Bend 2 is a new programming language designed to be both extremely fast and formally unambiguous so humans can state intents that AIs must implement correctly. It combines a high-performance compiler (targets: C, CUDA, Metal, JavaScript) with an affine dependent type system, purity and linearity to generate code that rivals hand-written C on CPU and scales across thousands of GPU cores. The toolchain includes a fast proof checker that claims orders-of-magnitude speedups over existing assistants, enabling practical verification of program invariants. Built-in parallelism requires no explicit threads or locks: the runtime splits work automatically across available cores and GPUs.
Correctness is enforced by LAWS.bend and PROOF.bend: developers declare invariants as laws and the compiler demands machine-checkable proofs whenever code changes, preventing AI or human edits from violating those guarantees. Examples show laws like “players can never pass through walls” or “sum of balances is zero,” with proofs written or generated by agents to unblock merges. Installation is simple, but the system is young and limited: no backward compatibility with Bend 1, verbose annotations, few numeric types, no advanced libraries (TLS/HTTP/JSON), restricted platform support, no rich tooling or proof automation, and a largely AI-written compiler that hasn’t been fully audited.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
Orbital is an open source project that provides an open source version of the Claude AI model, allowing users to retain ownership of their data and context. It aims to offer an alternative to proprietary AI models with transparency and control. (github.com)
ShapeLearn's full models for Qwen 3.8 27B outperform the Lite versions in quality and speed benchmarks, with all five models on the performance frontier. Using GPU-5, these models achieve up to 99.63% of BF16 scores, with options for faster, smaller models depending on memory constraints. (byteshape.com)
Claude Code from Source details the architecture and design of Anthropic's widely used AI coding agent, covering components like the agent loop, tool execution, and memory management. The book provides technical insights for engineers, leaders, and anyone interested in production AI systems. (claude-code-from-source.com)
Jev is a new AI model that produces only structured output, making it faster and more consistent than traditional LLMs. Its speed enables real-time applications like playing Doom and opens possibilities for new types of programming tasks. (seangoedecke.com)
Alibaba has launched Qwen 3.8 Omni Flash, an upgraded version of its AI language model. The release aims to enhance AI capabilities and performance. (qwen.ai)
Huawei's Ascend 960 SuperPoD uses NPO technology to connect multiple AI computing nodes, addressing data exchange bottlenecks in large-scale models. It supports up to 4,096 NPU cards and offers significant computing power for next-generation AI infrastructure. (technode.com)
Today's best Hacker News stories, summarized and screenshotted, one email a day.