The Japanese government stated that AI training data does not apply to copyright[1]
The Japanese government recently announced that it will not enforce copyright protection for AI training data, a policy that allows AI to use any data, "whether it is for non-profit or commercial purposes, whether it is an act other than copying, whether it is obtained from an illegal website or otherwise. ". Japan's science minister, Keiko Nagaoka, confirmed the bold stance at a local meeting, saying Japan's laws would not protect copyrighted material in AI datasets. The move is part of Japan's efforts to become one of the leaders in AI technology. The decision is controversial for anime and graphic art creators who worry that AI might devalue their work. Instead, academic and business figures have urged the government to use the country's lax data regulations to advance Japan's global AI dominance. This move may increase Japan's GDP (50% and above) in a short period of time, which is an exciting prospect for Japan, which has suffered from years of sluggish economic growth.
The Evolution of Slide to Unlock[2]
This article explores the history and evolution of the "slide to unlock" feature in smartphones. From the "slide to unlock" of the first-generation iPhone to the later Touch ID and Face ID, the author analyzes the advantages and disadvantages of this function, and puts forward some suggestions for improvement in order to better provide consumers with convenience and security unlock method.
OpenAPI v4 (aka Moonwalk) Proposal[3]
This article introduces the OpenAPI v4 (aka Moonwalk) proposal, discusses the goals and features of OpenAPI v4, provides some examples, and briefly describes its structure. This proposal aims to provide API designers and developers with better tools and experience to improve API quality and usage experience.
Notes apps are where ideas die (2022)[4]
We write notes not to remember them, but to forget them. Notes apps make us feel safe and we can save all our flash thoughts, but most of those ideas aren't too important and we just need to feel like they're still there. Still, note-taking apps are an incredibly valuable tool that makes us feel safe forgetting. In order to remember the more important things, we need to forget some less important ideas.
Mysterious man spying on theme parks from the sky[5]
Bioreconstruct is a Twitter account that seems to post almost daily aerial (and ground) photos of Walt Disney World and Universal's Orlando resort, especially construction projects. The account is a huge resource for theme park journalists, bloggers, and fans, but there is little public information about the people behind it. Even within theme park blogosphere, Bioreconstruct's identity is completely intentionally kept secret, even though everyone knows who he is due to their amazing photography skills (both on the ground and in the sky). In the world of Twitter, Facebook, YouTube and Tik Tok, this is extremely rare.
I decided not to commercialize nbdev[6]
The author of this article describes how he faced user adoption resistance and commercialization challenges in the process of building nbdev, a software development framework for Python. Eventually he abandoned his plans to commercialize nbdev and instead focused on other machine learning projects. His lessons include always having a strong conviction and leading an independent brand, while being brave enough to make the switch. In the future he may refocus on nbdev, or explore similar projects.
New microrobots can move quickly in liquids[7]
Engineers at the University of Colorado Boulder have designed a new class of tiny autonomous robots that can move through liquids with incredible speed and could one day deliver prescription drugs in hard-to-reach places inside the human body. The team's findings were published last month in the journal Small.
Teddit - an alternative Reddit frontend with a focus on privacy [8]
Teddit is a Reddit front-end replacement focused on protecting user privacy. It provides many configuration options such as defining a domain name, enabling Redis caching, enabling video playback, etc. It is also possible to enable HTTPS, set sorting options, set Reddit app ID, etc. Additionally, it supports features such as auto-topics, comment disputes, and displays recommended forums.
OpenAI according to Sam Altman's plan[9]
Last year, Sam Altman met with 20 other developers to discuss OpenAI's API and their product plans. Their short-term plans are GPU-bound, and their long-term plans are more social. Their plans include a cheaper and faster GPT-4, a longer context window, a fine-tuning API, and a stateful API. They don't compete with customers, but still need regulation and open source.
When LIMIT 9 works but LIMIT 10 gets stuck [10]
This article introduces an error case about PostgreSQL query in WebSocket connection. When using order by and limit 10, the query will be stuck, but it can run normally without order by. The author reads WebSocket F...
Falcon 40B LLM (over Llama) is now upgraded to Apache 2.0[11]
Falcon 40B LLM, an open source project for low-latency messaging, has been upgraded to Apache 2.0 with better performance and security. The project aims to provide an efficient and scalable solution by combining traditional message queue processing with the LMAX Disruptor framework. This update adds several new features, including modular programming, more performance statistics, and easier customization and extensibility.
What are shaders? [12]
This article explains that shaders are programs that run on the GPU, whose role is to process graphics in parallel and present them on the screen. Different operating systems have different shader writing methods, such as: GLSL (related to OpenGL), HLSL (Microsoft's shader language), Metal (Apple's shader language) and SPIR-V (binary standard shader specified by Khronos language). The shader pipeline includes a vertex shader and a fragment shader. The former processes vertices, while the latter draws the color value of each pixel of the triangle.
Compile Scheme directly to WebAssembly: Lambda, recursion, iteration [13]
The article introduces the Hoot project, a project that compiles Scheme into WebAssembly. Hoot achieves direct compilation through Guile's compiler and through the lispy structure provided by WebAssembly itself. Hoot has implemented functionality starting with standard r7rs-small. The Hoot project provides an effective path for more programming languages to enter WebAssembly, such as Python, Lua, Haskell, ML, etc. This project is funded by Consensys/Metamask.
AI21 Labs conducts the largest ever Turing test experiment[14]
AI21 Labs conducts the largest ever Turing Test experiment. The experiment involved more than 1.5 million participants from around the world conducting 10 million "human or machine" conversations, where they needed to judge whether to talk to a human or a robot based on AI language models such as Tornado-2 and GPT-4. Some popular testing strategies include checking for grammatical errors and use of slang, asking personal questions, asking about current events, and examining a machine's expressiveness on topics such as philosophy and ethics. According to the available results, people are more likely to recognize human conversations, with the highest correct recognition rate in France and the lowest in India.
Raiders "smart" toothbrush: Hackers discover secrets from it[15]
This article describes what a hacker discovered after purchasing a Philips Sonicare toothbrush: The toothbrush flashes an LED light when the head is inserted. After a quick online search, he found that the toothbrush head communicates with the toothbrush handle. When do you need to buy new brush heads. He used the NFC tool to find that the toothbrush head has a built-in NTAG213, and the data at address 0x24 stores the usage time of the toothbrush head. He eventually succeeded in several rounds of trying to get the toothbrush head password with the SDR device, but he didn't try to crack the algorithm to produce the password.
Improving Mathematical Reasoning through Process Supervision [16]
This article illustrates how complex trigonometric expressions can be simplified using trigonometric identities, with iterative application of the identities ultimately leading to the simplest form of the expression. This method can improve mathematical reasoning ability and problem-solving efficiency.
quote link
[1] The Japanese government stated that AI training data does not apply to copyright: https://technomancers.ai/japan-goes-all-in-copyright-doesnt-apply-to-ai-training/
[2] The evolution of slide to unlock: https://cs.uwaterloo.ca/~csk/slide/
[3] OpenAPI v4 (aka Moonwalk) Proposal: https://github.com/OAI/moonwalk
[4] Notes apps are where ideas die (2022): https://www.reproof.app/blog/notes-apps-help-us-forget
[5] The Mystery Person Who Spies From The Sky Theme Parks: https://boingboing.net/2023/05/30/the-mystery-person-who-spies-on-theme-parks-from-the-sky.html
[6] I decided not to commercialize nbdev: https://hamel.dev/blog/posts/nbdev/
[7] New Tiny Medical Robots Could One-Day-Travel-through-your -body
[8] Teddit - An Alternative Reddit Frontend Focused on Privacy: https://codeberg.org/teddit/teddit
[9] OpenAI according to Sam Altman's plan: https://humanloop.com/blog/open_ai_talk
[10] When LIMIT 9 works but LIMIT 10 hangs: https://neon.tech/blog/when-limit-9-works-but-limit-10-hangs
[11] Falcon 40B LLM (better than Llama) is now upgraded to Apache 2.0: https://twitter.com/Thom_Wolf/status/1663986216771936263
[12] What are shaders? : https://thatonegamedev.com/other/what-are-shaders/
[13] Compile Scheme directly to WebAssembly: Lambda, recursion, iteration: https://spritely.institute/news/scheme-to-wasm-lambdas-recursion.html
[14] AI21 Labs conducts the largest Turing test experiment in history: https://www.ai21.com/blog/human-or-not-results
[15] Raiders "smart" toothbrush: hackers discover secrets: https://kuenzi.dev/toothbrush/
[16] Improving mathematical reasoning with process supervision: https://openai.com/research/improving-mathematical-reasoning-with-process-supervision
You must be logged in to post a comment.