Models, products and AI startups, near-duplicates collapsed to the most credible source
Qwen3.8-27B Tuned for Humanlike ChatA developer fine-tuned Qwen3.8-27B using 125,217 obfuscated human-to-human messages to produce a LoRA that makes replies shorter and more conversational; the model and a demo are published on Hugging Face.
Newsom signs law for third‑party AI auditorsCalifornia Governor Newsom signed legislation establishing a framework for third‑party AI auditors, creating an institutional pathway for regulatory review and compliance of AI systems.
AI inequity is developing in schoolsThe University of Chicago Booth School of Business warns that AI inequity is emerging in schools, suggesting gaps in access to and use of AI resources could widen educational disparities.
Judea Pearl at 90 still asking big questionsUCLA’s newsroom profiles Judea Pearl at age 90, emphasizing he remains engaged in core academic questions like causal reasoning and continues to be active in academia.
StartupHub.ai: OpenAI for Government and MoreStartupHub.ai summarizes industry updates including OpenAI for Government, Databricks Lakehouse, and Uber Eats search developments, offering a brief product and company roundup.
Tech whistleblowers warn AI could wipe out humanityThe Guardian reports tech whistleblowers warning that AI could threaten human existence; the piece urges taking such extreme-risk claims seriously and evaluating related policies and safety measures.
GPT-6 Astra helps Devin test softwareCognition says GPT‑6 Astra improves Devin's ability to test software and demonstrate that it works, aiming to reduce engineer review time and accelerate shipping.
Orukeet ASR, based on Parakeet, outperforms originalOrukeet is a 25-language ASR model built from NVIDIA Parakeet that reportedly outperforms Parakeet across many test splits, achieving about a 10% relative reduction in pooled WER on evaluated languages.
AI chatbot reduced mental health symptomsA report found an AI chatbot reduced symptoms of depression, anxiety, and risk for eating disorders, indicating AI’s potential in mental-health interventions; the piece is a research-news summary.
When AI goes rogue, who’s legally responsible?Bloomberg reports U.S. courts are handling cases where AI chatbots act as lawyers, witnesses, or are alleged to aid violent acts, highlighting new challenges for the legal system in assigning responsibility and evaluating evidence.
How dangerous is AI, really?The National Interest examines AI's potential risks versus real-world threats, assessing security, societal, and governance challenges. The piece aims to give a balanced perspective for the public and policymakers.
Hacker News, Without AIHacker News hosted a discussion titled “Hacker News, without AI,” with community members exchanging views on the topic. The discussion thread link was shared for readers to view the original conversation.
Timnit Gebru: doom talk distracts from real harmsTimnit Gebru told WIRED that apocalyptic AI rhetoric distracts from real harms, such as autonomous weapons, and is being amplified by companies.
Doing business in Canada: AI and dataDentons published guidance for doing AI- and data-related business in Canada, covering regulatory and compliance considerations for companies.
Meta to change AI suggestions after privacy gaffeMeta said it will change suggested prompts after its AI offered invasive questions on Instagram. The company admitted it “missed the mark” after a viral video by user Kalie Robins; spokesperson Dina El‑Kassaby said the prompts should not have appeared.
Meta meets European credit investors in roadshowMeta Platforms has been holding a non-deal roadshow with European bond investors as hyperscalers seek varied debt markets to fund AI expansion.
Bipartisan push for AI regulation after Anthropic warningAfter a viral extinction warning from an Anthropic researcher, bipartisan lawmakers are pushing for stronger AI regulation, drawing political attention.
Sam Altman proposes industrywide pauseSam Altman floated an industrywide pause to address frontier AI safety concerns, reflecting growing unease about advanced model risks.
Singapore weighing joining China-led AI body risks US retaliationReporting says Singapore is considering joining a China-led AI body but must weigh potential U.S. retaliation or diplomatic risks. This highlights the growing intersection of AI governance and geopolitics.
Anthropic: Iran used its US AI model to target US shipsAnthropic says Iran used the company's US-developed AI model to target U.S. Navy warships. The incident raises national-security concerns and could prompt reviews of model access, oversight, and responsibility.
Can AI Aid Early Schizophrenia Diagnosis?Smithsonian Magazine examines the potential and challenges of using AI for early schizophrenia diagnosis, assessing implications for clinical detection and treatment decisions.
OpenAI says next-gen chips could be made at SamsungOpenAI says its next-generation processors could be manufactured by Samsung. Double-sourcing alongside TSMC suggests very large volume needs and a focus on production capacity and supply redundancy.
NYC, LA, Montessori schools limit AI useBloomberg reports US districts including NYC and LA, plus Montessori schools, are limiting AI and tech amid parental concerns; moves aim to boost enrollment, cut costs, and address declining student numbers.
Training a 210M text-to-image DiT on one GPUThe author trained a 210M-parameter DiT on one RTX PRO 6000 (3.5 days, ~4.2M 256² images) and reports three findings: learned null attention slots dominate cross-attention, flow-matching loss reflects training health not quality, and a training-time timestep shift notably improves sampling quality.
New Warning on Artificial IntelligenceWhat The Tech published a new warning about artificial intelligence, discussing AI risks and potential impacts and urging attention from the public and policymakers.
AI to Redefine Heart Failure CareVia Medica Journal discusses how AI could shift heart failure care from monitoring to understanding clinical meaning, and explores potential impacts on decision-making and patient management.
AI shows signs of lifting Britain’s slumpBloomberg says AI is showing tentative signs of helping lift Britain’s economic slump; City of London participants did not foresee this summer’s growth spurt, suggesting AI may have contributed.
Re/Done Opens First SoHo Store with AIWWD reports that apparel brand Re/Done opened its first SoHo store in New York, blending AI technologies with old‑world New York elements to explore retail‑tech integration.
AI transforming Latin American electionsLatinoamérica 21 reports how AI is affecting elections across Latin America, focusing on cases in Colombia and Brazil and examining how technology changes campaigning, voter interaction, and associated risks.
OpenAI Questions Legality of an Industry SlowdownThe Decoder reports OpenAI has asked Congress whether an industry-wide slowdown in AI development would be legal, seeking legal clarity on coordinated pacing of AI progress.
The A.I. Daily