samuelzxu

Awesome! Currently it looks more dev friendly, if you want go after PowerPoints I think it’ll be nice to make it look more MBA friendly ;)

Not yet! Still need to make the perspective correction more reliable first, but multilingual (Mandarin, French, Spanish and Korean!) are on my list :)

I trained a model to do perspective correction and now I've got an automated visually coherent AI tutor working from a propped up iPhone. Would love to discuss if anyone has questions!

Previous iterations of this were purely on MacOS. This one is on iOS. Link to both is below:

https://apps.apple.com/ca/app/knowable-the-ai-tutor/id676378...

A visually grounded AI tutoring application: https://knowable.ca Been working full-time on it for a few months, I'm trying to make the concept of an AI tutor better than just a chatbox.

Thank you! Let me know if you have any feedback, happy to chat :)

Thanks for the input! I'm aiming it at students and parents, highschool-level for now.

Ah it requires a special camera that only macbooks have :( But I'm working on a tutorial to get it working via an iPhone stand + continuity camera!

Thanks!! I have voice mode working on the MacOS app - working on getting it going on Safari as well :)

https://apps.apple.com/us/app/knowable-the-visual-ai-tutor/i...

I think we just haven’t spent enough time creating the systems to help people learn faster - so far, only the systems that teach AI to know more and for people to rely on AI to retrieve information.

AI + Education is really interesting but also pretty tough to get right. Working on something that is hopefully going in the right direction: https://knowable.ca

I'm working full-time on a personal AI tutor that just requires a recent Macbook, available for free on the Mac App Store! You'll get a full ~60 minutes. Vision LLMs are awesome.

Happy to hear some feedback!

Supported so far:

Macbook Pro 2024+, Macbook Air 2025+, or (a recent MacOS + iPhone continuity camera)

Code:

https://github.com/samuelzxu/Knowable https://github.com/samuelzxu/knowable-be

For feedback please email me at: samuel at knowable dot ca

Been using Shinka Evolve for a while now, but sometimes I just want it to use claude code only (they require lots of API keys for a model ensemble). So, I made claude-evolve, a claude code plugin for doing the same thing. The model ensemble is less powerful (thinking effort x [opus, sonnet, haiku]) but I've been using it and it's been useful enough to share it with some folks!

Sprites are AWESOME! They look so cool, does anyone know a good place apart from wikipedia to read about them?

Woah! How is this not more popular? I don't see it referenced in the naturalspeech2 paper anywhere.

Does anyone know how many words would correspond to the diffusion model's batch size of 6000 frames?

Curious to know whether anyone else here have read this, and their thoughts on it. I've just finished my first reading and found it entirely impossible to stop until I had read the whole thing. I've left a comment to kick off a discussion:

- There's an obvious symmetry with our own times, where stories and facts are smeared together. It's touching how clear it rings of "we see what we want to see" in the modern world of an infinitude of hot takes and narratives. Of course, even more recently, generative AI puts an even more literal spin in this direction.

- Desire and Will. In this book these two are differentiated in that desire cannot be changed, whereas you can will a desire into existence. I'm not sure about the philosophical basis of this - I've not read The World As Will and Representation - but I have a feeling it might strike some similar notes.

- In the interest of not spoiling the book, I'll say the carnal stuff is very in-your-face and deliberately so. The author was definitely shocking when she tried to be.

Thanks for the feedback! I tested the app out on my android tablet, and my S pen worked fine on my tablet. I'm not sure about Windows though ...

Thanks for the feedback! Yeah, it fails in some very simple cases, but does really well in others. I believe it does the best if your letters are tightly grouped, and can even recognize cursive pretty well, but space the letters out and it fails pretty easily ;)

This is a handwriting-to-text converter! Just follow the instructions on the page and you're good to go :)

Background: I've been tutoring on the side for a while and it's apparent that the whole process can be smoothed out, with the end goal being an AI tutor buddy with a stylus interface. This is a little step in that direction.

As for implementation details, I forked excalidraw (at https://github.com/excalidraw), got a gcp free tier instance running, and scraped together a Google K8s Engine cluster serving with torchserve. Luckily there's a great deal on the public preview of c3 cpus at the moment. For the model, I'm using trocr-base-handwritten ( https://huggingface.co/microsoft/trocr-base-handwritten ).

Let me know if anyone has any ideas, suggestions, and/or tips!

There's also this paper with very similar methodology called Mind-Vis, and also accepted to CVPR 2023. https://mind-vis.github.io/