Not OP, but I'd start with a one-click way to "install" the PWA apps using an API, e.g. with a banner (like on Android) instead of an item in the share menu.
A single file with a few functions, it seemed a good test to convert it to some other languages with GPT-4 (I tried Python and Ruby). Unfortunately, my access to GPT-4 is limited to the 2k version, and the first function is 4,500 tokens (800 minified, but losing names, comments, and probably the quality of the conversion).
With some language-independent tests in such a repository, you might be able to semi-automatically convert the code into different languages, and continue with code scanning and optimizations.
Your tl;dr shows how text+reader_context can generate the best summaries. Those 5 words are perfect if you know what they did for ChatGPT.
This makes me think that to get high quality summaries:
1) they have to be generated for each individual reader
2) the AI should know what the reader knows
You achieved 2 by imagining what HN people may already know, but the ultimate goal would be to know what the individual reader knows.
And this chain of thought leads to
- Perhaps all AI output should be generated on the fly for the end user with full (relevant and compressed) context.
- Giving AI what we know is extremely dangerous if someone wants to use it for something bad (so we really, really need local AI).
Sure, these are all things people in the industry already know, but there must be a lot of people like me who are just now thinking about it.
A note: "^2" added humor to make the reading light with very few bytes. Perhaps the machine would do this too, if the reader wants.
Only few things make me very very excited and sometimes very very sad like some AI developments. But the exciting part wins for now!
Wow, textbundle on disk instead of the database, right? It would be huge.
Another great feature that Bear and FSNotes share is the ability to insert hierarchy tags like #parent/child
For some reason, I find it perfect to organize notes without too much thinking.
Syncthing is very good, but the best iOS implementation, Möbius Sync[1], is not comparable with the Android one (the OS limits the syncs, you need a regular notification and so on).
The great things about Logseq are his weakness for me. Everything is so interconnected (you can say: "here put the paragraph of this other note") that I sometimes lose confidence in the system. It becomes too complex. With Obsidian I know that a note is a file. Less convenient but simple and reliable.
Logseq really excels with his outlining mode, I miss it (but I don't like the way it saves states in the markdown file). It has some problems with the code blocks too.
Love Obsidian. My main problem with it and similar markdown apps for notes is the way they store images and attachments. I find it very confusing to maintain multiple files per note and IMO the only app that nailed it is FSNotes[1] using the textbundle[2] format (with a custom implementation for encrypted notes too). I think it's elegant and future-proof.
But FSNotes is for the Apple ecosystem only and I can't tie myself to a single platform for something so important (I don't need another artificial reason to make OS switching so difficult).
I hope the textbundle feature request[3] gets some love soon. It would be great for Excalidraw files integration too[4].
(not op) Logseq is open source and really great in the way it connects notes and highlights pdfs, but the sync part is too “asynchronous”. It works well if you use a single device, but as long as you add something else you have to manually reindex and refresh a lot.
Obsidian is less opinionated on the txt file format and folders too, so I consider it more future-proof.
Italians have a precise word for that: "supercazzola". It's when someone voluntarily uses this nonsense bs to sell something or simulate some kind of knowledge.
Sure! This would be my approach (and tools) if I was smarter:
If you make the generations with some similarities and use the right interpolation, you don't need 1000 images like my video and can obtain a smooth movement.
First, generate images with some kind of visual anchor (background, an object). You can use frames generated using the previous frame as reference image, or the same seed but different prompt/parameters, or you can go wild using img2img/inpainting (btw I struggle to find an inpainting tool for Stable Diffusion: they seem to be just img2img with a mask, without contest).
The first link reports some free and paid implementation and a Colab, so depending on how deep you want to go, you have a lot of choices.
In the end, I'd use some good app to stabilize the image if needed, to get a more "calm" look. I use Luma Fusion, but it's a paid app (cheap, one-time payment, for iOS). I'm sure there are a ton of open-source implementations.
It's an approach similar to the animation on replicate, but it allows a lot of fine-tuning and you can add new animation ideas/tools to the process.
Nothing revolutionary, but I hope it helps!
> You have generated some pretty cool designs.
Thanks! I put in a lot of work in the last weeks. The project has a mission, I wrote something, but it's not ready yet. I believe it will be with the launch of Dall-E 8 :-/
Very cool! I generated this with 1000 images https://twitter.com/UnshushProject/status/156315821457709465... using the Deforum's Colab[1], it's really easy and now has interpolation too. It was the very first video, I could have made something great but, you know, awesome guys keep releasing AI tech and I'm like a child at Luna Park right now, not able to concentrate.
If you are interested in my project (I doubt, you are too busy playing like me) I'm posting a lot of things on https://unshush.com and on the Instagram account: https://www.instagram.com/unshushproject/
(Sorry for posting my stuff but I'm not very social so no one will ever see them)
If you want to generate videos I can share some links I bookmarked of software/code to make them more smooth.
There is always some cherry-picking, but prompt engineering is an art per se, you become better and better working on it.
I just started this experiment
https://www.instagram.com/unshushproject or without Instagram https://unshush.com
and spent A LOT of hours and patience to become good at it. Now I'm very proud of my results and I'm working on doing better.
It's a bit risky to invest too much time because every generator is different and they change the underlying model frequently (see the beta of MidJourney yesterday), but if you do it for passion or curiosity there is no problem.
Now I'm experimenting with a local installation of Stable Diffusion (well, not really "local" because I have an old computer) and the prompt is only one of the things you can tweak. There are num_inference_steps, guidance_scale and other parameters.