__A garden-path sentence is a grammatically correct sentence that starts in such a way that a reader's most likely interpretation will be incorrect; the reader is lured into a parse that turns out to be a dead end or yields a clearly unintended meaning. "Garden path" refers to the saying "to be led down [or up] the garden path", meaning to be deceived, tricked, or seduced.__
I wish he built a "Edit a form in a slideover". Modals are already part of the core_components and the Phoenix team has done a good job of covering edge cases around CSS transitions. But slideovers have one aspect that's not well covered in modal flow: in/out transitions. Slideovers have a slide-in effect when they are invoked and at the end of transition load the form component which is a live_component. When you cancel/close they slide-out along with the form component. At the end of the transition is when you want to remove the live_component from live_view. Handling these transitions with live_components is very tricky.
Just came off your Elixir talk photos on Twitter to find this on HN. Please share your ElixirConf talk recording. Producing a nice emoji/avatar for an AI agent seems like a cool idea.
I did some OCR tests on some 1960s era documents (all in English). Mix of typed and handwritten. I had as results:
Google Vision: 95.62% HW - 99.4% Typed
Amazon Texttract: 95.63% HW - 99.3% Typed
Azure: 95.9% HW - 98.1% Typed
Then if curious, TrOCR was the best FOSS solution at 79.5% HW and 97.4% Typed. (However it took roughly 200x longer than Tesseract which was 43% HW and 97.0% Typed)
When did you do this test? I don't have any numbers handy, but a couple years ago I compared google's OCR vs AWS's on "text in the wild" pictures. AWS' wasn't bad, but it was definitely outperformed by the google one. The open-source solutions I tried (tesseract and some academic deep-learning code) were far behind.
This was a couple months ago now, so not that long ago. For OCR I have found that it highly depends on the type of image you are looking at. In my case these were all scanned documents of good but not great scan quality, all in English. I expect if you were talking about random photos with text in them, you'd see the FOSS solutions do much worse, and much more variance in the Google vs Amazon vs Azure. I would be curious about the academic deep learning one you tried.
The main one was https://github.com/JaidedAI/EasyOCR, mostly because, as promised, it was pretty easy to use, and uses pytorch (which I preferred in case I wanted to tweak it). It has been updated since, but at the time it was using CRNN, which is a solid model, especially for the time - it wasn't (academic) SOTA but not far behind that. I'm sure I could've coaxed better performance than I got out of it with some retraining and hyperparameter tuning.
Interesting. I tried easyOCR, I found on handwriting it was about 35%, on typed it was 95.7%, so not bad at all with typed, but for handwriting pretty bad. I focused on Tesseract and TrOCR since it wasn't working out that well, still could easily have just been my particular use case.
I also tested paddleocr and keras ocr to round them all out.
At some point I really need to finish my project enough to write up some blog articles and post a bunch of code repos for others to use.
I got ChatGPT to write a python script to scrape entries in this post and output it as a csv file. Download it here https://github.com/subbu/ask-hn-candidate-list-chatgpt/blob/... and open in it Excel/Numbers. Finally I got ChatGPT to output my conversation with it into markdown https://github.com/subbu/ask-hn-candidate-list-chatgpt. I can host this script on modal and run it in a cron job, but that's overkill. I'll probably run this again and update the candidates list. This whole thing took me ~30 min.
Your youtube playlist combined with NanoGPT and your Lex Fridman podcast is like having a university level degree with a free internship guidance. Thank you!
This, totally. I manage a large B2B CRM with complex user-auth requirements and we use an off the shelf provider. It was an easy decision when we began, but now it has become a bottle neck in many ways. As your application becomes big you need a lot more customization to your auth logic. Large customers have specific customizations that off the shelf auth providers don't provide, or make it complex to implement. Over time auth and RBAC becomes crucial to your application. Engineering effort required to read through your providers documentation and implement workarounds exceeds a purpose built, native solution.
I really wish intelli-j provides a first-class editor. I could never get VSCode + ElixirLS to work smoothly. It's just broken. I have switched to NeoVim + plugins. It feels better mostly because of Vim's eco-system. But I am missing a lot of productivity due to lack of a good IDE.
