9.18.26 | Hister private search engine, Fujitsu's FUJITSU-MONAKA CPU, Astra for Law

SPEAKER_1 (0:00)

Welcome to the Hacker News Highlights, where we explore the top 10 posts on Hacker News every day. Today, we dive into building private, full-text search engines, launching made-in-Japan next-generation CPUs, and exploring Astra for Law. Let's get going.

Title, Histr, a private search engine for the pages you visit and the files you keep. Source, GitHub. The post introduced Histr, a private local search engine that indexes the content of visited webpages and local files, allowing quick search across browsing history, bookmarks, and files without relying on cloud services. It supports full text and semantic search, runs entirely on your own machine, and offers web, terminal, and assistant integrations. The project emphasizes privacy as it avoids telemetry and cloud sync storing data only on your local server. In the comments, the community was mainly positive, praising Histr as a useful tool for personal knowledge management and historical search. Users discussed integrating it with other tools like Linkwarden and Obsidian, and debated the importance of privacy, security, and multi-device support. Some expressed interest in new name options due to trademark conflicts, while others shared similar projects or ideas, such as archive tools or browser extensions. A common theme was appreciation for its offline, local-first approach, with some concerns about security and hosting options. Overall, the sentiment was supportive and eager to try out Histr's features.

Title, Fujitsu Launches Made in Japan Next Generation CPU Fujitsu-Monaka. Source, Global.Fujitsu. The news story announced Fujitsu's new CPU called Fujitsu-Monaka, a high-performance ARM-based processor designed for data centers and AI workloads. The processor features 144 cores, ARM-EVE-9 architecture, uses a two nanometers process, and is built with a focus on energy efficiency and performance, especially for inference tasks. The CPU is designed and developed in Japan, but is expected to be manufactured mainly through TSMC in Taiwan, with plans to use a Japanese fab in the future. In the comments, the community largely supported the new chip, emphasizing its potential for AI inference and HPC. Users debated the chip's actual manufacturing location, with most agreeing it was likely fabbed outside Japan, though designed domestically. There was discussion about Japan's strategy for technological sovereignty, with some viewing it as a move toward reducing dependence on US and Taiwanese supply chains. Community members also discussed the overall Japanese tech industry, the importance of software support, and the long history of Fujitsu and supercomputing. Some critics doubted the chip's market impact outside Japan, citing the crowded market and challenges in software ecosystem development.

Overall, the sentiment was positive about Japan's move into high-end CPUs, though with recognition of the complex geopolitics and supply chain nuances involved.

Title, Astra for Law, Source, openai.com. The article discussed OpenAI's new product, Astra for Law, which aims to help legal professionals automate and streamline legal research and document review processes. The system uses large language models to perform tasks that traditionally required significant manual effort, claiming to increase efficiency and reduce costs. The post highlighted that Astra for Law can search vast legal databases and analyze case law with a focus on the US legal system, but it also acknowledged limitations such as hallucinations and accuracy concerns. The product is mainly targeted at legal professionals, not laypeople, and community comments raised questions about reliability, the impact on legal jobs, and ethical considerations. In the comments, the consensus was mostly skeptical, with many users concerned about the accuracy and potential misuse of AI in law. Some highlighted that while Astra for Law might improve productivity for lawyers, it won't fully replace the nuanced judgment of experienced professionals. Debates centered around the risks of hallucinations, the ethical implications of AI in legal decisions, and whether the technology could democratize or further concentrate legal power.

A recurring theme was worried that AI might increase legal quantity rather than quality, leading to more disputes, misapplications and systemic issues rather than genuine justice. The community generally doubted the current AI capabilities, emphasizing the importance of human oversight and experience in legal work.

Title, Bend, a language that blocks AI mistakes via proof on CPU and GPU. Source bendlang.com. The post introduces Bend, a fast programming language designed to prevent AI errors using formal proofs. It runs on CPU and GPU, compiling to native code with performance comparable to C and CUDA. Bend employs type checking through proof verification to ensure AI-generated code adheres to declared laws, making it impossible to violate specified rules. The post provides installation instructions, explains core features like parallelism, proof enforcement and the use of laws and emphasizes the language's potential for bug-free fast applications. In the comments, the community mostly expressed skepticism about Bend's claims and implementation transparency. Several users questioned the legitimacy of the repository, criticizing the squashing of commit history and the lack of detailed documentation or empirical evidence supporting Bend's performance benefits. Debates focused on whether the proof system and performance metrics were genuinely credible, with many emphasizing the importance of transparent development and formal validation. Despite these concerns, some acknowledged the fascinating ideas behind Bend and showed interest in experimenting with it, though skepticism about the project's claims and methodology remained the dominant tone.