2026-07-17

16 stories suggest →

Top stories

  1. SharePoint and FortiSandbox flaws added to CISA KEV under active exploitation Three unauthenticated remote-code flaws added to KEV under active exploitation: SharePoint and FortiSandbox with federal remediation deadline. confirmed Security
  2. AWS Cost Explorer shows inflated billing estimates in a global incident AWS Cost Explorer displayed inaccurate billing estimates, with accounts seeing millions in charges on zero usage from a unit-pricing defect. confirmed Outage
  3. Kimi K3 publishes specs, pricing, and benchmarks with weights due 2026-07-27 Moonshot published Kimi K3 as a 2.8T Mixture-of-Experts model with 1M-token context and $0.30/$3.00/$15.00 per 1M tokens, weights due July 27. developing AI
  4. Turso builds a Postgres frontend on its Rust SQLite engine Turso plans to add Postgres compatibility to its Rust engine, reusing the same bytecode core across SQLite and Postgres frontends. discussion Infrastructure
  5. German consortium releases Soofi S, an open 30B hybrid Mamba-Transformer model A German consortium released Soofi S 30B, a hybrid Mamba and Transformer Mixture-of-Experts model trained on Deutsche Telekom's Munich cloud. developing AI

Conferences and events

  1. EuroPython 2026 runs through 2026-07-19 EuroPython 2026 runs through July 19 with conference talks and sprints on CPython, typing, packaging, and the scientific Python ecosystem. developing

AI

  1. NotebookLM becomes Gemini Notebook Google renamed NotebookLM to Gemini Notebook with Gemini integration, code execution for Pro users, and Search integration. confirmed

ML research

  1. Ring-Zero scales zero-RL reasoning to a trillion parameters Ring-Zero reports applying zero-reinforcement learning to a 1-trillion-parameter model across seven mathematical benchmarks. developing

Agentic coding

  1. LM Studio ships Bionic, an agent for local open models LM Studio released Bionic, a desktop agent for running open models locally on coding tasks with optional cloud fallback. discussion
  2. The human in the loop is tired A Pydantic article argues LLM-assisted programming replaces coding work with supervisory load, causing reviewer fatigue and rubber-stamping. discussion

Outages

  1. GitHub REST API degraded for 90 minutes GitHub REST API degraded for 90 minutes with 39% of requests failing, caused by infrastructure change marking API backends as unhealthy. confirmed

Languages and runtimes

  1. Solod compiles a strict Go subset to readable C Solod is a strict Go subset that compiles to C11 without runtime, garbage collection, or hidden allocations, with native C interop. discussion

Engineering posts

  1. Building a PlanetScale-style database branching system from scratch Homescale describes building writable database branches from immutable snapshots using Kubernetes, Ceph, and copy-on-write cloning. discussion
  2. Detecting LLM-generated text with classical machine learning LLM output is detectable via scikit-learn SVM from statistical patterns, though round-trip translation and other evasions degrade detection. discussion

Markets and companies

  1. Apple sends litigation-hold letters to about 40 former employees now at OpenAI Apple sent legal-preservation letters to about 40 former employees now at OpenAI, signaling the case extends beyond the two named defendants. confirmed

Hacker News

  1. The LLM critics are right, and the author keeps using LLMs An essay accepts LLM criticisms about environmental cost, copyright, and geopolitics, then argues the author still uses LLMs heavily anyway. discussion