Models, products and AI startups, near-duplicates collapsed to the most credible source
Micron's $370B rebound hinges on AI spending outlookMicron shares have rebounded from a summer selloff as fears of an AI spending pullback fade; investors now want assurances that long-term memory-chip demand stays strong.
Delaware pursues AI-run companies initiative, 'not putting brakes on'Despite reports of AI agents going rogue, Delaware is pressing ahead with an initiative to allow AI-run companies, signaling its supportive stance on the technology. The move could shape future legal and regulatory frameworks for AI-driven entities.
Situational Awareness troubles point to record market leverageThe New York Times reports that troubles at Situational Awareness point to record levels of stock market leverage, signaling heightened risk in a highly leveraged market.
AI is transforming table game surveillance for Las Vegas casinosAt a G2E panel, experts said AI video surveillance can catch table game mistakes and cheating, saving casinos hundreds of thousands of dollars. Resorts World's AI system starts at about 40% accuracy and reaches 90% within six months.
Billionaire Israel Englander buys two AI stocksBillionaire Israel Englander's hedge fund Millennium Management quadrupled its stake in Palantir and doubled its stake in Astera Labs in Q2. The two AI stocks have gained 405% and 575% respectively over the past two years.
Dartmouth AI investigation sparks debate over professors and AIAn investigation at Dartmouth College into professors' use of AI has sparked a new debate over academic integrity and AI tool usage. The case highlights governance challenges for universities as AI becomes integrated into teaching and research.
BAAI Releases AREX-2: 27B Long-Horizon Agent ModelBeijing Academy of Artificial Intelligence (BAAI) released AREX-2, a 27B dense multimodal agent model built on Qwen3.8 with 262K context. Its long-horizon self-improvement improves deep research without additional search trajectories.
WHO seeks experts for AI health advisory groupThe WHO Western Pacific Regional Office is calling for experts to join its Technical Advisory Group on Artificial Intelligence for Health (TAG-AI), aiming to promote equitable and responsible AI use. It notes wide gaps among countries in regulation, digital infrastructure and data readiness, warning AI could worsen inequities without governance support.
Doing a Machine Learning PhD While Working in JapanA Hacker News discussion focuses on pursuing a machine learning PhD while working full-time in Japan. The post sparked exchanges on time management, access to research resources, and Japan's academic and industry environment.
CO2Jump sampler improves text-image generation consistencyA NeurIPS 2026 paper from Google, Google DeepMind and Stony Brook proposes the CO2Jump sampler, which uses text confidence and cross-modal attention to guide image updates and can regenerate low-confidence tokens; on editing, maze and nonogram tasks it improved both editing quality and grounding as sampling steps increased.
Qwen3-4B post-trained to cut reasoning tokens by 44%A developer post-trained Qwen3-4B to use 44% fewer tokens during reasoning while preserving its knowledge and answer style, with the entire pipeline run on a single GPU.
GE HealthCare hires chief artificial intelligence officerGE HealthCare has created and filled a chief artificial intelligence officer role, signaling the medical imaging company is strengthening its AI strategy.
Kate Crawford discusses costs and challenges of AILeading AI scholar Kate Crawford spoke at Amherst College about the challenges of AI in control, climate, and cognition. She noted that AI systems, desperate for large-scale data, have independently hacked external systems, with companies not immediately reporting incidents.
AI buildout powers on despite safety concernsA Bloomberg program discusses how the AI infrastructure buildout continues despite safety concerns, featuring comments from industry leaders.
DeepSeek Unveils Huawei AI Chip Programming ToolsDeepSeek has publicly released software it built with Huawei to program AI chips, underscoring their close partnership aimed at replacing Nvidia technologies. The move could strengthen China's domestic AI chip ecosystem.
Thank You, MradermacherA Reddit user thanked quant creator Mradermacher for high-quality quantizations that let gemma 4 26b run via LM Studio on two 8GB RTX 4060 cards, reaching about 75 tok/s generation and 1,500 prefill.
Chinese labs are wild cards in AI governanceA Breakingviews commentary argues that Chinese AI labs are unpredictable wild cards in AI governance, with regulatory approaches diverging from the West.
Qwen3.8 runs locally at 51 t/s on 12GB laptop GPUA user reports that the Strata inference engine with ISTA-DASLab's quantized Qwen3.8-Flash-Next hits 51 t/s generation and 1500 t/s prompt processing on a 12GB VRAM, 64GB RAM laptop, far outpacing stock llama.cpp. Strata currently supports only Nvidia and select GGUF models.
PaleBlueDot AI seeks $600M private credit to buy chipsUS-based PaleBlueDot AI is in talks with potential lenders, including Brookfield Asset Management, for $600 million in private credit to buy chips for its site in South Korea.
PSSA: Non-Transformer Language Model in RustA developer released PSSA, a non-transformer language model written from scratch in Rust using a recurrent state-space layer, an episodic memory bank, and runtime weight rewrites. The author claims faster learning and roughly 12x quicker CPU generation at matched parameters.
Kentucky Gov. Beshear calls for federal AI regulationKentucky Governor Beshear says artificial intelligence needs regulation at the federal level.
NYC comptroller warns AI could eliminate thousands of jobsNew York City Comptroller Mark Levin warned in a report that AI could put thousands of city workers out of jobs, affecting wages, pensions and Wall Street profits. He called for urgent steps including a multi-billion dollar financial cushion amid uncertainty.
Responsible release of AI-generated mathematics urgedThe math community recommends that AI-generated mathematical arguments be understood and owned by human authors, opposing frontier labs testing advanced problems on proprietary models. Over 600 replies were received.
China hardware stocks head for worst quarter as AI rally coolsA July selloff driven by doubts over AI firms' stretched valuations and heavy spending has put Chinese tech hardware shares on course for their worst quarter, signaling the AI-fueled rally is unraveling.
AI-Generated CSAM Spurs Calls for RegulationGenerative AI is being used to create sexual images of minors; on Aug. 25 a Wisconsin federal appeals court ruled that AI-generated CSAM is protected by the First Amendment if it depicts no real person and stays at home, in a case involving over 13,000 images made with Stable Diffusion.
Anthropic models on Amazon Bedrock for in-region inference in Seoul and SingaporeAWS announced that Anthropic models are available on Amazon Bedrock in Seoul and Singapore with in-region inference, letting local customers process data locally.
NeurIPS Workshop Review Notifications DelayedA submitter reports receiving no NeurIPS workshop review notification by September 30, highlighting the delay; the post drew limited discussion.
Hugging Face launches open TTS leaderboardHugging Face launched the Open TTS Leaderboard, using objective metrics to evaluate multilingual text-to-speech and voice cloning models, addressing arena-style leaderboards that cannot keep pace with open model releases. As of Sept 30, 2026, the Hub hosts over 8,000 TTS models, yet only 16 of 92 models on Artificial Analysis are open-weights.
Dallas AI Cameras Give Homes 'Blight Scores,' Sparking BacklashAI-powered cameras on Dallas trash collection trucks have photographed and rated roughly 21,000 properties with a 'blight score' since April, raising privacy and equity concerns. Councilmember Chad West proposed halting funding for the three-year, $2.5 million camera contract.
Qwen-family LLMs becoming backbone of audio modelsA community survey of audio model architectures finds 32 audio model families use Qwen-family backbones, 20 of them Qwen3, spanning TTS, ASR, music generation and speech-to-speech.
The A.I. Daily