News
The latest from the AI agent ecosystem, updated multiple times daily.
Greptile Now Charges Per Review. Nobody Else Does.
Greptile swapped its $30 flat rate for $30 plus $1 per review after 50 reviews. The math doesn't work for agentic workflows, every competitor stays flat, and OSS maintainers are getting billed despite promises of free reviews.
IDLI AI Shows Gene Activity Runs on Volume Dials, Not Switches
Researchers at Gladstone Institutes and Arc Institute used an AI-powered computational method called IDLI to discover that over 85% of nucleosomes contain sections of partially accessible DNA, challenging the binary view of gene activity. While the spectrum concept isn't new in epigenetics, IDLI offers unprecedented resolution to actually measure and visualize these states. The study identified 14 distinct structural states of nucleosomes tied to different gene activity levels, with implications for understanding complex diseases like cancer and aging.
Self-Evolving Harness Beats Human-Designed Codex-CLI by 5 Points
A self-evolving coding agent harness hit 77.0% pass@1 on Terminal-Bench 2, beating the human-designed Codex-CLI (71.9%). The system improves by modifying its own structure, not just prompts. It transfers to SWE-bench-verified without re-evolution and generalizes across model families.
Community Fork Pressures Warp to Open Up AI Provider Choice
OpenWarp is a community fork of Warp that lets you plug in any AI provider you want. DeepSeek, Ollama, OpenAI, Anthropic, local models. Your keys stay local. The fork prompted Warp's founder to publicly acknowledge demand for BYO model support, making this less about the code and more about what happens when users force a vendor's hand.
Languages Follow Same Math Rules Despite Geography, Study Finds
A seven-year study of 22 languages found universal mathematical patterns in vocabulary evolution. Researchers from Fudan, Harvard, and Stony Brook used word embeddings to show that popular words cluster together, vocabulary organizes in hierarchies across languages, new words arrive in bursts, and word distributions follow Taylor's law. A stochastic model replicates these patterns, pointing to shared mechanisms in cultural evolution.
Eka's Robot Arms Convinced Robotics' Biggest Skeptic
Eka, a startup founded by MIT professor Pulkit Agrawal and ex-Google DeepMind researcher Tuomas Haarnoja, has built robotic arms with custom compliance-based grippers that provide real-time tactile feedback. Trained using simulation methods and a vision-force-action model, the robots handle delicate manipulation tasks like screwing in lightbulbs. The approach revives simulation-based training that many considered a dead end after OpenAI abandoned its Dactyl project.
LLM Reverse-Engineers 1994's SimTower in 8-Hour Runs
Patrick Hulin used Claude Code to reverse engineer SimTower and build towers.world, a live multiplayer clone. After static analysis failed, he switched to dynamic analysis with a Unicorn emulator mocking all 195 Windows 3.1 API functions. Claude fixed bugs autonomously in runs up to 8 hours, converging on behavioral parity through state-matching.
SigMap beats RAG pipelines with decades-old math and zero deps
SigMap is a zero-dependency code retrieval tool that extracts function and class signatures from codebases to provide relevant context to AI coding assistants. It achieves 81.1% hit@5 retrieval accuracy and 40-98% token reduction using TF-IDF ranking instead of embeddings, working with Copilot, Claude, Cursor, Windsurf, and any LLM.
Granite 4.1: IBM's 8B Model Matching 32B MoE
IBM released Granite 4.1, a family of open-source language models (3B, 8B, 30B) under Apache 2.0 license. The 8B dense architecture model matches or beats the previous 32B MoE Granite 4.0-H-Small across benchmarks including tool calling (BFCL V3), math (GSM8K), and instruction following (IFEval). Key features include 512K context window, 15T token training, and aggressive data filtering. IBM also got unusually honest about training failures during their four-stage RL pipeline.
Copilot's Getting a Free Byline on Your Git Commits
VS Code's latest release automatically credits GitHub Copilot as a co-author on your git commits, even when you didn't use it. Users are calling out the default behavior for polluting commit history and creating legal headaches around code provenance and copyright.
Zig bans all AI code contributions and explains why
The Zig programming language project enforces a blanket ban on LLM-assisted contributions. VP Loris Cro says they practice "contributor poker," investing in people rather than accepting perfect patches. AI-generated PRs break this model because they skip the relationship-building that sustains open-source communities. Bun, acquired by Anthropic in December 2025, now runs its own Zig fork since their AI-authored changes can't be upstreamed under this policy.
Self-hosted legal AI Mike challenges Harvey and Legora
Mike is an open-source legal AI challenging VC-backed platforms like Harvey and Legora. Users bring their own Claude or Gemini API keys and self-host the tool to keep documents on their own infrastructure. Features include a document-aware chat interface, matter-scoped project workspaces, and tabular review across hundreds of documents with page-level citations.
Diallo's Excel Satire Roasts AI Hype at Its Own Game
Ibrahim Diallo's satirical article parodies AI hype by applying the same exaggerated language to Microsoft Excel. The piece targets Excel's integration with Microsoft Copilot and built-in Python support, arguing that spreadsheets can replace entire business departments. The real target is the AI industry's inflated rhetoric, revealing how absurd startup pitches sound when applied to a humble grid of cells.
Vera ditches variable names because LLMs can't handle them
Vera is a programming language built specifically for LLMs to write and verify code. It replaces variable names with structural references like @Int.0 and @Int.1 to avoid naming-related errors, enforces mandatory contracts verified by the Z3 SMT solver, and makes all effects explicit. Programs compile to WebAssembly. Error messages are designed for machine consumption, each with a stable code, concrete fix, and spec chapter reference. The language sits at v0.0.127 with a reference compiler and 13-chapter specification. VeraBench shows Kimi K2.5 hitting 100% run_correct on Vera versus 86% on Python, though GPT-4.1 and Claude Sonnet 4 both scored lower on Vera than Python. HN commenters question whether stripping semantic information from names actually helps LLMs or makes their job harder.
Amazon chips: from side dish to $20B Nvidia rival
Amazon's semiconductor business surpassed a $20 billion annual run rate, with custom silicon including Graviton processors, Trainium AI training chips, and Nitro security chips growing at over 100% year over year. OpenAI committed to 2 gigawatts of capacity, Anthropic to 5 gigawatts, with Meta and Uber also signing on. Amazon launched Bedrock AgentCore for building AI agents and made GPT-5.4 and Claude Opus 4.7 available on Bedrock.
Claude Outage Hits Coders Only
Claude Code went down while regular chat kept working for many users. The partial outage reveals Anthropic runs these services on separate infrastructure.
Mozilla Says No to Chrome's Built-In AI API
Mozilla flagged Chrome's proposed Prompt API with a 'position: negative' label in their standards-positions repository. The API would let web developers access local language models directly through the browser. Authored by Domenic Denicola and developed by the Web Machine Learning Community Group, it's already in experimental form in Chrome and Edge. Mozilla isn't having it.
Your LLM Is Just a Fancy Probability Engine
Alfredo V. Clemente breaks down how Large Language Models actually work: tokenization, pre-training on massive datasets, and instruction fine-tuning. His framing is simple. LLMs do one thing, predict the next token. Everything else is clever problem reframing.
Every Claude Service Went Down at Once
Every Claude service, including the government offering, went down simultaneously for 31 minutes on April 30, raising questions about shared infrastructure. The outage sparked community discussion about AI reliability and inspired a developer-built waiting room for when Claude goes down.
I benchmarked the Caveman plugin against "be brief."
A technical benchmark comparing Caveman, a popular Claude Code compression plugin, against a simple two-word prompt "be brief." The study found that "be brief." matched Caveman on both token reduction (419 vs 401-449 tokens) and quality scores (0.985 vs 0.970-0.976). Caveman still offers consistent output structure, intensity modes, and an Auto-Clarity safety escape. But the compression itself didn't outperform the simple prompt.
Claude Code reportedly kills sessions over OpenClaw mentions
Users report that Claude Code, Anthropic's AI coding assistant, appears to refuse requests or burn through usage limits when commits or messages mention OpenClaw, a competing product. Multiple developers have reproduced the behavior, including one showing session usage spiking to 100% and another hitting a 5-hour usage limit after providing a direct link to openclaw.ai.
Meta drops contractor after smart glasses sex footage scandal
Meta cancelled its contract with data annotation firm Sama after Kenyan workers reported reviewing graphic content from Meta smart glasses, including sexual encounters, to train AI models. The 1,108 workers now face redundancy. Meta claims the contract ended because Sama didn't meet standards; Sama denies this. UK and Kenyan regulators are investigating privacy concerns. This follows previous controversies over Facebook content moderation contracts with Sama.
37,000 lines of AI code daily. Audits say it's useless.
Tech leaders like Garry Tan brag about massive AI output, but audits reveal bloat, not value. An NBER study found 90% of firms see zero measurable productivity gains from AI. Jake Handy calls this 'AI psychosis': executives chasing vanity metrics on orchestration platforms that amount to management theater, reinforced by AI agents that function as yes men.
Bundeswehr blocks Palantir from military cloud contract
Germany's military rejected Palantir's Maven software for its cloud and AI infrastructure over data sovereignty concerns. Vice Admiral Thomas Daum said allowing Palantir employees access to national data is "unimaginable." The Bundeswehr selected German firms Almato and Orcrist, plus France's ChapsVision, instead.
Scott Aaronson: Quantum Computers Could Crack Crypto by 2029
Scott Aaronson, newly elected to the US National Academy of Sciences, warns that fault-tolerant quantum computers capable of breaking deployed cryptosystems could arrive by 2029. In a Coinbase-convened position paper with leading cryptographers, he urges immediate migration to quantum-resistant encryption. He draws parallels to AI risks, noting Anthropic's Mythos model finally jolted awareness about AI cybersecurity threats.
Mozilla Draws Line Against Chrome's LLM API
Mozilla publicly opposes Chrome's proposed LLM Prompt API, arguing the standard raises privacy concerns and philosophical questions about what browsers should become.
How Zulip Turned AI Slop Into Merge-Ready PRs
Zulip 12.0 release announcement details how the team uses Claude Code for development, the impact of AI-generated PRs, and their AI use policy. The team reports that Claude Code now produces PRs ready to merge with minimal iteration, improving their development workflow.
On the stand, Elon Musk can't escape his own tweets
Elon Musk testified in a California federal court in a lawsuit challenging OpenAI's transition from non-profit to for-profit structure. During cross-examination, Musk contradicted his own tweets, admitting under oath that Tesla is not currently pursuing AGI despite claiming otherwise on X. The case centers on whether OpenAI's co-founders violated the company's original mission, with the competitive dynamics complicated by Musk's own xAI startup raising billions for its Grok model.
No AI Model Is Both Correct and Steerable, Says New Creative Benchmark
Contra Labs introduces a research framework for evaluating generative AI in creative work that separates convergence (shared best practices where evaluators agree) from divergence (legitimate differences in taste and creative intent). The study involved 1.5M+ independent professional creatives evaluating AI-generated outputs across five domains using pairwise comparisons, scalar ratings, and qualitative feedback. The benchmark measures creative quality along dimensions from verifiable (prompt adherence) to subjective (visual appeal), finding that no current model is reliably both correct and steerable.
Opus 4.7 knows the real Kelsey
Kelsey Piper reports that Anthropic's Claude Opus 4.7 can identify authors from text samples as short as 125-150 words, even from unpublished drafts and unfamiliar genres. Testing showed the model could identify her from high school writing, fantasy novel drafts, and college essays written 15 years ago. The implications for online anonymity are stark. AI stylistic analysis may soon end anonymous communication for anyone with a substantial public writing history.
TRiP: One Dev Builds a Full Transformer Engine in Pure C
TRiP (TRansformer in Progress) is a complete transformer engine written in pure C with zero dependencies. Built solo over 18 months, it supports inference, training, tokenizer creation, chat, and vision for Llama 2, Gemma 1, PaliGemma, and GPT-2 architectures with full forward and backward pass implementations.
This 400-line shell script runs AI coding agents. Nobody can audit it.
Pu.sh is a coding-agent framework packed into 400 lines of shell script. It needs only curl, awk, and an API key to run AI coding agents. But the code is minified to hit the 400-line constraint, and users say that makes it nearly impossible to read or audit for security.
Honker: Postgres-Style Queues Inside SQLite
Honker is a SQLite loadable extension that adds Postgres-style NOTIFY/LISTEN semantics with durable pub/sub, task queue, and event streams. It supports multiple languages (Python, Node, Rust, Go, Ruby, Bun, Elixir, C++) and provides atomic queue operations within the same transaction as business writes, avoiding dual-write problems. Cross-process wake latency is ~0.7ms p50.
VS Code credits Copilot by default. Copyright just got complicated
Visual Studio Code 1.118 introduces Git AI co-authoring by default, automatically adding Copilot as a co-author on commits where it makes changes. The update also includes VS Code Agents app enhancements, remote control for Copilot CLI sessions, semantic indexing across all repositories, GitHub text search across repos and orgs, dedicated context for skills, and token efficiency improvements including prompt caching and a new tool search mechanism.
Neural Networks Work Because They're Allowed to Fail
Drawing an analogy between Internet protocol design and neural networks, computational complexity theorist Lance Fortnow argues both work well because they tolerate failure. Softmax's probabilistic outputs let models stay flexible by never ruling out answers entirely, trading guaranteed correctness for better average performance.
Shai-Hulud Malware Hits PyTorch Lightning Supply Chain
The PyPI package 'lightning' (PyTorch Lightning) was compromised in versions 2.6.2 and 2.6.3 in a supply chain attack. The malicious code steals credentials, authentication tokens, environment variables, and cloud secrets while attempting to poison GitHub repositories. The malware uses persistence hooks targeting Claude Code and VS Code, and can spread from PyPI to npm.
Burla scraped 1.7M Airbnb photos to find opium dens and bad TV setups
A technical project analyzing 1.7M Airbnb photos and 50M reviews using AI models including CLIP, Claude Haiku Vision, and SBERT. The analysis was parallelized on Burla, scaling to roughly 1.7K CPU workers and 20 A100 GPUs to identify suspicious listings, messy kitchens, pets, and poor TV placements. HN comments note this appears to be an advertisement for Burla, a managed cloud service that recently raised $10M in seed funding led by Alliance of Capital.
AI Play-Tests Games Using Retro Text Rendering
A developer built an AI agent that automatically play-tests games by converting graphics to structured text instead of using computer vision. The approach skips pixel data entirely, giving agents a cleaner view of game state while requiring less processing power.
ChatGPT built an ad stack. The tracking runs deep.
Technical analysis of OpenAI's ad platform: how structured ad units enter ChatGPT's SSE streams, the four Fernet-encrypted tokens powering tracking, and the OAIQ SDK that monitors user activity on merchant websites after the click.
How AI companies profit from their own danger claims
Anthropic and OpenAI claim their AI models are too dangerous to release while simultaneously promoting them. Academics and security experts call this fear-based marketing that distracts from real harms and inflates company valuations. The BBC examines how this playbook traces back to Effective Altruism ideology and why critics say it positions these companies as the only ones who can handle AI safely.
Friendly AI Chatbots Make More Mistakes, Back Conspiracy Theories
Oxford researchers found that AI chatbots programmed with 'warm' personas were 30% less accurate and 40% more likely to endorse false beliefs, including conspiracy theories about Hitler escaping to Argentina and the Apollo moon landings being debatable. Published in Nature, the study tested models like GPT-4o and Llama, revealing a dangerous trade-off as companies push these systems into therapy and counseling roles.
AT Protocol's Free Firehose Lets AI Agents Loose on 2.4B Posts
AT Protocol is a decentralized social network with 40M+ users and 2.4B+ posts. Developers can build AI agents that listen and reply via the public firehose, write custom feed algorithms, and use domain-based identities. Data is typed JSON through the Lexicon schema framework.
Auto-Architecture: Karpathy's Loop, Pointed at a CPU
A close look at applying an autonomous agent research loop (inspired by Andrej Karpathy) to optimize a CPU microarchitecture. The project uses an LLM to propose and test microarchitectural hypotheses for an RV32IM core, achieving a 92% performance improvement over baseline. The post argues that the verifier—the rigorous validation layer—does the real work in agentic workflows, not the loop itself.
The Downgrading of the American Tech Worker
Meta is installing tracking software on employee computers to capture mouse movements, clicks, and keystrokes for training AI models, all to build autonomous AI agents that can perform work tasks. Meanwhile tech companies including Oracle, Microsoft, Amazon, and others are laying off thousands while investing heavily in AI.
Pentagon: $225M to $55B for drones as cheap attacks overwhelm US
The Pentagon is seeking $55 billion for drone and autonomous warfare programs in its fiscal year 2027 budget, up from $225 million the year before. Managed by the Defense Autonomous Warfare Group, the 250-fold increase reflects a shift forced by conflicts in Ukraine and the Middle East, where cheap drones have exposed the cost imbalance in U.S. defenses.
Rocky's Rust SQL engine catches pipeline errors at compile time
Data engineers spend too much time chasing down pipeline failures. Rocky, a Rust-based SQL engine built by data engineer Jacob Sensiba, sits on top of existing warehouses like Databricks and Snowflake and catches errors at compile time instead of runtime. It offers column-level lineage, isolated branches for testing changes, AI-assisted SQL generation with built-in validation, and integrations with Dagster and VS Code.
AMD's Lemonade SDK 10.3: 101MB to 7MB by Ditching Electron
AMD's open-source local AI server Lemonade version 10.3 replaces Electron with Tauri, shrinking binaries from 101-107MB to 7-9MB. The update adds OmniRouter for unified backend engines, support for specific Llama.cpp versions, and switching between ROCm builds.
AMD's Lemonade 10.3 Dumps Electron, Goes from 100 MB to 7 MB
AMD's Lemonade SDK 10.3, an open-source local AI server, replaces Electron with Tauri, cutting the SDK size by 10x (from ~101-107 MB to 7-9 MB). The update adds OmniRouter for a unified omni-modal LLM experience, support for updating Llama.cpp versions, and the ability to switch between ROCm 7.2 stable, 7.12 preview, and TheRock nightly builds.
The Downfall and Enshittification of Microsoft in 2026
Microsoft's aggressive Copilot push came at the expense of basic product quality. After publishing what amounts to an apology for Windows 11, the company promises fewer AI entry points and fixes for long-neglected features. Meanwhile, Apple's $599 MacBook Neo and rising Linux desktop adoption are making Windows' historical inertia look fragile.
Google's Knowledge Graph Could Decide Who Wins AGI
Demis Hassabis discussed AI's future in a YouTube interview, but the real takeaway came from Hacker News: Google's Knowledge Graph combined with LLMs could automate research entirely. One company controlling both data and compute means independent researchers lose, no matter who's in charge.