barrell

I recently started learning a tonal language, and so far have not struggled too much wrt tones when everything is slow. There was an original strangeness and refusal for my vocal cords to want to work that way, but probably only for the first month or so.

At least, this is the case for slow text. Once the text is sped up it’s amazing how my brain just stops processing that information. Both listening and speaking.

I’m sure this will come with practice and time but for now I find it fascinating

I mean it’s closer, but I don’t think it’s right to equate commissioning an artist with paying a multi-billion dollar corporation to steal from artists.

These tools are just lazy shortcuts. And that’s fine, there’s no problem with taking the lazy way. I’m never going to put in the time to learn to draw, so it’s cool there’s an option for me.

I just take ire with pretending it’s something grand and refined, or spitting in the face of the ones who are willing to put in the work

Agreed. I don’t care for engineering or coding, and would gladly give it up the moment I can. I’m also running a one man business where every hour counts (and where I’m responsible for maintaining every feature).

The fact of the matter is LLMs produce lower quality at higher volumes in more time than it would take to write it myself, and I’m a very mediocre engineer.

I find this seperation of “coding” vs “building” so offensive. It’s basically just saying some people are only concerned with “inputs”, while others with “outputs”. This kind of rhetoric is so toxic.

It’s like saying LLM art is separating people into people who like to scribble, and people who like to make art.

I just rewrote all the graphs on phrasing [1] to webgl. Mostly because I wanted custom graphs that didn’t look like graphs, but also because I wanted to be able to animate several tens of thousands of metrics at a time.

After the initial setup and learning curve, it was actually very easy. All in all, way less complicated than all the performance hacks I had to do to get 0.01% of the data to render half as smooth using d3.

Although this looks next level. I make sure all the computation happens in a single o(n) loop but the main loop still takes place on the cpu. Very well done

To anyone on the fence, GPU charting seemed crazy to me beforehand (classic overengineering) but it ends up being much simpler (and much much much smoother) than traditional charts!

[1] https://phrasing.app

I’m personally a high fan of light mode, and rarely use dark mode. That being said, the comparison with paper isn’t very apt. LCDs produce their own light, and OLEDS exacerbate the difference by only emitting light for the background, and none for the text. It’s very different from using ambient reflected light.

If you take a non backlight e-ink display and put it next to a light mode OLED with the same brightness as the environment, the difference is (slight pun intended) night and day. There’s no configuration possible on my phone to make it more legible than a book it sits next to in low light conditions.

Also after years of having light mode only Phrasing, I recently added a dark mode. Not to look cool in dark mode, but because it’s often the first thing I use in the morning and last think I use at night. After a year of minim-brightness still-squinting at the screen, it was truly remarkable the physical difference a #000 background made in the dark with an OLED screen.

I’m not convinced the pen pot needs any more learning than anything else. Even the ones with the paper - is it a word processor, emailing tool, something about newsletters? Maybe a PDF or markup tool? Or a layout tool for print media? Or just a signature tool?

At some point, the user has to find out, in the same manner they find out about the pen pot.

I think users could easily associate the “pen and poison potion” with word processing for years until someone says “click on the pen and ink” and then they have a lightbulb moment.

I think we went from icons being “visually distinct” to “visually descriptive” to “visually uniform”. Personally I prefer the visually distinct. I’m not convinced we gained some massive leap forward in usability moving away from it; I know I struggle substantially more to find an app or tab that I’m looking for nowadays than when I first got a Mac.

Thanks!!

Omg I just realized you commented on my last post and I spelled your name wrong in my post. I’m so embarrassed, and that comment makes so much sense now XD I didn’t realize that was you

Yeah I really spare no expense on the expressions. Makes them expensive, but also very effective

Thank! Regarding the translation, Phrasing is Bring Your Own Content. In my initial launch, I used an official translation for the demo, but apparently chose poorly haha

For languages without the Roman script - you can enter in qwerty (or configure it to work with any script). When you tap on a word, it shows you the phonetics; and if you tap on the shapes feature you can see that information for every word at the same time.

The app will be default display everything in its native script intentionally. With a bit of curiosity, you’ll pick up the script soon enough. Normally within a few weeks for easier-from-latin scripts like Greek and Cyrillic, to a few months for harder scripts (although you’ll pick up the basics still in a few weeks).

For Japanese and Chinese, I have more suggestions though. Which language(s) are you learning?

All great feedback! Thanks. Two questions if I may:

1. Where did the complexity start on the page for you? From the top it did it lose you somewhere?

2. Did you watch the video, and was that included in the complexity comment?

Can’t say I’m able to quickly get to any of your feedback but I’ll take it to heart and apply it over the coming weeks/months :)

There’s part of this that I agree to - I tend to disagree with most anti-capitalist (or anti-profit) sentiment. However, I disagree that builders “owe” anyone anything, and I strongly disagree with goal of as much profit “as possible”.

I miss the days when someone would make a service where the user would benefit as much as possible and the creator got compensated fairly. I feel like that system worked for hundreds of years. It’s only in the last couple decades that we’ve made this obligation for maximal profits - something that I personally hold responsible for all the mass enshittification going on these days.

I do this a lot as well. I have a bad habit of just freezing physically when I start to think. Since my work is conducted over video, my colleagues will often think I dropped off the call XD

Personally I find it a bad habit of mine, I have no idea how people gracefully take time to think. Whenever I do say something like “hold on, let me just think for a moment” my brain completely freezes and I get no thinking done.

x

Everyone keeps making religious connections, but it’s worth pointing out linguistically most countries in Europe (and the world) refer to red meat, poultry, and seafood as entirely separate. Meat often just refers to red meat. English is in the minority to bucket them all under “meat”

I think LLMs are just another commodity. I think AI would be something different.

Emphasis on the conditional tense.

I like the comparison with fossil fuels, that feels very accurate. I hope we do not develop the same dependency.

“Nothing holds any value” and “nothing costs any money” are two entirely different ideas. The former is nihilism. The latter is communism.

Neither paint a pretty picture in my book

My understanding of 4.5 was that it was released long, long after the initial training run finished. It also had an older cutoff date than the newer 4o models

Not everyone is a nihilist

Yes

Yes but what happens to society when art, labor, "intelligence" and productivity all become abundant? These are not comparable to salt. You are comparing an apple to every orange tree that has ever existed.

I do not think value and scarcity are identical, merely that they go hand in hand. In the desert, I would pay a lot of money for water. In the west, I would never pay for a glass of water, even if it's blistering hot and I'm parched. The requirement is the same, the value only changes due to scarcity.

What you are talking about is needs, which is an entirely different discussion. Value can also be used to discuss needs (how valuable is something to survive), but I think I'm quite clearly using it to describe something different.

Price is just a proxy for value. Diamonds do not have inherent utility (to the layman) but they are expensive because we societally ascribe value to them.

But that is exactly the promise that the heads of the AI labs are making. Sam Altman is repeatedly saying he wants to be able to ask it "go discover a new branch of physics".

Perhaps it's not how you use LLMs, but it is the promise of AI.

For the record, I make a distinction between LLMs (a current tool that we have today) and AI (a promise of some mystical all-powerful science-fiction entity).

There is nothing intelligent about what we have today, artificial or otherwise.

An idea that has been living rent free in my head is that "AI is ultimately nothing but a pure destruction of value". It's promise is unlimited value to everyone on demand; but if everyone can do everything without any effort, it is no longer valuable. Value and scarcity go hand in hand.

I realize the hyperbolic framing of the idea, but none-the-less I haven't been able to get it out of my head. This article feels like it's another piece of the same puzzle.

If I cannot tolerate a failure rate, I do not use LLMs (or and ML models).

But in that case the larger the better. If mistral medium can run on your M2 Ultra then it should be up to the task. Should eek out ministral and be just shy of the biggest frontier models.

But I wouldn’t even trust GPT-5 or Claude Opus or Gemini 3 Pro to get close to a zero percent success rate, and for a task such as this I would not expect mistral medium to outperform the big boys

Reasoning was set to minimal and low (and I think I tried medium at some point). I do not believe the timeouts were due to the reasoning taking to long, although I never streamed the results. I think the model just fails often. It stops producing tokens and eventually the request times out.

What's your acceptable error rate? Honestly ministral would probably be sufficient if you can tolerate a small failure rate. I feel like medium would be overkill.

But I'm no expert. I can't say I've used mistral much outside of my own domain.

Heh it's a quote from Archer FX (and admittedly a poor machine translation, it's a very old expression of mine).

And yes, this only happens when I ask it to apply my formatting rules. If you let GPT format itself, I would be surprised if this ever happens.

Thank you :) and you're definitely not the only one.

Full transparency, the first backend version of phrasing was 'vibe-coded' (long before vibe coding was a thing). I didn't like the results, I didn't like the experience, I didn't feel good ethically, and I didn't like my own development.

I rewrote the application (completely, from scratch, new repo new language new framework) and all of the sudden I liked the results, I loved the process, I had no moral qualms, and I improved leaps and bounds in all areas I worked on.

Automation has some amazing use cases (I am building an automation product at the end of the day) but so does doing hard things yourself.

Although most important is just to enjoy what you do; or perhaps do something you can be proud of.