Today I am trying a bilingual version in Chinese and English. Writing in Chinese allows me to express my thoughts more clearly in my native language, while writing in English is mainly for practicing my writing skills.
What cool things have I done?#
- Finally launched the first version of LinguaTale on the App Store
- I am extremely excited about the release of this product, which took a month to develop. Apple was also very efficient in reviewing and approving the version I submitted. It caught me off guard because I hadn't properly deployed the backend service.
- There are definitely many issues with this app at the moment, but it marks a fresh start for my long-term plan. Rome wasn't built in a day, and a good product also requires sufficient time to be polished. I adhere to a long-term perspective and will spend a lot of time gradually refining this product. The goal is to provide enough value for language learners and make LinguaTale a great help for them.
- I would also like to express my sincere gratitude to the friends who are currently using this app, especially those who have provided feedback and valuable suggestions.
What articles have I written?#
- I posted several answers on Zhihu and did some marketing for the app [🐶]
What videos have I watched?#
- RoboPianist: A Benchmark for High-Dimensional Robot Control (Teaser)
- This is an interesting animated demonstration of the paper RoboPianist: A Benchmark for High-Dimensional Robot Control. It is a study on multi-task learning, zero-shot generalization, multimodal (sound, vision, touch) learning, and imitation. It can currently generate fingerings for over 150 music scores.
- Watch the video here
What cool things have I discovered?#
- BookAi
- This app is interesting as it allows you to read and understand a book by having a conversation with it. However, it actually just creates a table of contents for the book, and everything else is likely directly accessed through the GPT3.5 or GPT4 API. During the training process of the GPT model, the content of these books has already been learned and embedded into the model, so no additional information or processing is required.
- Segment Anything
- Meta has finally released its AI product, which is an AI model that can identify and segment any object. I tried it with many pictures and the results are amazing.
- In the picture below, the model accurately recognized the stick carried by two dogs, even though a part of the stick was covered by one of the dog's mouth.
- In the picture below, the model easily recognized various vegetables in the image, such as a hyacinth bean.
- In the last picture, it was difficult to distinguish between several brown bears. I clicked on the head and butt of the bear at the back (with a large part of its body covered by another bear), and the model accurately recognized them.
- Bartosz Ciechanowski
- This is a very long blog post, but it is full of valuable content. Starting from the basics of force analysis, it explains the principles of bicycle dynamics, structural analysis, and design in a clear and interactive way. The article is in English, but if you don't have a good command of English or don't want to spend time reading, you can just watch the animations to get a good understanding.