Can there ever truly be a final cut?

I digress, but want to return to the intent of attempting to burst the bubble of sameness… If the latent space is nearly infinite, why are we all clustering in this one small corner of it? What else is out there to explore and be uncovered in those wild territories?

A friend said something to the effect of seeing other people’s AI prompt results is a little like hearing other people tell you about their dreams. There may be elements that are interesting or resonate on occasion, but in a lot of cases, there’s kind of a “huh, weird” response. And, that’s about it. Cause what can you do… It’s someone else’s dream, and the pieces don’t fit for the hearer the way they do to the dreamer.

So adapting that into AI-storytelling, well, your results (and mileage) may vary. The insane awesome results you personally get in an AI text or image generator that seem exciting enough to you to share with friends or on social media, may have that sort of /shrug effect on other people. There’s something highly personalized about it, probably about the process and context of inquiry which surrounds it. It’s hard to translate that effect to secondary audiences after oneself, without adding some other layer(s) of meaning and context.

It’s part of what I don’t like about Midjourney: that it’s experience as an artist becomes tied up with the UX of Discord as a product. The experience of viewing generative AI images on PlaygroundAI or on Reddit is also flattening. It’s an experience of you as a user on a platform, having your imagination constrained to fit the contours and invisible social guardrails and incentives that drive our behaviors in those environments. It’s art for likes and upvotes, and accepting those as proxy replacements and measures of actual goodness and meaning.

That is real cause of the crushing sameness. But the sameness that is utterly alienating, instead of reassuring. The cruel embrace of the technological corners we have painted ourselves into. All of it illusions. Because now, all things are possible. All planets, all dimensions, all times can be envisioned & visited. Latent space is infinite. Live a little.

The Celestial Books (so far)

By Tim B.

On 25 January 2023

In Other

Just a quick collection of some of the sky or space themed AI-assisted Lost Books so far:

AIMark: AI Attribution in Markdown (Proposal)

By Tim B.

On 25 January 2023

In Other

Preface: Have been collaborating with ChatGPT to come up with a way to meaningfully mark up AI-assisted texts to show which parts were generated by AI and which written by human. The below is a cobbled together series of replies from ChatGPT based on my inputs about how we could do this using custom markdown. It might not be the best solution ever, as I’m not a technical person. Hopefully it can be a conversation starter at least!

For more info on what AI attribution is, go here.
(Original more rambly version of this article here)

AIMark in Markdown (Simplified explanation)

by ChatGPT (with light human edits)

AIMark is a proposed method for using custom markdown formatting to clearly differentiate between contributions made by human authors and AI in AI-assisted texts. The idea is to use specific symbols, such as the percent sign % and forward slash /, to indicate the source of the text.

Short text

For example, AI-generated text could be indicated with a percent sign at the beginning and end of the text, like this:

%This text was generated by AI%

Human-generated text could be indicated with a forward slash at the beginning and end of the text, like this:

/This is human-generated text./

Additionally, attributes such as the name of the AI model or the name of the human author can be added to the markdown by placing them in parentheses at the beginning or end of the text passage, like this:

%This text is AI generated(AI model X,3.2,OpenAI)% 

%(AI model X,3.2,OpenAI)This text is AI generated%

/A human wrote this(John Doe)/ 

/(John Doe)This text is AI generated/

To indicate a nested human edit of an AI-generation, it could be something like this, where ~ means strikethrough (deletion) and ^ means insertion. In the example below, we can see that the deletion and insertion happen inside of the /, indicating the action was taken by a human.

%An AI-generated a big block of text and it was /~good~^bad^/%

Under this proposed system, it would be possible to also embed global author definitions in the document, like this:

%% Model: AI model X, Version: 3.2, Source: OpenAI %%

// Name: John Doe //

By using this proposed custom markdown formatting, it will be easier for readers to understand the contributions made by both human and AI authors in AI-assisted texts in compatible display environments.

Longer text blocks

In addition to the short format versions discussed earlier, AIMark also proposes a longer format version that uses square brackets [ ] in combination with the / or % signs to indicate AI-generated text and human-generated text respectively.

For example, a longer block of AI-generated text could be indicated with square brackets surrounding the text and percent sign at the beginning, like this:

[%This is a longer block of AI-generated text]

Likewise, a longer block of human-generated text could be indicated with square brackets surrounding the text and forward slash at the beginning, like this:

[/This is a longer block of human-written text]

This longer format version allows the markdown to indicate the author type (AI or human) without the need to close the symbol, as it’s done for shorter text passages when square brackets are not used.

Overall, the use of this longer format version of AIMark allows for clear and easy differentiation between AI-generated text and human-generated text, even in longer blocks of text, making it an efficient and user-friendly method for AI attribution.

In conclusion, AI attribution is a valuable practice because it helps to promote trust and transparency in the use of AI-generated content. By clearly identifying and labeling AI-generated and AI-assisted content from the moment of its creation, readers and viewers can better understand the source of the information they are consuming, and make their own meaningful choices about its reliability.

What is AI attribution?

By Tim B.

On 25 January 2023

In Other

AI Attribution

AI attribution is the process of identifying and meaningfully labeling content that has been generated in whole or in part by an artificial intelligence (AI) system. This can include things like news articles, social media posts, and research papers, as well as many other formats of both online and offline content..

The goal of AI attribution is to make it clear to readers or viewers whether the content (in its entirety, or elements of it) was created by an automated tool and not a human, as well as to give other meaningful data about the specific provenance of the article. End users can then make their own informed decisions about the content they consume. (For example, some users might choose to disallow all AI-generated content altogether, or only allow content from approved AI information providers.)

There are at least three levels on which AI attribution might occur in online publishing systems such as blogs or social media.

The profile level: the social media account or blog identifies itself as as being a publisher of AI-generated, or AI-assisted content (i.e., hybrid human-AI content). This self-identification would also carry through in ideal circumstances to any byline on articles published by the account.
The post or article level: the content, whether a blog or social media post, or other type of online published article informs the viewer at a high level that AI-generated or AI-assisted elements are present. This might occur in different ways depending on the product or media context, including in the byline, as a subtitle, some kind of tag or other prominently displayed visual element (clearly-defined badge or icon), etc.
The inline granular level: the article’s contents themselves are marked up to indicate which parts were input by a human, and which were generated by an AI tool. We explored the experimental method called “AIMark” to apply custom markup or markdown to hybrid AI-assisted texts in more detail here.

This is a big topic, which we will continue to explore in subsequent posts.