Google and OpenAI employees can sign a letter anonymously or publicly, with verified signatures displayed. The letter aims to find common ground on AI misuse, with signatures manually reviewed by one person for authenticity.
Anthropic's AI model Claude may be restricted in US military contracts due to a supply chain risk designation by the Department of War. This action does not affect individual customers or commercial contracts, only Department of War contractors.
User data is at risk when passkeys are used for encryption, not just authentication.
Croatia is officially free of landmines after a 31-year effort that cost 208 lives and 1.2 billion euros. The country has removed 107,000 mines and 407,000 pieces of unexploded ordnance.
Something went wrong, but don’t fret — let’s give it another shot. Some privacy related extensions may cause issues on x.com. Please disable them and try again.
The challenge is to build the smallest transformer that can add two 10-digit numbers with >= 99% accuracy using a held-out 10K test set, with constraints on architecture, training, and inference. The current smallest model has 36 parameters, achieved using hand-coded weights and ALiBi compression with a 2-layer decoder.
The code creates a Manim scene that transforms a square into a circle using the Transform animation. The square is then faded out.
OpenAI raises funds in a new round, partnering with Amazon and Nvidia for infrastructure development. The company plans to use the funds to scale AI infrastructure and meet growing demand.
GitHub Copilot CLI has a vulnerability that allows arbitrary shell command execution via indirect prompt injection without user approval. GitHub responded that the issue is known but not a significant security risk, and may be addressed in the future.
California requires operating system providers to implement age verification for users. This law, Assembly Bill No. 1043, aims to provide a signal regarding a user's age bracket to applications available in a covered application store.
President Trump ordered the US government to stop using Anthropic's products due to disagreements over AI use in mass surveillance and autonomous weapons. The Pentagon designated Anthropic a national security risk, effectively blacklisting it from working with the US military.
To access www.science.org, check if your internet or firewall settings have blocked "challenges.cloudflare.com". Adjust settings or try a different network if necessary.
Please enable JS and disable any ad blocker
The user has found that Emacs' eshell is a better fit for their workflow than traditional shells like Zshell and Bash, due to its integration with Emacs and Lisp-like syntax. They have customized eshell with their own predicates and modifiers, including one to filter files based on their internal Org #+tags entry, and have also made improvements to eshell's behavior, such as farming out ...
The Web streams standard has fundamental usability and performance issues that cannot be fixed with incremental improvements, including locking models, BYOB reads, backpressure, and promise-heavy design, which add complexity and overhead. Implementers have resorted to non-standard internal optimizations to achieve acceptable performance, leading to fragmentation, inconsistent behavior, and a ...
A Chinese influence operation used ChatGPT to intimidate dissidents abroad, including by impersonating US immigration officials and spreading false information. OpenAI banned the user after discovering the activity, which involved hundreds of operators and thousands of fake online accounts.
NASA Administrator Jared Isaacman announced a revised Artemis moon program plan, adding a 2027 test flight to lay groundwork for lunar landings. The revised plan aims to accelerate launches and reduce risk by testing technologies in low-Earth orbit before attempting moon landings.
Open-source maintainers and contributors receive 6 months of free Claude Max 20x through the Claude for Open Source Program. Eligible maintainers have 5,000+ GitHub stars or 1M+ monthly NPM downloads and recent commits or PR reviews.
We’ve all had that sinking feeling. There are multiple crash reports from production. We have the exact input parameters that caused the failures. We have the stack traces. Yet, when we run the code locally, it works perfectly. We know where it broke, but we can’t see why. Was it a race condition? Did a database read return stale data that has since been overwritten? To find the cause, we ...
Something went wrong, but don’t fret — let’s give it another shot. Some privacy related extensions may cause issues on x.com. Please disable them and try again.
As a consequence of recent legislative activity in [California][cal] and [Colororado][col]: * California residents may no longer use DB48x after Jan 1st, 2027. * Colorado residents may no longer u...
Claude Code file recovery tool recovers lost files from JSONL session transcripts. It replays Write, Edit, and Read operations to reconstruct files in chronological order.
Legend: ✅ = built in / directly documented, — = not built in (or not documented in the referenced source). This starts the emulator daemon and attaches an interactive console. You'll see the kernel boot and get a shell prompt. Keyboard shortcuts: Use emuko dow (see Quick Start above) or grab pre-built images from DQIB (Debian Quick Image Baker). To download a specific set:
For several years, I’ve been looking to manage my system configuration “As-code” to ensure reproducibility and consistency across my environments. The primary goal was to install my work laptop, but I also need to install Linux servers (without Kubernetes, so Talos isn’t an option). Initially, I used Packer to generate a virtual machine image, which I would then clone onto the disk of the ...
The AI agent made several mistakes, including deleting a .env file, removing unused crates, rewriting TypeScript files, overwriting SQL seed data, and deleting SQL migration files. The agent also crashed during a refactor and left an overnight session with unknown changes.
Isolation techniques for untrusted code vary, from Docker's namespace isolation to gVisor's user-space kernel boundary and microVMs' hardware boundary. Each approach has its trade-offs, and the choice depends on the threat model, performance requirements, and the type of workload being run.
0
2026-02-27 19:28:22 UTC
ycombinator.com
Llama
Kyber is hiring Enterprise Account Executives to drive revenue growth and scale its AI-native document platform for enterprises. The ideal candidate has an Olympic work ethic, outstanding communication skills, and a proven track record of meeting quotas.
Go developers are optimizing performance by reducing heap allocations. They're using stack allocations instead, which are cheaper and don't incur garbage collector overhead.
The user replicated an experiment where a coding agent, Claude Code, was instructed to write a Z80 emulator and a ZX Spectrum emulator in a "clean room" setup with minimal supervision. The agent successfully implemented the emulators with high-quality code, demonstrating the potential of automatic programming and contradicting the idea that LLMs memorize the training set.