AI news

AMD AI director says Claude Code is becoming dumber and lazier since update (theregister.com) AI

AMD AI director Stella Laurenzo says her team’s long-running use of Anthropic’s Claude Code has degraded since February, with evidence from thousands of sessions suggesting the tool is ending early, avoiding ownership, and reading less before making edits. She links the changes to “thinking” content redaction introduced around Claude Code version 2.1.69 and asks Anthropic for transparency about whether it is reducing or capping thinking tokens and to expose thinking-token counts per request. Laurenzo says her team is switching to another provider and warns Anthropic may lose ground unless the behavior is fixed.

The AI Great Leap Forward (leehanchung.github.io) AI

The article argues that many corporate “AI transformation” efforts mirror China’s Great Leap Forward: top-down mandates push teams to ship impressive but poorly validated “AI” outputs, while metrics and incentives encourage inflated claims. It warns that eliminating people and processes (middle managers, QA, documentation, operations knowledge) creates second-order failures once real-world edge cases appear, and that attempts to “distill” expertise into agent skills can backfire by making workers strategically indispensable. Overall, it calls for evaluation, data, monitoring, and maintainability rather than demos and paperwork.

Show HN: Skrun – Deploy any agent skill as an API (github.com) AI

Skrun is an open-source tool that turns an “agent skill” (defined in SKILL.md) into a callable API endpoint using a POST /run interface. It supports typed inputs/structured outputs, multi-model backends with fallback (e.g., Anthropic, OpenAI, Google, Mistral, Groq), and stateful agent runs via stored key-value data. The project ships with a local runtime and includes CLI commands to init, develop, test, package, and deploy agents, with an architecture intended for cloud deployment.

Databricks co-founder wins prestigious ACM award, says 'AGI is here already' (techcrunch.com) AI

Databricks co-founder and CTO Matei Zaharia won the ACM Prize in Computing, with the award highlighting his work on Spark and related contributions. He said he believes AGI is already here in a limited form, arguing people should avoid judging AI by human standards. Zaharia also discussed how AI could better support research and engineering, while warning that current agent systems can introduce security risks.

The demise of software engineering jobs has been greatly exaggerated (cnn.com) AI

CNN argues that claims that AI will wipe out software engineering jobs are overstated. The article says developer postings are still rising and that AI is changing what engineers do—shifting work from routine coding to overseeing AI-generated code and focusing more on design and customer problems. It also notes a short-term transition that may be difficult for some workers, as companies cut costs and require engineers to keep learning new skills.

Show HN: Voxcode: local speech to text and ripgrep = transcript and code context (github.com) AI

Voxcode is an open-source macOS app for local speech-to-text tailored to coding agents: you select code in your editor, speak instructions, and it pastes the transcript back with a ripgrep-style file/line reference (or the selected snippet when exact lines can’t be resolved). The project indexes local git repositories and uses an optimized parallel file-walking approach plus a local ONNX Runtime transcription model (Parakeet TDT) to keep searches and transcription fast. It’s designed to work across IDEs and agent tools without direct integration by operating purely on clipboard/paste and filesystem context.

Claude Managed Agents (claude.com) AI

Anthropic announced Claude Managed Agents, a set of composable APIs meant to help developers deploy cloud-hosted AI agents faster by handling production concerns like secure sandboxing, long-running sessions, permissions, and tracing. The company says teams can move from prototype to launch in days instead of months and that Managed Agents are available in public beta on the Claude Platform, with multi-agent coordination in research preview.

Façade (2005 Video Game) (en.wikipedia.org) AI

Façade is a 2005 interactive drama in which the player converses via text with an AI-driven married couple, Trip and Grace, in an open-ended story about their deteriorating relationship. Built using natural-language processing and an AI “behavior language,” the game was praised at launch for its conversational design and storytelling ambitions, and later developed a cult following fueled by awkward moments from its reactions. A planned sequel, The Party, was paused after 2013 and later resumed in 2024.

A Digital Compute-in-Memory Architecture for NFA Evaluation (dl.acm.org) AI

The paper proposes a compute-in-memory digital architecture aimed at evaluating nondeterministic finite automata (NFA) more efficiently by performing computation closer to where data is stored, reducing data movement overhead compared with conventional approaches.

Show HN: TUI-use: Let AI agents control interactive terminal programs (github.com) AI

The GitHub project tui-use ("Show HN") proposes a way for AI agents to operate interactive terminal programs by running them in a PTY, rendering the screen with a headless xterm emulator, and sending keystrokes based on clean screen snapshots (including TUI selection “highlights”). It targets use cases like REPL sessions, CLI wizards, database CLIs, SSH-driven interactive workflows, and full-screen TUIs such as vim/htop/lazygit, with a command-line interface plus plugins for agents like Claude Code. The article notes it works on Unix-like systems and strips most terminal styling, relying on metadata to identify active selections.

Digital Hopes, Real Power: How the Arab Spring Fueled a Global Surveillance Boom (eff.org) AI

The EFF argues that the 2011 Arab uprisings’ digital tactics spurred a global surveillance industry: governments upgraded monitoring, expanded cybercrime and protest-related laws to criminalize dissent, and relied on spyware markets to hack targets at scale. It also describes how biometrics, facial recognition, and “smart city” systems helped normalize automated tracking and risk profiling, including in migration and humanitarian settings. The piece warns that these tools and legal frameworks—often sold without meaningful safeguards—have been exported beyond the Middle East to support digital authoritarianism worldwide.

Meta debuts Muse Spark, first AI model under Alexandr Wang (axios.com) AI

Meta has launched Muse Spark, a new homegrown AI model (code-named Avocado) built over nine months under Alexandr Wang’s leadership, aimed at narrowing performance gaps with top rivals. The text-only model takes voice, text, and images as inputs and will power queries in the Meta AI app and on Meta.ai, with plans to expand across Facebook, Instagram, and WhatsApp, alongside several “modes” including a shopping mode. Meta says all versions will be free with possible rate limits and plans an open-source licensed release, while noting privacy rules may allow broad use of data shared with its AI systems.

Muse Spark: Scaling Towards Personal Superintelligence (ai.meta.com) AI

Meta introduced Muse Spark, a multimodal reasoning model designed to support tool use, visual “chain-of-thought,” and multi-agent orchestration. The company says it will be available on meta.ai and via a private API preview, with “Contemplating mode” rolling out gradually. Meta also outlines how it improved compute efficiency through changes to pretraining, reinforcement learning, and test-time reasoning, and it reports safety evaluations showing strong refusal behavior in high-risk scientific domains.

Muse Spark – Meta Superintelligence Labs (meta.ai) AI

The linked page appears to introduce “Muse Spark” and Meta’s “Superintelligence Labs,” describing the lab’s mission and related research efforts, though no specific article text was available to summarize in detail.

Show HN: We fingerprinted 178 AI models' writing styles and similarity clusters (rival.tips) AI

Researchers report a study that fingerprinted the writing styles of 178 AI models and identified similarity clusters among them, aiming to characterize how models differ in generated text.

The Future of Everything is Lies, I Guess (aphyr.com) AI

The author argues that today’s AI—especially large language models—is less like human-like intelligence and more like a “bullshit machine” that statistically imitates text while frequently confabulating, misunderstanding context, and making factual errors. They describe why LLMs can’t reliably reason about their own outputs, how “reasoning traces” and generated explanations can be misleading, and why models can be both astonishingly capable and still repeatedly “idiotic” in practical tasks. Overall, the piece frames current AI progress as creating major real-world risks alongside potential benefits, without offering a single definitive prediction of the future.

LLM plays an 8-bit Commander X16 game using structured "smart senses" (pvp-ai.russell-harper.com) AI

Russell Harper describes bringing his 1990 8-bit PvP-AI game to the Commander X16 (first running well in an emulator but slower on real hardware due to a rendering issue). He also explains how an LLM can play the game using structured “smart senses” instead of raw visual input, with the game converted to turn-based and equipped with text-based state inputs. The write-up covers the PHP-to-emulator integration and reports a progression in LLM strategies across recorded games.

MegaTrain: Full Precision Training of 100B+ Parameter LLMs on a Single GPU (arxiv.org) AI

MegaTrain is a proposed training system that enables full-precision training of 100B+ parameter LLMs on a single GPU by keeping model parameters and optimizer states in CPU host memory and streaming them layer-by-layer to the GPU for computation. The method uses double-buffered pipelining to overlap parameter prefetching, gradient computation, and offloading, and it avoids persistent autograd graphs via stateless layer templates. Reported results include training up to 120B parameters on an NVIDIA H200 with 1.5TB of host memory, and improved throughput versus DeepSpeed ZeRO-3 with CPU offloading on smaller models.

Mario and Earendil (lucumr.pocoo.org) AI

Armin Ronacher announces that Mario Zechner is joining the Earendil team, praising Pi as a thoughtful, quality-focused agent infrastructure and contrasting it with the industry’s rush for speed. He links the hire to concerns about AI systems producing “slop” and degradation, and describes Earendil’s Lefos effort to build more deliberate tools that improve communication and human relationships. Ronacher says he and Colin want to steward Pi as high-quality, open, extensible software while clarifying how it may relate to Lefos.

Multi-agentic Software Development is a Distributed Systems Problem (AGI can't save you) (kirancodes.me) AI

The post argues that multi-agent software development with LLMs is fundamentally a distributed systems coordination problem, not something that “smarter agents” will eliminate. It models prompt-driven code synthesis and agent collaboration as an underlying consensus task constrained by an underspecified natural-language spec, then relates the setting to classic impossibility results like FLP (showing limits on deterministic consensus under async delays and possible crashes) and discusses possible parallels to failure detectors. The author concludes that building scalable tooling/languages for agent coordination remains necessary even if future models become extremely capable.

The Downfall and Enshittification of Microsoft in 2026 (caio.ca) AI

The article argues that Microsoft’s 2026 “enshittification” is driven by shifting focus from core product quality to aggressive, AI-centered Copilot integration across Windows, Office, and GitHub. It points to Windows 11 promises to fix long-standing desktop usability issues, recurring complaints and outages affecting GitHub’s developer workflows, and the perceived tradeoff between reliability and Copilot placement. The author also suggests competitive pressure from Apple’s lower-cost MacBook Neo and Linux’s gradual desktop legitimacy is making Microsoft’s strategy look less like leadership and more like defensive retrenchment.

I've Sold Out (mariozechner.at) AI

Mario Zechner says he has joined the Earendil team and will “take pi” as a coding agent, explaining his history of OSS-to-commercial transitions and the pain he saw when key projects like RoboVM went closed-source after being sold. He describes growing interest from VCs and large companies in pi, but says he does not want to run a VC-funded company focused only on pi, prioritizing family time and avoiding the stress and community-betraying dynamics he experienced before. The post also recounts how Zechner met Armin and others in the “Vienna School of Agentic Coding” circle and how collaboration around agentic coding led to this decision.

Open Models have crossed a threshold (blog.langchain.com) AI

LangChain reports early Deep Agents evaluations showing open-weight models such as GLM-5 and MiniMax M2.7 can match closed frontier models on core agent abilities like file operations, tool use, and instruction following. The post emphasizes lower cost and latency, and describes how their shared eval suite and Deep Agents harness let developers compare and swap models across providers with minimal code changes.

An Arctic Road Trip Brings Vital Underground Networks into View (quantamagazine.org) AI

A Quanta Magazine field report follows biologist Michael Van Nuland and colleagues as they sample Alaskan tundra to test machine-learning predictions about rare mycorrhizal fungal “hot spots.” The article describes how underground fungal networks connect to plant roots, exchanging nutrients and carbon, and how recent imaging and robotic tracking suggest the fungi actively regulate this system rather than merely serving plants. Because these networks help store vast amounts of carbon in permafrost but are vulnerable to warming, wildfires, and thaw, the researchers argue that better mapping and protection of soil biodiversity could matter for climate resilience.

Japan relaxes privacy laws to make itself the 'easiest country to develop AI' (theregister.com) AI

Japan has approved amendments to its Personal Information Protection Act to remove the usual opt-in consent requirement for organizations using low-risk personal data for statistics and research, aiming to speed AI development. The changes include provisions for some sensitive categories such as health-related data (for improving public health) and facial images, with additional conditions like parental approval for children under 16 and stricter requirements around handling facial data. The rules add penalties for fraudulent data acquisition and improper use, but reduce requirements to notify individuals after data leaks deemed unlikely to cause harm.