Took 5 seconds to generate this in Chat. Something’s different and I can’t wait to get ahold of it in the API.

So apparently you can have a machine review all the generations for quality instead of doing it yourself. 😆 🤯

Orwell feared a future where machines churn out soulless fiction, but tools like Subtxt prove AI can enhance creativity rather than replace it. By guiding writers through their story’s core, Subtxt ensures authors stay in control, crafting narratives rich with meaning—not just hitting “generate.”

🎬 Loglines often capture just the plot, but they can reveal so much more! Our latest post goes Beyond the Logline: Why Your Story May Have More Depth than a Single Line. 🌟

Curious about the latest in Subtxt? Check out the October 2024 walkthrough where we explore all the recent updates and features designed to supercharge your storytelling. Catch up here: https://narrativefirst.com/recordings/255 ✨

Drafting isn’t just about getting words down—it’s about uncovering the heart of your story. At Subtxt, we share the belief of many published authors: rewriting is where the magic happens. Reshape your writing process with a little help from AI: Read more ✨

The Searchers is now a part of Subtxt’s Narrative Matrix. 🐴🤠 subtxt.app

The o1 difference:

“When I started with the Subtxt app a little over a month ago I mentioned here that I was having problems constructing the storyform for my somewhat challenging concept…”

“I’ve been too busy to work on it in the meantime, and just got back to it, since when the AI has been upgraded. And I’m blown away. The depth of its insights in navigating these problems… Well, they’re not problems anymore. Complexities maybe, but complexities that are adding to the richness of the story. I can’t say how impressed I am. I’d been a little discouraged, and now my mind is racing.”

New post on the process here at Subtxt: Training off the Subtext not the Storytelling: The Journey Towards AI-Enhanced Narrative Development

Uploaded files to OpenAI are now coming back with .txt referenced files in the annotations…Tuesday is going to be so much fun! 🤩

Gleefully pushing to production on a Friday night. With o1, Muse rewrote the entire plot of the Kenobi series giving it a meaningful progression of events that addressed many of the story holes and inconsistencies. Blog post next week, Discord here if you can’t wait: discord.gg/8YV6RsCJ

Subtxt with Muse fixes Disney+. Details tomorrow.

New o1 models in Muse pointed the path forward to fixing the Kenobi series. 🤣 Blog post as soon as I can get it out!

For storytelling, the difference between o1-mini and o1-preview is like the difference between Paramount+ and A24. 😊 From Popcorn Flicks to Profound Narratives: How Muse Finds the Heart of Your Story

just discovered the perfect analogy:

• storyform reasoned by o1-mini = FX
• storyform reasoned by o1-preview = A24

same story, completely different thinking

did some more testing this morning. definitely feels like we’ve jumped timelines. it’s weird. o1-mini reasons storyforms that your typical writer would pick. o1-preview captures all the nuance and subtlety of a great author. find myself just staring at the screen in awe.

in-depth blog post on our first test with o1: Subtxt and System 2 Thinking: The Power of Precision with the o1 Model

First test of o1 without any tuning up? story idea to complete thematic Storyform in 30 seconds (storyform is a holistic representation of narrative conflict across 75 datapoints). Completely reasoned. 100% accurate. Insane implications from screenplays to gaming.

pricing on the new o1 models is not unreasonable if you’ve ever tried agentic/chain-of-thought reasoning with previous models. the repeated calls of the latter make the former quite attractive.

This new series of AI models can reason through complex tasks and solve harder problems than previous models in science, coding, and math.

They forgot to add “story” too. 😊

This new model has huge implications for Storyforming.

something changed significantly with gpt-4o in the last couple of hours. the rote “Absolutely!” and “You got it!” are gone (thankfully) and the rest of the response sounds like a completely new voice.

I know endpoints were down around 6pm wonder if there have been some changes.

Beautiful day for coding. Brand new Max Richter on repeat: In A Landscape.

I forgot Y tu mama tambien had hand-off Main Characters. It’s rare, and this film does it right.

New landing page drop. 😊 subtxt.app

Degraded performance in the gpt-4o-mini model right in the middle of demo/class!! When it works it’s 100% pure magic, when it doesn’t, oh man does it fail spectacularly!

Creating characters and setting them loose within your story is a mere question away with the new Players functionality just added into Subtxt Muse. Watch them behave with integrity to the underlying themes of the narrative while maintaining their core motivations.

What it’s like to use Muse to expand a generic 5-page sequence of scenes into a fully formed and thematically consistent narrative: From Scene to Story: How a 5-page Screenplay Became My Narrative Blueprint.

Used Muse to analyze a screenplay I wrote a decade ago. Pretty much nailed everything that was wrong with it from a thematic standpoint. 😅 Just uploaded it and we started breaking it down together. Will post a follow-up later.

And this is too insane to not share right away, but I went through and took the narrative intent of First Blood and gave it a slight tweak, and Muse never buckled ONCE as it helped me form a new thematically consistent narrative. youtu.be/bsdsWSIBN…

Excited to share a Welcome Tour of the new Subtxt Muse - a way to develop a story that perfectly balances thematic structure with chaotic inspiration!

The most important line from today’s OpenAI blog on how Voice Engine works:

It’s important that people around the world understand where this technology is headed, whether we ultimately deploy it widely ourselves or not.

OpenAI held a Build Hour this week on Function Calling and hardly anyone talks about it. The presentation was absolutely phenomenal and will likely give you a ton of new ideas for your apps (it did mine).

I love that I can keep Muse up-to-date with the latest movies. Here’s one about Dream Scenario.

Building with AI is so ridiculously fun. Latest addition? Muse can now build new “empty” stories for authors (creating new model on backend). The idea that this is even possible, to just ask in plain language what you want–and have it work? How is this even possible?!

Latest episode of “Fun with Subtxt” is out. Continued my development of my Elf/Christmas Carol inspired story “A Tale of Two Spirits”. Kind of like “Shakespeare in Love”, but about how Dickens came up with the idea for “A Christmas Carol”. Developed with Subtxt (naturally!).

OAI fine-tuning engineer with the win: 🤣🤣

RedditAMA with OpenAI CEO. Good news for those using AI to develop their stories…no matter how spicy. 🌶️

All of us after Monday’s 10am presentation.

One of the fun things about building up an AI based on decades of narrative study, is remembering that you wrote something five years ago and forgot to publish it. 😊 The latest is this review of “Locke”, which was a really great one-man film starring Tom Hardy.

Managed to fold in Subtxt’s ReviewAI tool into Muse over the weekend, and it’s turned out way better than I thought it would. All prepared for some self-reflection…Muse: Your Story’s Best Critic and Coach

OpenAI, teach me LLMs as if I was five:

AI models learn from relationships in information to create something new; they don’t store data like a database.

Repeat as needed.

User feedback is best feedback:

I’m convinced the new Muse in Subtxt Infinite is nearly psychic. All I have to do is feed it my stream of consciousness notes and it turns it into something organized and usable, which I can plunk into the Storyforming engine. It’s magic. When I query it about my MC it’s like it’s picking the thoughts straight out of my head.

Finally saw The Iron Claw. Crazy story, can’t believe it was real. Also can’t believe that a story analysis that typically takes 2-3 hours only took 5 minutes with Muse. 😄

Didn’t take half as long as I thought it would implementing this…Muse speaks! 🔊

I stand corrected. I guess Sandra was innocent!

Subtxt Infinite, my newly released version of my story development AI, just helped answer a complex question for a writer in a clear and succinct manner and in a way that I could never match. This is going to free up so much more time for greater things. 🙏🏼

Pretty sure I just unlocked an entirely new artform - crafting narrative before writing it out (Narrative First??!), it’s almost more enjoyable than watching or reading someone else’s story. And you can constantly change it and re-dream it…😃

Big release day for Subtxt Infinite 🚀

Use Subtxt to Refine (that’s me!) 😄

GPT-4 at GPT-2 prices would change the entire world.

I guess gpt-2-chatbot is better, but all I notice is how much longer the response is…who’s paying for all this?!

Figured out a way to concretely explain the difference between Subtxt and other paradigms of narrative “structure” like Save the Cat! and Hero’s Journey. Putting this at the front of the line for the next video. 🎥

John Scalzi speaks the truth in his one-year anniversary post:

I think the Bluesky technical and cultural schema confuses the famous and/or influencer and/or shitty people who come onto the service to _be _famous, or to influence, or to be shitty for clicks.

I finally got Memory in Chat! 🎉

Turns out there are others who deeply care about great storytelling. ❤️

Posted another short video on narrative theory. Perfect for those looking to write better stories.

Finding the Source of Conflict in your Objective Story Throughline

The new Assistants API is so crazy good that my only concern now is being able to afford what will almost certainly be a huge uptick in usage. Responses are blazingly fast and context held so well that churn is going to be a thing of the past.

GPT-4 Turbo (4.2 or 4.5…) is now teaching me NEW things about a narrative theory I’ve been studying and using for almost three decades now. I’ve never thought of the connection between these two thematic touchpoints of a story until today. New frontiers 🚀

Everyone wastes so much of their time posting about the deficiencies of AI, projecting their fears of irrelevancy onto a technology they know nothing about. Seth Godin is the latest, calling it “dumb.” Anyone who has built an AI app will tell you: it’s much “smarter” than you could possibly imagine.

Just speechless with this one. Mr. Mazzucchelli 🙏🏼

Have to say, this Assistants API from OpenAI is really astounding. Not having to manage conversation threads, and tool calls that just work the first time. This is everything I ever wanted when it comes to delivering an amazing creative writing app like Subtxt. It frees up a lot of memory cycles…

So apparently, publishing YouTube videos is a GREAT way to advertise your product. Who knew?!

OpenAI picked a helluva day to completely blow up the Assistants API. 😮 The new improvements are insane, and basically makes the switch a no-brainer now.

As someone who considers himself technically-inclined, I found it interesting that I could never quite grasp Bluesky’s atproto thingee. This Introduction to Atproto from K-NKSM certainly helps:

Since all AppViews have access to the same data, it’s impossible to gain a competitive advantage by monopolizing data. Therefore, anyone (given they have the money and time) can create a social networking service to compete with Bluesky.

I know we’re all waiting for 1pm Pacific to see what happens, but on a lighter note: I’m surprised more aren’t commenting on the new GPT-4 release this week. I had a conversation with Chat, asking it to help me with a CSV, where I could’ve sworn I was talking to a real person.

Published a short little video on Three Body Problem’s Problem.

For real, the greatest class EVER teaching writing meaningful narratives just now played out. So so very energized now! 🙏 Interactivity is EVERYTHING.

Hahaha 😂 404 page sung by AGI

This is absolutely incredible. Feel like I’m living in a sci-fi movie. Feel the AGI

Finally. Something new from OpenAI. Just switched out the models for Subtxt and the inference speed difference is insane! Won’t know about the reasoning until after a few days, but already happy to see some improvements.

Very excited to reveal the first ever VIDEOGAME storyform into Subtxt! After my video yesterday, a super-smart writer sent me his take on the narrative thematics of God of War 2018 - so here you go. 🪓 God of War (2018) Storyform

AI Storytelling and Video Games - new video on the Subtxt YouTube channel, all about modeling the meaning of a story and what it means for the future of gameplay.

Kindof sorta love Sugar on AppleTV.

Getting used to that full-frame action, covering how to directly input your story’s subtext into Subtxt.

Writing stories is how we make sense of our lives. Off-loading that process to an LLM without an author at the helm robs narrative of its true purpose. Subtxt aims to preserve the author’s voice and agency in the manifestation of a work of art.

Cool. Thanks to YouTube livestream technology, I just got to ride Star Tours and see the new Ashoka scene that came out today.

And I didn’t have to spend $500 to see it. 🤫

Thanks to Suno, Subtxt now has a theme song. 😄 app.suno.ai/song/8862…

Seven scariest words in the world: “A new version of Descript is ready!” 😱

Just unlocked a great use case and time-saver for AI. Spent the last hour or so personally writing emails to those who expressed interest in Six-Figure AI App. Took all of those, and asked Chat to draft the landing page based on what I wrote. SO different than if I had started with AI first.

Easily, the GREATEST class ever teaching this material and going over the Storyform. We had so much fun building up all the intricate thematic details of my skeleton sheriff idea. Lots of laughs and great collaboration. 🙏

Now that Flare can display Laravel 11’s context I’ll definitely be taking advantage of this new feature. Perfect for tracking evals on agentic requests. 😊

the special thing that Context does is make all of this information available to any job dispatched in the request where context was set.

Just added a mind-blowingly awesome feature into Subtxt. You could always Develop Storypoints with Muse, but you kinda had to know what to ask. With the new “Review” feature, it’s just one click – and the answers so far are SPOT-ON! This will really make a difference. 🎉🚀

Very interesting. OpenAI’s former Developer Relations Lead Logan Kirkpatrick is now at Google. Almost makes me interested in seeing what they have as Logan was/is great.

Still would love to know why he left though.

Got my first block on Bluesky! My crime? Letting a game developer know they had nothing to worry about w/ AI, and that lots of really smart people are working very hard to make sure their artistic intent was at the center of new tech. Can’t say much more than that now, and can say even less on 🦋😆

Built a little step-by-step progress bar to help with onboarding new users. Hoping to make it easier to understand just what Subtxt is capable of from the very start.

Now that it makes sense to make this live again (since you don’t need a ChatGPT account to use it), here’s the link to Subtxt Muse, Jr. to get an introduction into what it would be like developing a story with Muse.

Chat is now free to use without an account, which finally makes offering a CustomGPT to advertise Muse’s capabilities a sensible choice:

Starting today, you can use ChatGPT instantly, without needing to sign-up. We’re rolling this out gradually, with the aim to make AI accessible to anyone curious about its capabilities.

I am completely fascinated with people who livestream from Disneyland for hours at a time day after day. I can’t stop watching them. 🏰

I miss the days when OpenAI actually released something. Election-year buzzkill makes this spring season feel like pandemic 2.0.

So many great ideas in this LangChain newsletter. Right now, Muse is all about remembering, but the idea of reflecting on the conversation and altering its system instructions is quite intriguing.

Subtxt, with the killer explanation as to how both Hale and Mollie stood on the same side of the dramatic argument - both essentially arguing for “tolerating harsh conditions”, with the latter putting up with self-delusion and the former self-deluding in the name of success. Bravo.

Jotted down some notes on Tuesday’s class, where we uploaded the first Act of “The Holdovers” into Muse and asked thematic questions based on the film’s Storyform. Muse responded with ways to improve the text! 😄 (no, we don’t really think the screenplay needed improving)

I always say this after teaching this one particular class to cohort students, but boy, do I LOVE doing this. So much fun. 🙏

Man, if I had known Vue 3 was this much fun to work with, I would’ve switched over a long time ago. 🏎️ 💨

At Sequoua’s recent AI Ascent event, LLM architect Andrej Karpathy says to build more ramps, as well as projects. Well, I’ve done the latter and now, just pulled the trigger on the former: Introducing Six-Figure AI App. 🚀

The newly rebranded Subtxt University! For all your narrative needs (really, just a way to bring all the courses/recordings/community together under one roof for Narrative First).

LOL, was teaching a class on character arcs just now and someone asked if you could “retro-actively” go over narrative structure with a finished work. I dropped in the first Act of “The Holdovers” into Muse, not knowing if it would really work, and it actually had constructive feedback on it! 😄

Inspired by the new Sora videos, I went ahead and showed how you could use Muse to take the “Balloon Man” idea from boutique short to feature film…🎈 The Resurgence of Visual Storytelling: Exploring New Horizons with Sora and Muse

Not a coincidence at all that this blog post from OpenAI about Sora First Impressions releases at day one of their journey to Hollywood this week. 😉 The balloon head guy is probably the closest thing to being able to string together a cohesive narrative…

Ilia Malinin landing SIX QUADS in a single performance is one of the most awe-inspiring things I have ever seen. Such a bright shining light and celebration of humanity. Bravo!

Found a way for me to more easily train Muse when it comes to the relationships between thematic items within a complete narrative. Excited to correct all its mistakes all in the service of a better story. 🎉🚀

Just used my own creation Muse to frame the plot of an example story I’m building in the SubtxtU cohort (ghost town with a skeleton sheriff!). So much fun, and looks like the memory features I have installed are firing on all cylinders as it never once lost context.

Spent the evening tumbling down the YouTube rabbit hole of red heifers, third temples, and Texas farmers and now there’s no way I’m going anywhere near the totality of this solar eclipse. 🌀😵‍💫

Rebuilding Subtxt’s frontend from the ground-up to take advantage of all the latest tech and to get ready for the next model releases from OpenAI. Amazed at how much fun all of this is, but man–there’s a lot to update: over six years of development! 😅

This month’s moment of Mazzucchelli zen:

I choose green. “To war then!” House of the Dragon 🐉

Gotta say, rebuilding the entire frontend of Subtxt using ChatGPT to update components to latest and greatest best practices is a breeze (and that Vue3 Composition API, yeash). Super fun, learning and working things out. Could possibly have this all done before May, which is crazy.

It makes me so happy that Muse is constantly updated and ever-evolving to include more examples than just Star Wars and The Wizard of Oz! 😄

Kinda feel like this is going to be my last relaxing weekend for a couple months given the rumors of next week’s release…🤖🚀

Asked Chat about “abiding centerlessness” and not sure if I can give up my favorite hobby:

The paradox comes into play when someone loves the practice of meditation itself. The pursuit of abiding centerlessness requires letting go of all attachments, including the attachment to meditation as a practice or hobby. If one reaches the ultimate goal of meditation, the need or desire to meditate as a distinct activity might dissolve, as one would theoretically be living in a constant state of meditative awareness. This could be perceived as the “death” of meditation as a hobby because it no longer occupies the same role in one’s life—it’s no longer an activity you do, but rather a state you embody at all times.

Finally figured out that Kacey Musgrave’s “To Good To Be True” is just a ripoff of “Breathe 2AM”, so now I can finally sleep. I knew I heard it somewhere before!

Nothing better than a great new group of cohort students. 🎓🙏🏼

The Storyform for Academy Award-winning Screenplay Anatomy of a Fall is now up in Subtxt.

Most compelling observation is that usually, I need to go in and adjust some of Muse’s responses. Not sure if this is an indication of how great the screenplay was, but I didn’t have to change a thing. 😬

AI is amazing! Building out a new feature, it generates a character called “Sir Hoots-A-Lot”. I say it out loud – and now Siri is playing “Baby Got Back” from “Sir Mix-A-Lot”. The future is magical. ✨😄

Anatomy of a Fall was great if for no other reason than it explores the difference between what we perceive to be reality, and what we know to be reality.

Was going to respond to this post about 1M context windows and how now you can “put your whole book in”, but as I scrolled down I noticed every single response was just a bot. I mean, people can tell this now right?

I don’t know if it’s just because I work with LLMs all the time, but 90% of the posts on Twitter/X are generated.

So what’s the point of interacting with them?

With Bluesky, I still get the feeling that real people are here, which is a nice feeling.

Wanting to know more about the B&W scenes in Zone of Interest, I stumbled across this magnificent article about the film: Viewing the Ob-scene: On Jonathan Glazer’s “The Zone of Interest”

These scenes are shot without lighting and in infrared, so the girl’s body appears as white against the sheer black of her surroundings. It’s a startling, uncanny reversal of the daytime scenes, and the only act of compassion we witness in the film, which must necessarily take place out of sight.

Pretty stoked, after a rather quiet winter, enrollment in the Spring edition of my Maven cohort is heating up! 🔥 The Future of AI and Storytelling

Spending the next quarter upgrading Subtxt to the latest and greatest version of Vue. This article from Tighten on State Management with Vue/Pinia couldn’t have been more perfectly timed.

Zone of Interest was quite surprising to say the least.

Published the Storyform for Poor Things into Subtxt. Required a fresh build from scratch as it is one completely unique narrative. 😅 I now have the ultimate example for what a Dark Comedy Female Mental Sex narrative looks like!

Still love movies. 2024 Oscars 🎬🎞️

Muse finally gets it. You don’t know how difficult it is to “un-train” an LLM from all the nonsense it learned out in the publicly-available internet.

New frontend on the way. 😃

It’s official, heading back to Vue for frontend. Everything comes full circle. ⃝

As Twitter becomes more and more irrelevant, seems like it was about time to move my company’s social media over to Bluesky as well. Can’t resist the inevitable. 🔮 Narrative First on Bluesky

Uploaded the Storyform for American Fiction into Subtxt (really great film if you haven’t seen it yet). First time I was able to use the new “hand-off” Throughlines in the Complete Story section. Works great if you have a more complex narrative where the perspectives are handed-off.

American Fiction was quite delightful. 📚

I would not be surprised if Elon shut down Twitter so no one could talk about the latest OpenAI post.

Rolled out Infinite Memory for Subtxt Muse yesterday. Even added the ability to “mute” specific messages so writers can tailor their brainstorming sessions to their liking.

When I was in 8th grade, I would put hidden notes in my giant history projects to make sure my teacher was really reading the whole thing. At the time, Mr. Estes did not appreciate it, and it looks like Claude–and future AGI models–won’t like it either. 😄

Who turned EVERYONE on today?!!?! 🥳🥂🎉🚀

I invite everyone to watch the latest drops from #Sora with this Terminator ambience. The surreal nightmare of the musuem and alien in the city take on a whole new level of existential fear.

Dark mode on the OpenAI playground - finally! MY EYES! 😭

If this is your definition of subtext:

“Subtext in writing is the underlying meaning or message conveyed by a text that is not explicitly stated by the characters or shown in the narrative. It’s the layer of meaning that lives beneath the surface dialogue, action, or plot, offering depth and complexity to a story. Subtext can be created through dialogue, character actions, symbols, and the overall context of the story, allowing readers or viewers to infer deeper meanings and emotions that aren’t directly expressed. It enriches the audience’s experience by engaging their intuition and imagination, inviting them to read between the lines to uncover the true essence of the narrative. Subtext can reveal character motives, thematic elements, and emotional undertones, making the storytelling more nuanced and resonant.”

Then yes, you absolutely can train an AI on it. I already did and it’s called Subtxt. The app is trained on the underlying meaning of a story, not the text itself, and is designed primarily to help authors discover and refine the message of their work.

Bluesky hit an inflection point for me this week. All of a sudden it’s very very much alive and I can finally put Twitter/X to rest permanently.

Had to watch 21 Pilot’s Overcompensate twice in a row it was that good. 🥁

It’s OK to be hopeful about Horizon: An American Saga, right? 🐴🤠

Nothing more important than supporting your very first client that showed up ten years ago and took a chance on you when you had no history to back up what you had going on. 🙏

OpenAI simplifies API pricing page, listing per/1M tokens instead of per/1K tokens. Two things: a) this is obviously a precursor to larger context windows. b) I can’t believe how many millions of tokens run through Subtxt a day. 🤯

Feeling pretty good about this rebrand of my cohort: The Future of AI and Storytelling. Reflects a lot of recent developments and makes it clear that it’s just not about writing a screenplay.

Wonderful! You can check out all the hashtag posts from a single user. I love Bluesky so much 😊 I can see myself using this all the time now.

Bluesky now has mute words and I’m in heaven!!! Muted X and now I can mute the last semblance of insanity of my phone forever.

This is the most incredible thing I’ve ever seen. And it’s not hyperbole or hype. Muse kept the entire conversation in context for over 30 pages of story development. Do you know what that means? Introducing Subtxt with Muse: Infinite Memory

I can’t believe what I just made. Blog post coming.

So I guess the theme for 2024 is businesses who want to hook into Subtxt for their own storytelling applications…didn’t expect that, but OK. I can do that.

There should be a rule that people can’t say “orders of magnitude” when referring to AI.

This month’s killer full-page shot:

When you’ve been fine-tuning your story development chatbot for 15 years, and didn’t even know it! LOVE that writers can just talk with Muse to find something I wrote years ago to expand their knowledge of narrative structure.

Recorded a big video going over Subtxt’s Narrative Agents (Auto-GPT for story development). Though long, I like this new format of showing everyone what it can do—because even long-time users aren’t even aware of is ultimate power!

That’s a wrap on another round of the AI-Powered Storytelling cohort workshops! Always a wonderful experience teaching how to best leverage AI to better understand the context of the stories we tell. 🎉🙏🏼

James Briggs on Semantic Routing and Pinecone. One of the more frustrating aspects of building an AI app are the tool calls from OpenAI. So much potential, yet they don’t quite work as expected (especially with older models). SR looks much better. Thinking of rolling my own for better control.

Just spent the last hour or two animating CSS with natural language. Took my animation skills and just “talked it out” to GPT-4 to get the timing right. Really fun.

Made some improvements to Subtxt’s image generation. Whole different ballgame than using DALL-E in ChatGPT, much more freedom, but means you’ve really got to be specific (with more power, comes more responsibility…) 🎨

Emerged from meditation with the thought that I need to delete Twitter and LinkedIn. Did so, then found Ben Werdmuller explaining so eloquently what I intuitively thought:

Posting in a space I control isn’t just about the principle anymore. It’s a kind of self-preservation. I want to preserve my attention and my autonomy.

Random Twitter voice asks:

projection seems like a basic fact of psychology but i still don’t understand how it works, why it happens

To which I answer:

Because we can’t step outside of ourselves to gain the objectivity needed to make meaning of our experience. It’s why stories exist.

16-year old animator frets about the future of animation after seeing Sora. Animator several decades ahead of him offers advice.

Eyvind Earle drone photography is my new jam! 🤩 GanWeaving

With Subtxt, you can ensure those five seconds are not wasted.

When a new user starts subscribing during the .19% of downtime from OpenAI over the past 90 days. 😵‍💫

Really fantastic and accessible explanation of how Sora works

Previous text-to-video approaches required images and videos used in training to all be the same size, which required significant pre-processing to cut videos down to size. But because Sora trains on “patches” instead of the full frame of the video, it can gobble up any video or image without requiring it to be cut down.

Gilfoyle. Always giving me crap, even when it’s clearly a typo.

Guess what I’m reading before bedtime tonight?! Video Generation Models as World Simulators Might as well understand how this all works. 🚀

Pretty convinced that 10-min. video recorded feedback of student’s submissions are the way to go. Just showing how to develop a story in conjuction with an AI (Subtxt), opens up so many new understandings for the writer. Might have to rethink my whole approach!

Instead of writing about the experience of deep-diving into the narrative of Oppenheimer with my AI Assistant, Muse, I went ahead and asked Muse to write about it for me: Unraveling the Oppenheimer Epic with Subtxt Muse

Asked once again about the difference between ChatGPT and Subtxt (“Why can’t I just use the former?”), I went ahead and published an entire page in the Guide about it: The Narrative Expertise Layer.

In short, there is a “narrative expertise layer” that works as intermediary between the writer’s intent (question) and the response they receive from the AI (OpenAI’s API services). Many believe they can replicate the same experience in Chat with uploaded files and prompting…many find that their wasted hours spent trying to remake Subtxt could’ve been better spent writing 🙂

I KNEW Google’s 1M token claim was because OpenAI was about to drop something incredible. I did not expect this: Sora: Text-to-video 🥳

Pinecone’s new Serverless Vector Database is more than just a cheaper AWS implementation. Some nice thoughtful additions: the Freshness layer is something I KNOW my users will love, new RAG considerations are also welcomed. Nice to have someone focusing on practical and usable functionalities.

The recording of last night’s analysis of Oppenheimer, hosted by Dramatica theory co-creator Chris Huntley, is in the Narrative First Workshop (free for all): Oppenheimer Narrative Analysis.

Uploaded the narrative structure of Oppenheimer into Subtxt. Interestingly enough: same Pivotal Elements as Inception only this plot is less heist and more a “memory” play. 🎞️

The Dramatica Users Group analysis for March 24 is going to be Teenage Ninja Mutant Turtles: Mutant Mayhem. The director was a student of mine at CalArts, so you can be guaranteed the story is going to be great! 😄

Ebon-Moss as the Thing in Fantastic Four is such perfect casting. I’d actually go see it just for him (phenomenal as Richie in The Bear).

There is just nothing quite like developing a story with both imagery and text at the same time, bouncing back and forth between the two, and getting inspired both–knowing full well that this isn’t the final product. The playground of generative AI.

From Paul Graham:

Historically, letting technology eliminate their jobs has been a sacrifice people have made for their kids’ sakes. Not intentionally, for the most part, but their kids ended up with the new jobs created as a result. No one weaves now, and that’s fine.

I LOVE this. A bit concerning when some of that generation don’t want all the benefits of the tech, but I do like seeing this as a potential sacrifice on our parts for a better tomorrow.

Probably not a good sign that I only now just got access to the @ GPTs. Means I won’t get memory for weeks 😭

A trip of happiness on Valentine’s Eve. 💘

Oh, you know. Just asking a super-intelligent story analyst to help me better understand the narrative behind Oppenheimer. 🤓

Descript’s constant updating and changing things is a constant reminder to me to not do the same thing for my own users (but then, I go ahead and do it anyways). 😎

Guess I picked a good week to work on extending Subtxt’s memory. OpenAI rolls out memory and personalization.

I really like Discord’s design aesthetic, but man, most people just use it as their own personal help desk–as if they are completely unconscious of the community’s experience. Gotta come up with something else.

Everyone is rightfully excited about uploading drafts into Subtxt and getting feedback. But what makes me most happy is how nuanced the response is: instead of just blowing smoke up, it gives you options of ways YOU could improve it. Unexpected tbh, but inspiring.

Uploaded the textbook-perfect tragedy, Griselda into Subtxt. Will be using this one for a long-time to come as its easily digestible and shows the forces at work in a Changed/Failure narrative so blatantly. Wonderful example of an MC Antagonist as well.

I work really hard to make Subtxt great, so seeing this random tweet out in the wild (and in Spanish!), will never not be cool. (Translation: “10 AI tools that I really use.”)

One of the many surprises of building with LLMs is how it opens up collaboration across cultures. Working with Subtxt and Muse across different languages is as simple as just asking.

How is this even possible?!!

Guess everyone really enjoys uploading their entire manuscripts for analysis. 😅

Received an email from a writer wondering who the Antagonist of Good Will Hunting but couldn’t remember if it was Stellan or Will. Then I remembered I built a super intelligent AI trained on 30 years of answering this question and just asked it instead. 😇

Big morning to wake up to.🤩 Subtxt could already do what it does today, but something about selecting a paper icon and uploading a file makes ALL the difference to everyone. Who knew UX was a thing?

First report back from user uploading entire screenplay into Muse (“This is insane!”).

Chris Surdak on AI:

LLMs are context-less. #promptengineering is nothing but making sure that uses provide enough context in their questions so that the LLMs actually knows what’s what.

And that’s how Subtxt works.

One of the first things I teach when it comes to narrative is that context is meaning. Or, from another angle: life is meaningless without context. Same with stories.

If you want a meaningful story you need to set the context.

And that’s what Subtxt does in the background with its “prompts”: converts author’s intent into context.

(And most importantly: holds that context)

Trying to explain to writers in the workshop yesterday how a narrative works when the Main Character’s “change” leads to tragedy, and then Griselda just goes and does it. 😇

Fun little super hack, just tell ChatG the top 3 things you want to do today, and receive back a detailed itinerary on how to accomplish all those tasks (if coding, you even get libraries, step by step, etc).

Doing my part in the battle against the machines and bad storytelling. Here’s ChatGPT’s generic “story”:

vs. a narrative analysis from the all-knowing Subtxt Muse. 😎

Great Storytelling: 1 Generic Weights Story: 0

OpenAI adds metadata to images now (and by extension, vis dev created in Subtxt has the same).

Everybody can sign up for Bluesky now. No invite needed. Everyone who is anyone: make the switch. SO MUCH nicer over here.

The headline I’ve been working for the past seven years.

Oppenheimer soundtrack hits differently when it’s raining outside. ⛆

Many requested a weekend version of my AI-Powered Storytelling Workshop, as they found it difficult to show up during the week. This Saturday, February 10th, is the start of that 6-week workshop. If you want insight into all the narrative theory behind Subtxt,,I’d love to see you there. 😃

Blog post on Narrative First about this weekend’s AI-storytelling breakthrough.

MAJOR key unlocked today! Took the first Act of Saltburn and fed it into Subtxt to have it evaluate the narrative structure. It correctly identified the Four Throughlines of conflict AND where their sources of conflict are coming from BUT it was wrong about the Main Character and Obstacle Character–and that’s because it hadn’t read the whole thing!! 😄

I can’t even believe this is possible and I built the whole thing!

Everyone said this ability was years away - it’s not.

Welcome to February 2024. 🚀

If this weekend is any indication of the next eight months, I am going to be VERY VERY busy. 😅

From Vox, Everybody has to self-promote. No one wants to:

You’re getting worse at your art, but you’re becoming a great marketer for a product which is less and less good.

I totally get this (Subtxt is my art)–in that, I always feel like there’s so much I could be doing to make the app better instead of marketing or promoting it, and that time away from development isn’t mixing the needle

BUT

i’ve always found that in the marketing and in the promoting, and in the teaching that I end up finding ways or thinking of ideas that I never would have if I was just self-consumed with the creation of the art.

If you make the promotion fun, it’s all part of the same creative process.

This is unreal. In the same way that blogging about story for 15 years set me up perfectly for retrieval augmented generation (RAG) this year, building Subtxt for the past six years put me in the perfect place to take advantage of OpenAI’s tool calls this year. 🚀

Just unlocked a whole ‘nother level of Muse here. Now, instead of random “motivations” that are based on default training, Muse will suggest Motivations from Subtxt that line-up and sync up with other Throughlines throughout the story (melding character with plot!).

Whoever plugged everybody back in, thanks! More emails/chats/new customers/comments today than practically all of January. 🥳

Oh man! This new interaction offers up so much more detail and variance of experience for storytellers using Subtxt to develop their stories. It might not look like much, but when they ask for two conflicting resources, I’m using natural language to change the scope of what is requested in a way that just doesn’t dump out and say “I don’t know.”

The implications of this are mind-boggling. 🤯

Starting tracking individual API calls to OpenAI with different keys for different features. WAY COOL. 🚀

Added some clarity to the Subtxt Guide that I think is quite nice: Narrative Agents are for quickly building up the scaffolding of your story, Muse is for deep analysis, reflection, and review.

Love this quote today from Louie Bacaj (SmallBets.com:

The risk is, and always was, that you might waste your time making something that helps nobody & that nobody wants.

But somehow, they convinced us in the last decade that the risk is: it might not scale.

I spent so much time in the beginning worrying about what would happen when Subtxt accelerates. And the idea that VCs were only interested in products that had massive scale behind them only added fuel to that fire.

But guess what? You don’t have to worry about that in the beginning AT ALL. And you might even be building a product that doesn’t scale to ridiculous proportions.

And that’s OK too.

You can just quietly build something that users love and watch their enthusiasm and joy scale instead. ❤️

Added a quick little section to the Guide explaining how Subtxt’s AI tools work (and how they differ from Subtxt Muse, the story development chatbot). 🤓

Woke up to a high-pitched whine in the air. Either the NTIs have finally decided to show themselves OR there’s another prison break at my local maximum security prison. 🛸🚔

When you finally roll out something you’ve been working super hard on for two days, then find out Safari doesn’t like lookbehind assertions in regex but you don’t have much hair to pull out anymore. Still amazes me after a year how quickly I can fix this kind of thing thanks to Chat.

Published a quick little video tut on Setting the Focus of Conflict in a narrative whilst in Subtxt (not all narratives are created equal!). I get this question quite a bit, so I hope it helps!

Man, I love teaching. Such a phenomenal last two hours and SO GRATEFUL that I found a way to teach writers everywhere (and that there are some who continue to show up!) 🙏

Been waiting for this. Now all my virtual employees can virtually use ChatGPT as their virtual Slack. 🤖😎

Starting soon: (because Laravel is really all you need)

Set the date for Feb’s Deep Narrative Theory Workshop. Films with similar thematic structures: Legally Blonde, Arsenic and Old Lace, and Disney’s Jungle Book (‘67). See you there. 😃

Perfect timing. The guy in charge of creating the framework I’m intending on using to redo the frontend of Subtxt is giving a talk tomorrow on just that: Go ahead. Rewrite your codebase. 😄

I heard JK (my old boss) made a prediction that animation would suffer 90% job loss due to AI. Then I found the video. Animation peeps: don’t base your existential crisis on someone who claims “prompting” is a thing. It’s not.

Rolled out a massive new update to the Subtxt Guide: Illustrating Storybeats. Everything you ever wanted to know about the temporal aspects of a narrative…and then some. 🤓

January was excruciatingly slow, but can already feel it picking up her towards the end. OpenAI leaker:

January was a bit sleepy on the surface, but February will be full steam ahead. As you know, I already hinted in October that February would be wild 🌸

Finally decided to do a group analysis of All the Light We Cannot See by Anthony Doerr. We’re going to do the novel (the series was unwatchable), so if you’re interested in an appreciation of the deep narrative thematics of an amazing work of art, take Feb to read it and we’ll meet in March. 📚

Brand new tutorial on what it means to alter a Storyform halfway through the writing process, and how you can manage it better with Subtxt. This gets asked ALOT, so I hope it makes sense (I took Ex Machina and showed how you could change the underlying meaning).

“If we don’t put tentpole movies in movie theaters, there won’t be movie theaters in the future.”

Hasn’t this already happened?

Director Doug Liman on his Road House remake

Oh ok, Jeremy Allen White. Big deal. So you can scale the fire escape of a NY flat—but what do you know about the stochastic gradient descent and backpropagation algorithms that are fundamental in training AI?

Based on requests from users, I added Dan Harmon’s “Story Circle” technique to the Plotting Agents in Subtxt. Took about 20 minutes as it’s just a fork of Hero’s Journey/Eight Sequence method (whereas Subtxt’s framework has taken over seven years to develop). 😄

The deadline for Converge 2 (OpenAI’s $1m startup fund) is this evening. And while I would usually jump at the chance, there is this whole other world out there where the only backing you need are your customers. The pastures of the quiet 🦄 are lush and green–and open to everyone.☀️

Why I prefer to roll my own RAG:

Lastly, this isn’t to do with cost but there is a LOT more to deal with RAG than simple semantic comparisons. If you notice that the results aren’t satisfactory your only option is to try and improve the document. In comparison to a vector DB you have a massive amount of tools & metrics to work with.

As cool as it is, it doesn’t make sense to offload all the quality-of-life improvements you can get with tweaking into the black box that is the Assistants API (especially if you don’t have millions in VC to back up your experiments).

OK, wow. Yeah. GPT-4 is DEFINITELY not lazy anymore. Jeez. Leave some code for me. 😅

Life is mundane without an OpenAI release–so today is exhilarating! ☀️ New GPT-4 Turbo out today, new 3.5 Turbo which is practically free next week (hopefully it makes tool calls better than current), AND new embedding models. Very interested in seeing the performance increase. 🏎️💨

Finally was able to publish my latest video on “Storyforming Past Lives”. Beautiful film with an amazingly unique narrative structure.

Always amazes me how just a few small tweaks in an AI app makes such a HUGE difference in the quality of output. Haven’t completely given everything over to OpenAI as I like having granular control of what shows up context, so making some tweaks with Pinecone elevates everything to another level.

This month’s comic book post of greatness:

Uploaded the Storyform for Past Lives into Subtxt. This was a unique one unlike any of the others (now 575 of them!).

The score for Saltburn is surprisingly exquisite. Highly recommend. 🎻 🎹 🎶

We’re about to enter the exponential world of Hero’s Journey/Save the Cat! If anything, it’s free advertising for Subtxt. 😊

Of course, the narrative structure of Past Lives is so unique that Subtxt has to build it from scratch–but it will be well worth it.😇

After doing the Storyforms for The Holdovers and Saltburn, I somehow woke up with the Storyform for Past Lives in my head. I saw it months ago, but just now (for reasons) put out all together. This explanation of the premise in Subtxt–just based on thematics sounds like the film.

Uploaded the storyform for the magnificent Saltburn into Subtxt.

Took me about a year, but I’m starting to see the art of building a non-deterministic AI app with large-language models. No different from sculpting, sketching, animating, editing. I think I may have discovered a new passion. 🎨

Bizarre change in my morning social media catch-up: I find way more value in the OpenAI Dev Forums and other private Discord servers than I do on Twitter/X. The latter is usually just MRR porn interspersed with the occasional helpful post, whereas I always learn something with the former.

Something about this just looks like pure art to me. It’s the logic for the Assistants API from OpenAI. I’ve yet to switch over as it doesn’t support streaming yet (and I’ve got enough on my hands switching everything over to the new tools methodology), but love the simplicity of the complexity.

So I guess with Saltburn we’re just back to making great movies again?

Published a new video on finding the unique narrative Storyform for The Holdovers. Wonderful film with–as expected–a great story.

Another reason why running daily evals while serving an AI app is so important. A few tweaks can cut your costs in half without affecting the quality of responses.

Dang it! Users are starting to catch on that basic story generation is defaulting to a classic Western-culture Storyform (e.g., Looper where pursuing things is the problem). Guess it’s time to move for phase 2 where the AI “reads” the synopsis for intent (opens up Linear…)

To be honest, running an AI app is all-encompassing. There really isn’t any time for anything else, especially if you want to run evals daily and make sure it just continues to get better and better. Hardest, most rewarding thing I’ve ever done.

15 years later, still have to show writers and producers that not every Main Character has to “change”. All grow, but only half grow out of their resolve. The other half grows into their resolve. 2-min video montage of famous MCs who don’t change: What Character Arc Really Means

May December: flipping 180s to put you right in the center of it all. 🎨

What it’s like to develop a story visually with Subtxt and Subtxt Muse: Visualizing a Historical Romance Unlike Any Other

As someone with a productive AI app built with LLMs, I certainly don’t agree with this:

RAG is fundamentally flawed approach. human memory doesn’t refer back to original source material and re-infer connections between everything whenever you think

But I do love this response:

That’s because sleep is our daily finetuning job which incorporates the day’s data into our weights

Thinking about Newton’s 3-Body Problem and education this morning. The past 15 years have been quite predictable between me, teacher, and student. With the introduction of Muse (AI storybot) in January 2024 that relationship has become wholly chaotic and unmanageable. It’s also become exponentially more delightful for all involved.

So wait. You’re telling me that the musicians behind Oppenheimer listened to click tracks that several bars before a tempo change would start clicking that change into their headphones while they were still on the old tempo?!! MASTERFUL.

Ok. Attention to detail alert: Bluesky offers dark mode splash screen of the clouds now. So good. 🚨

When image generation goes horribly wrong. 🤣

ChatGPT + Arizona State University. Who else doesn’t want to be left behind?

“ASU recognizes that augmented and artificial intelligence systems are here to stay, and we are optimistic about their ability to become incredible tools that help students to learn, learn more quickly, and understand subjects more thoroughly.”

So funny to me how so many of us are on the same trip. Uplimit’s AI-Powered Learning feature looks exactly like what I’m building for Six-Figure AI. 😄 This course on building apps is pretty close (though I would add Laravel in there for maximum happiness). ☀️

Just the beginning.

The winner of Japan’s most prestigious literary award acknowledged that about “five percent’ of her futuristic novel was penned by ChatGPT, saying that generative AI helped her unlock her potential. 🤖

Literary laureate unashamed about using ChatGPT

AI is just a quicker way to get to what a lot of writers refer to as a “vomit draft.” You can always turn in your first draft if you want to, but you’ll likely be called on it.

Thanks to Laracasts and its AI-Generated Images episode, I was able to get my app to start generating pictures for stories in less than 3 hours. Also, Jeffrey Way is crazy smart on those keys. Learned SO MUCH.

Coming soon to Subtxt…

Just published a big write-up about starting to develop a modern-day “Casablanca” set in Ukraine using Subtxt AI. Builds on the thematic relationships rather than the actual text.

Finally figured out how to run the monthly “Wide World of Narrative” workshops today. REALLY great discussion from everyone, and the new approach to build a story based on what we learned in class is going to make a huge difference I’m sure. What a week! 💫

So apparently this is possible now. With the new breakthrough this week, you can easily draw upon Subtxt’s knowledge of narrative structure to quickly spin up a new story.

My Trinity…BG Music

Did not expect to be so excited about starting to teach again the AI-Powered Storytelling workshop. Great group of students from all the over the world. Not sure what it is about today, but man–super exciting!

Reading through the Pinecone Serverless announcement. Hybrid search is in there, which means I’m definitely going to use it:

Just like the pod-based indexes, Pinecone serverless supports live index updates, metadata filtering, hybrid search, and namespaces to let you have the most control of your data.

Hybrid in Subtxt this past year has made a huge difference in accurate RAG (retrieval augmented generation).

You know I’ll be trying out Pinecone’s new serverless vector database. While this approach certainly saves costs, it also opens up a whole new world of features I was holding back on…

Added a little bit of social proof to Subtxt’s landing page. 😊

I’ve found that the reduction in overhead frees up more conceptual levels of thinking, how it all fits together instead of how it all is

The biggest benefit GPT-4 gives for coding tasks is simply reducing cognitive overhead IMO. It removes instances of “ugh I know how to do this but it’s going to take a bit of thought and 20 lines of code.” Instead I just ask GPT-4 and it gets me 95% of the way there.

Mark Tennenholz

I know Python is the hotness when it comes to AI, but to me–there’s no lower barrier to entry than building with Laravel. New release in Q1 reconfigures middleware.

For those building on the Assistants API: biggest speed improvement: setting up thread ahead of time.

Runs have messages on them and are within a thread. My biggest speed improvement came from splitting part one. Basically when the user opens my bot by clicking on the + symbol, it sends a message to the backend to initalise a thread and have it ready. Then when the user send the message, the run is added to the thread. Thread ends when the user closes the bot

New page in the OpenAI docs about how they built the GPT Builder is the single greatest page on the whole site. Totally eye-opening the way they set up the prompts and instructions. 👀

Devs behind OpenAI’s GPTBuilder did an AMA this week. Discord, as usual, is a nightmare to navigate; luckily screenshots of the events are here in the DevForum.

Biggest takeaway for me:

• Knowledge files are client side and for the most part deconstructible.
• Custom actions are much the APIs of a backend and more defensible.

Yet another reason to build your own app

Well, this is certainly good news 😃

The new spinny thing when ChatGPT loads up its voice assistant is so freaking good.

That moment when your finely-tuned AI-powered storytelling machine just works. ⚡️😇

Google DeepMind-er speeds up his AI timeline predictions. Better than 95% of economically valuable jobs in 5 years.

Overheard: At ycombinator W24 kickoff today, Sam Altman suggested ppl build w/ the mindset GPT-5 and AGI will be achieved “relatively soon”; most GPT-4 limitations will get fixed in GPT-5, per YC founder Richard He.

Expect GPT5 in 2024 and #AGI in 2025…

Is it healthy to keep listening to the “Destroyer of Worlds” track off of Oppenheimer over and over again?

“Tech bros” is such a great RegEx filter. No one who uses that phrase should be taken seriously.

Well, here we go! With the official release of OpenAI’s GPT Store, expect to see more and more “screenwriting” and “storytelling” AI GPTs. There will be an exponential increase in stories written by an LLM trained on a generic and collective understanding of narrative. You thought the last 30 years of Hero’s Journey and Save the Cat! was problematic, just wait till you get a load of GPTs trained on Pixar’s 22 Rules of Storytelling! 🤣

2024 is going to be the year of the LLM “tool_call”. Everywhere I look I see, “hmm that could be a tool call and that could be a tool call”. Non-deterministic guardrails ftw.

Enrollment is picking up for the AI-Powered Storytelling cohort that starts next week! Get a jumpstart on the future of narrative 🏃‍♀️💨

My favorite kind of morning. ☀️

Over-engineering vs under-engineering. Context is everything (in other words, do what works for you if you’re the one doing the work).

Another thought for everyone bemoaning the current state of generative AI art and its “soulless” nature: I invite you to compare the animation of humans in the original “Toy Story” with humans in “Encanto”. We used to say the same things about CG animation in ‘94–we don’t say anything about 2D animation anymore.

Well, that worked exceedingly well the first time. 😄 Added an “explanation” property into tool_checks for OpenAI, and asked it to explain why it called the function: and now can totally fix the function’s definitions to account for the misinterpretation. Highly recommend!

Pinecone (the vector store behind Subtxt’s smarts) walks you through using Canopy, their open source framework for streamlining Retrieval Augmented Generation this Tuesday, live at 9am. I built my own over a year ago, but always up for learning something new.

To everyone jumping up and down with joy about The NY Times lawsuit: OpenAI educates you on the reality of our new world.

Training AI models using publicly available internet materials is fair use, as supported by long-standing and widely accepted precedents. We view this principle as fair to creators, necessary for innovators, and critical for US competitiveness.

The principle that training AI models is permitted as a fair use is supported by a wide range of academics, library associations, civil society groups, startups, leading US companies, creators, authors, and others that recently submitted comments to the US Copyright Office. Other regions and countries, including the European Union, Japan, Singapore, and Israel also have laws that permit training models on copyrighted content—an advantage for AI innovation, advancement, and investment.

Andrej Karpathy on naming conventions:

I’m playing around with calling our tech, as it is today, IA (intelligence amplification) instead of AI. IA have the vibe of tools for thought, needing human interaction, and resemble a lot more what we actually have today. AI feels more like independent long-running agents.

Intelligence Amplification is spot-on and seems less gimmicky a buzzword.

Unfortunately in Europe that’s what AI means and IA means “artificial intelligence”. 😄

“Attention is all you need” spelled out in this great step-by-step tutorial on building your own transformer (GPT).

Attention was the key discovery that enabled the transformer architecture. The idea is that each token should be able to communicate with or look at each previous token in the sequence but not future tokens. For example given token number 4 in a sequence of 8 tokens, token 4 should be able to access token 1, 2 and 3, but not tokens 5 through 8.

This concept is what makes it possible for a computer to have an intelligent conversation with a human.

I figured out the one person I’m “writing” the Six-Figure AI App for…(old Stephen King trick). Makes it so much easier to keep moving with that person in mind…and the clarity makes the process a joy.

The GPT Store from OpenAI releases next week. With a couple years behind me, I can say that without evals engagement falls off. The initial excitement of “wow, look at this” dies down once hallucination and frustration sets in. Custom GPTs don’t have this necessary step for GPT creators.

Just got my first “multi_tool_use.parallel” in action. The tools it picked–I’m not even sure I would do that manually, but pretty much everything the LLM requested was necessary for the conversation and development of the story. This is crazy.

Ahh! This is GREAT news. 🚀

Huge news for OpenAIDevs: API key based usage is here 🥳 To get started, head to the API key page and generate a tracking token for each key which will enable per key tracking in the usage dashboard for all new requests!

Now, to resist the urge to micro-manage all my requests so I can track every last request! 🤗

I can’t wait until the GPT Store rolls out next week and everyone finds out that there is SO MUCH more that goes into creating a great AI experience than just uploading some PDFs.

Subtxt: it’s not just for screenplays.

Ethan Mollick has a completely generated AI video of himself (with a crazy-realistic and random mic adjustment with his hands). All I can think about are the hundreds of hours I have of me teaching story that I can now use to give Muse a face. 🤯

Crazy to me the differences between GPT-3.5 Turbo and GPT-4 Turbo when it comes to the recent function calling (tool calls) and JSON mode. The former is unworkable, whereas the latter–while slower–delivers more reliable results.

Overheard today: Midjourney CEO in office hours just said he thinks they “can get to the holodeck” by 2024 😳

“We’re gonna build a lot of stuff this year. I think we’ll build more stuff than I’ve ever built before…By the end of 2024 hopefully we have real-time open worlds”

Great article about a great writer (“Logan” and “The Queen’s Gambit” scribe Scott Frank, Script Doctor to Showrunner). Though I will say, $300,00 a week to add a Main Character Throughline seems a bit excessive. 😁

The JSON response from GPT-3.5 Turbo has degraded so much that I can’t use it anymore in production. Had to hardwire GPT-4 as the responses back were so inconsistent. Looks like I can go back to it if I provide examples (one-shots) or fine-tune, but nor now, its persona non grata in Subtxt.

Wish Brockman from OpenAI posted elsewhere besides X. Until then I’ll just copy and paste these words of motivation:

One trick to making ML progress is repeatedly increasing the precision of the engineering and decreasing the number of mysteries being papered over – i.e. just focusing on improving your understanding of the system you’re working with.

I can’t tell you how many times I have gone over and over the AI in Subtxt, conveniently refining and shaving off the edges.

It takes some getting used to.

Posted the Dramatica Users Group analysis of Barbie event into the Narrative First School

Join us for this month’s Dramatica Users Group analysis, where Chris Huntley leads an insightful discussion on the narrative structure of “Barbie.” Discover the intricate storytelling elements that bring this iconic character to life. Engage with fellow narrative enthusiasts and deepen your understanding of Dramatica theory in action.

Published a new article on Narrative First, all about How to Use a Storyform for Developing a New Story. What started out as an innocent enough attempt to explain an objective point-of-view of narrative devolved into a simple case of projection and misunderstanding.

While most of Bluesky cries over AI-“plagiarism”, there are others building the future with it:

March of last year, I didn’t know how to code in python. I asked GPT-4 to implement the basic functionality of getting HTTP data from an API endpoint back in April/May of last year.

Then, I put the code base on a secret Gist and kept pointing the GPT-4 Plugin to its own source code with new feature requests, improvements, and bug fixes. Directing its attention to the parts it was to work on next.

I repeated this process several thousand times over the course of the next 7 months to today. And now it’s a fully Asynchronous Quart ASGI service pushing 5,000 lines of code, and can do all this

It’s AI building itself.

It also taught me Python in the process. I am quite proficient now. And the Python it wrote is quite capable. Here’s an example of how we persist sessions with no authentication necessary just by maintaining a working knowledge model of the Ephemeral User ID’s and Conversation ID’s.

Vue 2 hit EOL on December 31st. Prior to AI, upgrading Subtxt would’ve taken months. With AI, I’ll have it done by the end of the month.

Might even use it to convert the friend to a Swift app, and publish that in February.

Absolutely great quote from Arvid Kahl:

Making money is a consequence of making a difference in people’s lives.

Molick posts a paper covering AI reasoning:

One contentious debate in AI research is about the ability of LLMs to reason like a human does. While there is still a lot of uncertainty, this paper from Google has a handy chart on the kinds of reasoning tasks AIs are tested on & how they do (GPT-4 wins)

This is more about people enjoying being a part of the creative process. They’re actually enjoying it more than if they were just watching.

Everyone so excited about potential lawsuits over “stolen” artwork have no idea what’s coming next.

Spatie.be with a great new package for creating PDFs in Laravel apps. Wondering what it would mean to have a one-button pitch packet available in Subtxt now…

OpenAI’s chief engineer Greg Brockman’s prediction for AI in 2024:

Prediction: 2024 will feel like a breakthrough year in terms of AI capability, safety, and general positivity about its potential impact. In the longer term, it’ll look like just one more year on an exponential that can make everyone’s lives better than anyone’s today.

If the person at the center of it all is saying this, you know that there are some pretty big things coming our way this year.

I “work” every day of the week, but am happier than when I only worked Monday through Friday. The secret? Working on something I’m deeply passionate about (good stories).

If you’re not fulfilled? Carve out something of your own. Choose the hard that lights a fire under your ass and gets you out of bed with excitement every morning.

Jay Clouse with a great list of strategies for starting to your own digital business. I did every one of these over the past ten years and can say quite confidently that they all work. It takes time, but if you’re consistent you’ll be the architect of your own future.

I believe what Yann is suggesting is that there exist more lucrative methods for disseminating information than traditional books. Overlooking this evolution in info delivery would be akin to me steadfastly adhering to traditional 2D animation while the rest of the industry transitioned to 3D.