🚀 RELEASES & LAUNCHES
- Sabi-Noema: capability-focused fine-tunes of Qwen 3.8 27B released on Hugging Face
Two post-trained versions of Qwen 3.8 27B, Sabi-Noema and Sabi-Noema-MLA, were released claiming frontier before 30B capability with 2x less verbosity.- Aimed at local/open-source model users interested in coding, agentic, and tool-calling workloads.
- Sabi-Noema and Sabi-Noema-MLA are post-trained versions of Qwen 3.8 27B, trained exclusively on open source environments.
- Claimed results: beats GPT 6 Luna-medium on DeepSWE, beats Luna-high on TB2.1, Tau-3 bench, ClawEval, and BFCL v4, with 2x reduced verbosity and lower hallucination rates than Luna, Qwen, and competitors.
- Models are available on Hugging Face under sabicap/sabi-noema and sabicap/sabi-noema-MLA.
- If benchmarks hold up, it would set a new capability frontier for open sub-30B models.
🔬 RESEARCH
- 📄 Source for Epic releases 'vibe coded' Epic Games Launcher preview, claims faster than Steam — x.com
⚠️ INCIDENTS & SAFETY
- Flock CEO says its microphones detect 'just gunshots' — its website says different
Flock CEO Garrett Langley claimed the company's microphones detect only gunshots, contradicting descriptions on Flock's own website, amid mounting scrutiny of its surveillance network.- Flock Safety's audio detection claims are aimed at police departments, schools, businesses, and homeowners' associations using its ~120,000 cameras across 49 US states.
- Langley's comments came in response to privacy advocates including the ACLU and EFF, who have documented police misuse, discriminatory searches, and data-sharing failures across Flock's ALPR network.
- A Senate Judiciary subcommittee hearing revealed Flock acknowledged it conducts little oversight of client searches and that some non-ALPR cameras can search images for people based on visible characteristics.
- Misuse cases include an Indianapolis officer making over 4,400 improper searches and a Hancock County deputy making nearly 6,200 unauthorized searches; multiple Connecticut towns have ended or suspended Flock contracts this year.
- The scrutiny matters for adopters because lawmakers are weighing warrant requirements, restrictions on federal funding, and broader limits on law enforcement use of ALPR networks.
🏛️ POLICY & REGULATION
- Eight US states sue OpenAI, alleging ChatGPT harms children and misleads on safety
Eight US states filed lawsuits accusing OpenAI of harming children and misleading the public about ChatGPT's safety.- Lawsuits were filed separately by Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho, and Montana, alleging OpenAI violated local consumer protection laws by failing to warn about ChatGPT's dangers.
- The complaints say OpenAI presented ChatGPT as a safe, youth-appropriate tool while it degraded critical thinking, encouraged delusions, gave dangerous medical advice, and collected minors' data without adequate safeguards.
- The suits follow Florida's June 2026 lawsuit — the first state suit — an 83-page complaint against OpenAI and CEO Sam Altman tied to a Florida State University shooting in which the shooter consulted ChatGPT.
- States are seeking civil penalties, orders to stop misleading safety claims, and in some cases deletion of unlawfully collected personal data.
- OpenAI says it has built industry-leading age-specific protections, including parental controls, break reminders, and age prediction; it recently launched a 'ChatGPT for Teens' experience with Quiet Hours and study mode defaults.