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

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.”

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).

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

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.

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.

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…

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.

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.

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.

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.

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! 😅

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.

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.

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. ⃝

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. 🤯

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.

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…) 🎨

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

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

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.

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

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.”)

How is this even possible?!!

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

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?

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

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!).

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. ❤️

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.

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)

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. 😄

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.

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.

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.

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). ☀️

Coming soon to Subtxt…

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

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

It will never not blow me away that this is even possible. I will be Function Calling everything in Subtxt.

Unbelievable. From nothing to full-blown web app integrated with my decades old blog in just 5 days! 🤯 The new Laravel/Folio/Volt method is such a delight. ⌨️

The problem with developing with an AI assistant (GPT-4 for me) is that now, I can do so much in one day that I really have to find ways to take a break. It used to be that I would take a break when things get frustrating–but now, I never run into roadblocks so it just keeps going and going!

While others are focused on unraveling the mysteries of existence 🙄, we’re over here changing what it means to develop a story!

OpenAI added Function Calling, making it even easier to tap into Subtxt’s extensive knowledge-base. 🚀

Revolutionizing Storytelling

Leaked paper on “moats” and AI and how there’s no way what you build will remain a secret, nor will your models be a strategic advantage. Goes for those studios building out their AI🤫. It’s you and who you are that attracts your audience (i.e., your artists) not your IP.

This one is starred and saved for periodic reading, Why we prioritize support over sales in our SaaS business. And while at times I feel Subtext is one that requires a 68-video course (and that video demos are key), I appreciate the focus on putting people over product.

Don’t mind me. I’ll be re-installing everything after switching to Monterey a bit too soon 🙃 (I never learn).

Working on the new Storybeat Cards based on all the discoveries and new developments from last week. This is looking pretty good, especially when you consider this is pretty much how that opening with scene at the review board plays out.

When using your predictive narrative application to show exactly why The Shawshank Redemption is a story you can watch (or read) a thousand times, it helps to have Thomas Newman’s soundtrack on infinite repeat.

Judging by the responses in my inbox this morning, Subtext’s new Scene Resolution feature is going to be the most welcomed innovation in storytelling ever! 😃 youtu.be/RWyT605iO…

Pretty much Neo’s final speech on the phone.

Seriously?! And this is what Subtext predicts Ben should be talking about on the trip to Alderaan - you know, controlling your impulses by reaching out with your Senses! And changing your perception!!

Just going to leave this here…as this is what Subtext predicts George should be doing in the latter half of the second Act…when he finally musters up the courage to punch Biff…

When you solve a computational problem that has plagued your mind for over 22 years, there can be no better backing track than Demi Lovato and Sam Fischer’s What Other People Say.

It’s crazy how much impact a single = sign can have on an entire application. ☹️

Second app is officially profitable. 🎉 😃

I could really get into this—what am I going to build next?!

Who would’ve thought that breadth-first order and depth-first order would have anything to do with telling a story? One floods the landscape to find the topology. The other is like a rat in a maze constantly backtracking. Wonder which is plot and which is character 🤔

So cool building an app with someone else (my first after Subtext! 😃) Much easier the second time around once you figure out where everything goes. Rolling out this Tuesday to a select group of people (5,000!)

The view from your writing desk when you don’t use Subtext.

Just had my first support request where I could point them to my brand new feature that saves a ton of time and avoids convoluted workarounds. Hat a great unexpected side-effect 😃

I’ve received more emails of congratulations and appreciation after today’s release than anything I’ve done in a long time. Feels amazing to know so many appreciate the work and I’m so grateful that I can connect with so many!

Witness the power of this fully armed and operational battle station!

youtu.be/qomkz8fCK…

(Really, just an insanely cool way to find narrative structure through artistic intent!)

Best soundtrack for rolling out a major release or submitting a project you poured your soul into.

And just like that, it is done.

The Premise Builder in Subtext is now live.

Have at it everyone. 😊

Holy F. If you have failed jobs in the Queue, you fix it locally and push it up, Laravel will finish off the jobs with the new code 🤯🤯🤯 who the heck came up with this stuff?! @laravelphp

My favorite button in all of Subtext. 🤓

Having @tailwindcss notify of me new components feels like I’ve hired really great designers to help out on things I didn’t even know I needed 😃

You know I’ve got a problem when I stay up late to learn more about branchless programming!

🤣 When even your complex algorithmic method for finding narrative structure can’t make heads or tails of this Premise (from “I’m Thinking of Ending Things”). Charlie Kaufman broke my server!

Subtext’s Premise Builder is pretty much feature complete now. 3+ years in the making. I feel pretty good about the balance between power & simplicity. Just need to iron out a few things and then you’ll be able to generate a complete outline from a thematic statement.

Looking back over old code, it almost breaks my heart to have remove the pieces I don’t need anymore. It’s either that, or the soundtrack from The Prestige on infinite repeat tonight…

When a single typo unravels the entire universe 🤣

When you’re using ‘mindset’ as a variable, do understand that ‘mindest’ is not the same thing (but totally looks the same the first 596 times you looked at it).

Best part about building Subtext? User requests ability to copy InstantScenes to clipboard. Brilliant idea! Time from suggestion to live on server? Less than 50 minutes. 😃

Crazy watching old classes in Subtext where I talk about crowd-sourcing narratives (which it does now). It’s great seeing your dreams come to life 😃

Watching VueJS Amsterdam conference during a late night session. Nitro for Nuxt looks incredible…and Chopin’s live-coding of a Nuxt “notes” app has me spinning with ideas for Subtext! What a great team behind this framework.

Well, it’s settled. Subtext will be used to help design the narrative of a feature film to be streamed at a later date. 😃 Not that this hasn’t been done before, but this will be with the new improved algorithms. Nothing like having a real world example to test against.

Giving HelpScout a go after taking a break from the craziness that is Intercom (way too much for what I need). So far so good, and I certainly appreciate the streamlined interface.

Currently running Lord of the Rings through Subtext. The following “arc” for Frodo was based on simply choosing Temptation for Character and Altruism for Plot. And this is what Subtext came back with…😊

Pretty happy with the new Storypoint cards in Subtext. Had to do this now that creating a story could conceivably add well over 70 new Storypoints at once! Storypoints On-Demand are now live in all accounts.

Unbelievably crazy good news about Subtext and AI just happened about an hour ago 😃 I CANNOT WAIT to tell everyone about it!

Successfully installed the new version of Laravel Spark into Subtext. Been waiting for this for awhile. I used the OG version when I started a couple yrs. ago but ran into issues once I went SPA and had to roll my own.

Feel more confident with this behind the scenes.

My new video series on Building Subtext in Public is starting here on my YouTube channel. Follow along as I show off the latest and greatest in predictive narrative storytelling.

Just used this post from Raycast (cool app) to help co-ordinate some of my distributed work this week. Not used to being my own project manager when working with others (production always did that in animation). But I can see the value in spelling it out each cycle.

Just watched the 30-min on-boarding for circle.so thinking I could build my own version–until they explained the integrations with current sites. Guess I don’t have to build anything 😃

On my application to OpenAI’s GPT-3 integration:

Subtext is an app that generates meaningful stories based on a complex psychological model of the mind. With an integration with OpenAI I can take my single sentence outlines and turn them into full-blown stories PERFECT for inspiring creativity in writers, novelists, filmmakers, and game designers. The combination of my model with yours would be absolutely insane.

Connecting AI to Meaningful Storytelling

And this is freaky scary…

I just plugged in Subtext’s random illustrations for a story based on the structure of Moulin Rouge! into GPT-3 abs this is the result:

In a futuristic society, a rebel fighter and his lover have just escaped after being held in prison for two years. While attempting to reach the safety of the Rebels, they discover that their relationship has changed–for the worse. They journey to the one place that can help them save their love: The Scarred Lands…where monsters rule.

The lovers intentionally misunderstand the other person. In the first place, they are lovers, and lovers by their nature do not understand what is said to them; they hear only what they wish. For instance lovers talk about “the beauty of nature”; but to lovers themselves the beauty of nature is simply an occasion for talking together. Again, if you call a lover bad-tempered he is not offended: he knows better than anyone that a lover ought to be in a rage at everything which hinders his setting forth or returning to his loved one; when he has no excuse for making a fuss (as in poetry) he

The Lover’s Swore : We make love with someone else, Yet they still are ours. The two lovers were asked to be honest about how they felt, but neither could bring themselves to say what was truly on their minds. In the end, they settled for silence.

Their life together was a long one, and they always said they were meant to be together. Now the man is alone, and doesn’t know how to go on without his beloved wife. He starts to wonder why he never told her how much he really loved her. He thinks about how he used to talk to her each day at lunch, but then got too busy with work and let their communication slip. One day while staring out the window of his office, he has an epiphany: he’ll wait for her under the clock at noon tomorrow. If she’s late, it means she didn’t get his message because…

🤯

This took all of five minutes. And most of that was copying and pasting.

Guess I know what I’m doing for the rest of 2021.

Seeing all the great feedback from copy.ai I am FORCING myself not to start adding GPT-3 to Subtext. The idea of generated text matched with a meaningful Storyform? 🤯🤯🤯

Must fix what’s there now, must fix what’s there now, must fix what’s there now…

The great thing about coding your own narrative theory is it forces you to reconsider initial thoughts. With previous versions of Premise, I tried to force the Story Driver (type of plot point) in when really, it plays a very minor role in a Holistic Steadfast story.

After slaving over the intricacies of narrative theory for my re-write this week, it’s time to code my latest thoughts into Subtext’s Premise Controller. In short, the app looks at the dynamics of a story and then generates a meaningful Premise from them.

First thing I do every morning is approve user-submitted Illustrations of story structure in Subtext. Just checked, since launch (10/17) I’ve added over 2600 new & unique interpretations. Essentially “crowd-sourcing” story structure. 🤓

The new Laravel Spark looks great. Had to hand-roll my own last year and I’ve never been completely happy with it. I’m hoping it won’t be too difficult to rip out the current system and replace with this. We’ll see. 🤓

Well, this is fun.

Building an app for someone instead of building an app for my story dev biz is almost…relaxing. Totally reminds me of animating scenes for Disney or Dreamworks. Not as much pressure when you’re just making it work.

Invoker sure is mighty pretty. Much nicer than TablePlus, plus ZERO config. Would have saved me HOURS doing the same with Nova…will be interesting to see how it compares.

My life is complete - Descript, the ultimate screencast/podcast/recording thingee - is now availble for BigSur. Time to get back to recording.

Cloning repositories and cleaning up code on a morning walk, as you do.

Switching colors in Linear to match my app’s brand colors makes it all that much more real. 😊

Somebody thought this would be really funny to put on my desk.

Can’t wait for Black Friday/Cyber Monday desperation to end. As someone who makes a living off selling apps and services, trust me—if you have a great product, I have no issue paying full price. If what you’re selling is good enough, you won’t need sales gimmicks. ✌️

So super excited when I finally figure something out after beating my head against the wall all day. I can sleep now.

Home office vibes courtesy of Phillips, Vari, and Apple.

Well that was easy. Told myself last week I wanted to start focusing on developing new apps with everything I’ve learned about Nuxt and Laravel over the past couple of years and less than one week later—someone hired me to do just that 😊

Full circle. Now I’m animating transitions with VueJS using what I know about spacing and timing from my animation days.

New Divide utilities in Tailwind are awesome. Feel like I could spend a couple days catching up on all the new features (and find a bunch of things I don’t have to by hand anymore!) tailwindcss.com/docs/divi…

Been using Tailwind for 2 years now, and just learned sm: is NOT for mobile devices, but SMALL screens. 🤯

When you spend your usual writing time searching for “remote NuxtJS jobs” because it sounds WAY more fun, you know you’ve got a problem with app development.

Having conquered my PWA splash screen issues, I basically just open and close my app all day. Both for validation and amazement that it actually works 😆

Figured it out. I incorrectly assumed the @2 and @3 settings in Figma were supposed to match pixel density of device. Nope! Just export at 1x and custom splash screens come up as expected.

I have great users. 😊

Witness the power of this fully-operational, fully-optimized app (thank you Nuxt :))

Raycast now opens GitHub repos in a flash…🥰

Ah, much better. DankMono re-installed in new VSCode.

Upgrading to TailwindCSS 2.0 and praying for easy DarkMode/Typography integration (so people can read my app at night). This promo video is ridiculous.

And by ridiculous, I mean wonderful.

blog.tailwindcss.com/tailwindc…

This week marks the 3 year anniversary of my writing app Subtext. 3 ½ years ago I had no idea what Laravel was, or Vue, or Nuxt, or MySQL, or GitHub, Netlify, Tailwind, or the JAMstack. Now, I use these every day to deliver an amazing app across the entire world. 😃

What used to take a minute to build in Nuxt now only takes 9 seconds on an M1 MBA. I have a feeling this might have an impact on my productivity.

I can’t stop laughing about how much faster this M1 is 🤣 and my own app I built? Ridiculous. I never realized how good it was until now. All the self-hating inner dialogue can now stop!

Really surprised at how much I can run my entire business off my iPad Pro. Previous Mac died 3 hours after last week’s keynote, new M1 its way. Last 7 days has been surprisingly productive…only thing is rebuilding app, but Netlify builds filled in the gap nicely while I wait…

My theme song when I finally get local and prod servers to work nice in Nuxt after months and months of beating my head against the wall in regards to proxy, SSL, axios, etc.

music.youtube.com/watch

2 hours from idea to deploy, Nuxt to Netlify, and now there’s an app out in the wild that perfectly encapsulates my imagination.

What a world. 🌍

Updating a 1200+ page site/app that serves thousands across the world with my phone while kicking back on a couch is my Independence Day. Thanks to WorkingCopy, Netlify, and Vue 😃🚀☕️

Very much looking forward to utilizing new Render functions in Vue 3. Evan does a great job of explaining them on VueMastery.

Upgraded the Discuss Dramatica forums to the latest and greatest version of Discourse, still my all-time favorite forum platform.

Going to tinker around this weekend with color schemes and logos 🤓

Two minutes into using @linear_app and I already know it will stick. Super slick app development and built for speed. A joy to use so far.

First pull request merged! 😊 Not at all as difficult as I thought it would be. This one a simple request to not force uppercase in a Remark plugin (Markdown stuff).

Helps make Narrative First look great.

Super happy with development on the latest version of Subtext. A preview of what’s coming next. narrativefirst.com/blog/the-…

Participating in the LIVE JAMStack Conference, put on by Netlify. It’s out of control what these guys are doing. If you want to hook up with those who are as annoyed with the recent bout of anti-JS sentiment as I am, definitely check it out.

Spent the holiday weekend conforming 1235 pieces of content to Nuxt’s new Content module. Not having to manually update index files meant I could publish three new pieces in the time it used to take me to do one.

Over and over again, I am reminded of how smart it was to convert my entire ecosystem over to Nuxt. Today’s release of their Content module (Markdown in Vue Components) is just another in this endless developer-Groundhog Day of mine.

The ease with which I can now integrate bits of my app with over 15 years of content…my entire Friday just flew out the window 😃

JAMstacking everything in 2020. Already half way there with Narrative First. Nothing but a great experience so far, and the community support is crazy good.

Top 15 from 2019: Features, Resources, and Netlify & JAMstack News

So awesome how the thing that didn’t make any sense at the beginning of the week (persisting state in VueJS when offline), now makes perfect sense on Friday.

I always feel like I’m constantly learning new things each and every week, and it really makes life fun and enjoyable.