How to Get Hired in 2025

In a hypothetical 2025, a software engineer's test assignment might be rejected for appearing AI-generated, even if human-made. The author highlights that many standard good programming practices—like fully understanding and implementing requirements, using industry-standard tools, writing clean, modular, and well-commented code with descriptive names, robust error handling, reasonable organization, a polished UI, and comprehensive tests—could be mistakenly perceived as "AI giveaways" due to the increasing sophistication of AI.

Read more

The Reverse Centaur's Guide to Criticizing AI

The author argues that the current AI and LLM boom is a speculative bubble driven by tech monopolies to sustain growth, not by genuine technological breakthroughs. He contends that AI is primarily deployed to create "reverse centaurs"—humans serving machines—by displacing high-wage workers, with humans becoming "accountability sinks" for AI's subtle errors. The text criticizes expanding copyright for AI training, advocating instead for the US Copyright Office's position that AI-generated works are uncopyrightable and promoting sectoral bargaining for workers. Ultimately, the author predicts the bubble's collapse will yield cheap GPUs and useful open-source AI tools, urging focus on the economic forces fueling the bubble.

Read more

Grok 4.20 beats all other AI models in Alpha Arena test

Google is adjusting free access to its Gemini 3 Pro and Nano Banana Pro AI tools due to high demand, while simultaneously expanding Gemini's integration and capabilities across its ecosystem. Upcoming features include direct image annotation within Gemini for enhanced editing, and a new "Circle Screen" tool to streamline on-screen search, potentially replacing Circle to Search. Additionally, Gemini is replacing Google Assistant across all Google Maps navigation modes, and Circle to Search now leverages AI Mode for more intelligent follow-up queries.

Read more

Running With Scissors cancels game over AI-generated assets, days after reveal

Running With Scissors cancelled Postal: Bullet Paradise two days after its reveal, citing "extreme damage" to its brand and reputation due to the game's "flagrant use of generative AI" assets. The publisher stated that significant negative community feedback and broken trust with developer Goonswarm Games over AI integration led to the project's termination.

Read more

Investigating a Possible Scammer in Journalism's AI Era

An editor uncovered a widespread journalistic fraud by "Victoria Goldiee," a freelancer who submitted numerous AI-generated articles with fabricated interviews and bylines to prestigious publications. This case exemplifies how LLMs can exploit vulnerabilities in the media industry, where overworked editors and reduced fact-checking make it difficult to detect sophisticated AI-powered content fraud. The incident underscores the increasing challenge of content provenance and authenticity in the age of generative AI.

Read more

Zebra-Llama – Towards efficient hybrid models

Zebra-Llama introduces a family of 1B, 3B, and 8B hybrid LLMs designed to improve inference efficiency without costly retraining. It combines SSMs and MLA layers, transferring knowledge from pre-trained Transformers using only 7-11B training tokens. This approach achieves Transformer-level accuracy with near-SSM efficiency, dramatically reducing KV cache size (down to 2-3.9%) while preserving high zero-shot performance and delivering superior throughput compared to other hybrid models.

ArXiv

High-Performance DBMSs with io_uring: When and How to use it

This study investigates how modern database systems can leverage Linux's io_uring interface for efficient, low-overhead I/O. io_uring is an asynchronous system call batching interface that unifies storage and network operations, addressing limitations of existing I/O methods. The research evaluates its benefits in storage-bound buffer managers and network-bound data shuffling, analyzing advanced features like registered buffers. It provides practical guidelines for designing I/O-intensive systems, validated by a 14% performance improvement in PostgreSQL's io_uring integration.

ArXiv

When does a physical system compute? (2013)

This paper introduces a formal framework to determine if a physical system is performing computation, addressing the lack of consensus in defining computation, especially for non-standard systems. It defines physical computing as the use of a physical system to predict the outcome of an abstract evolution, drawing parallels between abstract computational and physical device levels. The framework provides conditions for computation, defines 'computational entities', and covers broader contexts beyond human users.

ArXiv

Quantum theory does not need complex numbers

This work refutes the recent assertion that quantum theory fundamentally requires complex numbers, demonstrating that a real-number quantum theory is consistent with its postulates. The proposed real-number theory retains representation locality, leading to the conclusion that real and complex quantum theories are experimentally indistinguishable.

ArXiv

Individual and Team Drivers of Developer GenAI Tool Use

Uneven developer adoption of generative AI tools in software engineering hinders productivity and creates uncertainty. Research with 54 developers revealed usage differences are driven by tool perception (collaborator vs. feature), engagement approach (experimental vs. conservative), and challenge response (adaptive persistence vs. quick abandonment). This leads to a "Productivity Pressure Paradox," where organizational expectations for rapid gains without sufficient learning support undermine actual productivity benefits.

ArXiv

Advent of Code 2025: The AI Edition – By Peter Norvig

Peter Norvig's "pytudes" is a collection of short Python programs and Jupyter notebooks designed to refine programming skills. For an AI/LLM-focused audience, the project includes studies on leveraging LLMs for solving coding challenges like Advent of Code, assessing their "theory of mind" in logic puzzles, and analyzing their efficacy in code generation and natural language tasks such as generating word game clues. It also features an analysis of AlphaCode's automated programming.

Repo

A fork of Calibre called Clbre, because the AI is stripped out

clbre is a fork of the cross-platform e-book manager calibre, specifically designed to strip out its AI integration. calibre provides functionalities for viewing, converting, editing, and cataloging e-books, along with fetching metadata and downloading news.

Repo

Show HN: SFX – A language where 0.1 and 0.2 = 0.3 and Context is first-class

SFX (Situation Framework eXchange) is a Rust-based, standalone Context-Oriented Programming language that natively supports Situation and Switch constructs, enabling objects to dynamically adapt their behavior based on context, unlike traditional OOP. It features a Python-like syntax, a tree-walking interpreter with Cranelift-powered JIT compilation, reactive When observers, and async primitives. Its standard library includes an LLM module for OpenAI API integration, alongside modules for networking, data parsing (JSON, XML), and file I/O, making it suitable for AI-related development.

Repo

Show HN: Nano.noq – Experimental key-container format

NANO.NOQ is an experimental, client-side file format (.noq) designed to store AES-GCM keys separately from ciphertext. Its binary structure includes a header, the raw AES-256-GCM key, a SHA-256(key) slice for integrity verification, and random padding. Designed for local browser operation and educational purposes, it also employs a "mutation layer" for light obfuscation of Base64URL-encoded ciphertext, which offers no cryptographic security. The project explicitly states it is not a standard security format and is unsuitable for production, serving as a personal exploration of file format design and key separation.

Repo

Show HN: AgentPG – Stateful AI Agents in Go with PostgreSQL Persistence

AgentPG is a Go toolkit for building stateful AI agents powered by Anthropic's Claude, leveraging PostgreSQL for persistence and transaction-safe operations. It features a streaming-first architecture for long-context support, automatic context management with smart compaction, a robust tool system including nested agents, and extended 1M token context handling.

Repo