Generated about 20 hours ago.
VibeOS: First ever AI-native operating system (vibeos.sh) AI
VibeOS is presented as an “AI-native” operating system that uses an Anthropic Claude-powered agent (Claude Code) to control the computer from prompts, enabling instant creation of apps and tools like live-edit NextJS UI, MCP-based utilities, browser handoff to the AI agent, and AI-curated news feeds, with a Dockerized option aimed at privacy by not giving the agent hardware access by default.
Agentic AI solved coding and exposed every other problem in SE (venturebeat.com) AI
The article argues that agentic AI can solve coding tasks but also highlights broader software engineering problems that still need addressing.
Data centers consumed 264B gallons of water as drought hits nearly 63% of US (barchart.com) AI
The article says drought affecting nearly 63% of the U.S. is intensifying water conservation warnings while AI hyperscale data centers consume massive amounts of water—estimated at about 264 billion gallons in 2025 (roughly 1 trillion liters), according to Mordor Intelligence.
What Are Tokens in LLMs? (bearisland.dev) AI
The article explains that in LLMs, text is converted into model-specific integer token IDs rather than raw characters or words, using tokenizers built from algorithms like (byte-level) BPE. It walks through how BPE incrementally builds a vocabulary by repeatedly merging frequent adjacent pairs, including an example showing how words like “cat” can become single tokens. It also clarifies the “strawberry” effect—models may split a word differently because their vocabularies differ, and byte-level tokenization avoids out-of-vocabulary characters by starting from UTF-8 bytes.
Billions spent and hypothetical returns: the AI boom explained with six charts (theguardian.com) AI
A Guardian explainer uses six charts to argue that the AI boom is driven by rapid spending on datacentres, chips and other infrastructure alongside accelerating corporate and consumer adoption, even as critics warn of rising costs, infrastructure bottlenecks and potentially dangerous market valuations.
The week I lost the plot at a startup where the tools worked as advertised (nlkw.de) AI
An employee at Cogentiv.ai describes how a culture of reflexive AI use and “tool trust” spiraled into a loss of understanding and a staging incident where an automated notification service sent thousands of emails about events that hadn’t happened yet, leaving them unable to explain what they themselves had shipped.
Leiden Declaration on Artificial Intelligence and Mathematics (lms.ac.uk) AI
The London Mathematical Society has published the Leiden Declaration on Artificial Intelligence and Mathematics, developed from a 2025 workshop, addressing how AI is being used in math research—such as formalising proofs—along with concerns about reliability, attribution, and impacts on publishing and peer review. The document recommends actions for individual researchers, professional bodies and funders, and policymakers, including disclosing AI use, ensuring correctness, developing publication/review policies, and considering regulation and public investment.
My automated doubt development process (alexself.dev) AI
Alex Self describes an “automated doubt” workflow for AI-assisted development that regains trust by front-loading multi-agent scrutiny of specs and code, repeatedly running validation, security, and interface checks until issues converge to the author’s readiness threshold.
Agents, Agile, Communism, Coercion (elliotmorris.net) AI
Elliot Morris argues that “agentic/agile” AI systems and even communally minded economic models will fail if designers ignore real human capabilities and integration needs, and he suggests that achieving such systems may require coercion—potentially authoritarian—to make people comply.
The ROI of AI coding looks different when you are a bootstrapped founder (octavio.substack.com) AI
The piece argues that AI coding’s ROI looks different for bootstrapped founders because it effectively creates “capacity” to ship and iterate features faster when time and engineering budget are limited, whereas for large companies it can mainly mean higher, harder-to-track software costs.
The OnlyFans Economy of American AI (leoveanu.com) AI
The piece argues that parts of the U.S. AI market operate like an “OnlyFans economy,” driven by hype, high-priced rate limits, and valuation pressures rather than evidence-based performance, using examples from Anthropic/OpenAI spending and pricing. It claims U.S. frontier models have hit a “plateau,” and highlights Qwen 3.7 Max (via a $100/$100K-credits plan) as an economical alternative that provides extended “thinking” and access to other model providers.
7 Ways New Engineers Can Flourish in the Age of AI (spectrum.ieee.org) AI
IEEE Spectrum advises new engineers to thrive in the AI era by prioritizing core fundamentals, learning to use AI as a productivity partner (with judgment and debugging), building end-to-end projects, strengthening system design and communication skills, staying continually curious, and focusing on problem-framing, architectural judgment, and ethical awareness as routine coding becomes automated.
Show HN: Lathe – Use LLMs to learn a new domain, not skip past it (github.com) AI
Show HN highlights Lathe, a GitHub project that uses LLM “skills” to generate hands-on, multi-part technical tutorials from prompts, then provides a local UI and workflow for learners to work through the material themselves (rather than having the model do the thinking). The project describes a Go-based CLI for storing and serving tutorials, integration with tools like Claude Code/Cursor/Codex, and features such as source documentation, tutorial verification/extension, and built-in library management to reduce hallucination risk by keeping users actively typing and asking questions.
Anthropic/OpenAI may be spending more than $1000 for every $100 you pay them (ea.rna.nl) AI
The article argues that while coding with LLMs like Anthropic’s Claude Code can be impressive, the underlying token costs can make it economically unviable for sustained, agentic coding—estimating that model usage could amount to more than $1,000 in API costs for what users pay about $100 per month.
Anthropic, please ship an official Claude Desktop for Linux (github.com) AI
A GitHub issue asks Anthropic to provide an official Claude Desktop for Linux (not just the CLI), arguing that Linux users currently lack an officially supported GUI/extension testing environment and resort to third-party repackages. The requester points to Claude Code’s existing signed Linux package pipeline and to Claude Cowork’s Linux execution inside macOS/Windows setups as evidence the capability already exists, while proposing an official Ubuntu LTS/Debian .deb distribution and asking for either a roadmap commitment or a clear explanation for why Linux isn’t planned.
Automated QA and Testing with AI (antirez.com) AI
The article argues that while AI-assisted programming can trade off structure and efficiency, LLMs can substantially improve software QA by letting an AI agent act like a manual QA engineer—specializing tests based on new commits and running integration-style checks such as distributed inference coherence and speed-regression detection.
LLMs are eroding my software engineering career and I don't know what to do (human-in-the-loop.bearblog.dev) AI
A 10-year software engineer argues that rapid advances in LLMs are eroding the value of their finance/payments domain expertise and especially their debugging and distributed-systems know-how, making them more like an “off-the-shelf” agent reviewer than a differentiated specialist.
Efficient and Training-Free Single-Image Diffusion Models (arxiv.org) AI
The paper proposes a “training-free” single-image diffusion approach that models an input image using a finite dataset of its multi-scale patches, enabling an analytic closed-form denoiser instead of neural network training and improving generation quality and diversity; it demonstrates applications such as unconditional generation and text-guided stylization, and reports accelerations for megapixel to gigapixel outputs.
Why Aren't We Measuring How AI Affects Humans? (spectrum.ieee.org) AI
IEEE Spectrum reports that Imran Khan argues AI evaluation is overly focused on technical performance while under-measuring downstream psychosocial effects on humans, warning that harms could take months or years to emerge and urging longer-horizon, human-outcome metrics and data access for external researchers.
Speculative KV coding: losslessly compressing KV cache by up to ~4× (fergusfinn.com) AI
The post proposes “Speculative KV coding,” a lossless method that compresses an LLM’s KV cache by using a cheaper predictor model to estimate each KV scalar’s value and uncertainty, then arithmetic-coding the exact target cache based on how well the predictor fits; experiments on Qwen3 suggest up to ~4× lossless compression (on top of ~8× from FP8 cache compression).
Arithmetic Without Numbers – How LLMs Do Math (alvaro-videla.com) AI
The article explains how large language models can produce exact arithmetic despite having no human-style “variables” or scratch space, by describing internal mechanisms like residual streams, attention, and next-token generation—then discusses experiments showing how numeric continuations fail at carry boundaries due to token/chunk coordination limits.
Human-Like Neural Nets by Catapulting (gwern.net) AI
Gwern.net proposes a speculative “catapulted” training paradigm for overparameterized neural nets that uses very high learning rates and regularization on small, filtered datasets to jump into a basin of human-like generalization, with claims of improved robustness to adversarial examples and better alignment prospects. The article frames this as a bias–variance tradeoff idea—LLMs minimizing variance while human brains may minimize bias—and discusses related anomalies such as sample inefficiency, why active learning/embodiment don’t fully explain human learning, and why current architectures or learning rules haven’t yielded clear “magic” from biology.
I design with Claude more than Figma now (blog.janestreet.com) AI
Edwin Morris of Jane Street says he now uses Claude more than Figma by writing a problem description, prompting Claude in an editor to rapidly build and iterate working prototypes in the real codebase, and then collecting user feedback—shifting effort from mockups and spec docs to functional artifacts. He notes benefits like fast, unlimited iteration on features (e.g., adding LLM prompting to an internal SQL input) and easier evaluation by engineers and users, while also acknowledging downsides such as reviewers receiving “fully baked” code and a concern that the workflow may limit more open-ended creativity for novel problems.
AI is the greatest money-wasting scheme humanity has ever invented (telegraph.co.uk) AI
The Telegraph argues, based on the article title, that AI is an ineffective and wasteful use of resources, framing it as a major “money-wasting scheme.”
Show HN: TakoVM – Isolated model and tool execution used by enterprises (github.com) AI
Show HN introduces TakoVM, a self-hosted system for isolated AI workload execution that bundles Docker-based sandboxing (with optional gVisor) plus built-in job queueing, retries, and persisted execution history in PostgreSQL.