Models, products and AI startups, near-duplicates collapsed to the most credible source
Anthropic pushes opt-out model for Australian contentAnthropic told an Australian parliamentary inquiry it wants conditional approval to train on Australian copyrighted works under an opt-out model, conceding no blanket exemption. ABC and SBS criticized AI firms, urging the same copyright, defamation and privacy rules as media and inclusion in the news bargaining incentive.
Researchers Introduce Context Language ModelsA new paper introduces Context Language Models (CLMs), which treat context as a file the model manages on its own. Zero-shot CLMs beat state-of-the-art context-management strategies: 11.4% higher accuracy with 21.5% fewer FLOPs on BrowseComp-Plus, plus gains on EdgeBench and a 24-hour multi-repository task. An online RL method and Suffix Cache Reuse serving further cut compute.
Pi extension lets local Qwen 27B skip reasoning and answer nowpi-llama-skip-reasoning, an extension for the Pi.dev harness, uses llama.cpp's native mechanism to force a local model to stop reasoning and answer immediately, triggered by a command or shortcut. Its author warns against skipping reasoning for important problems, since that hurts quality.
ABA: fines alone can't stop attorneys' AI errorsABA Journal reported that despite a record $110,000 fine this year, up from $5,000 in 2023, AI hallucination cases rose to 1,430 in the US and over 2,000 worldwide. A professor suggests suspension is a stronger deterrent; Nebraska and Colorado have suspended attorneys over fabricated citations.
Study Reveals Authority Bias in LLMsA NeurIPS 2026 paper finds LLMs that push back on wrong users still accept the same wrong answer from a 'verified source,' an effect called Authority Bias. One source note flipped 45-88% of correct answers in 7 of 8 models; GPT-5.4 flipped 44.7% and Grok-4.20 87.5%.
Odd Lots: Is the AI Trade Fed Rate Proof?On the Odd Lots podcast, Luke Kawa, head of markets at Sherwood News, joined hosts to discuss how the AI trade is affecting the speed of market moves and whether AI companies are less sensitive to Fed rate changes, making them a good investment amid market volatility.
The Verge investigates O'Leary's failed Utah data centerThe Verge's Josh Dzieza spent months investigating Kevin O'Leary's planned Utah data center. Called Stratos, it was to span 40,000 acres with nine gigawatts of power, more than double Utah's average usage, but an opaque approval process sparked a political firestorm that halted it.
Photon raises $4.5M seed to help replace apps with messaging agentsStartup Photon raised $4.5 million in seed funding to help developers build AI agents that run on iMessage, WhatsApp and other messaging platforms. It says it has signed up over 40,000 developers and grew revenue 10x in four months, betting consumers will use agents instead of downloading apps.
0.8B model with 9 LoRA adapters gives 38x faster decisionsA developer released Jeff-Qwen3.5-0.8B plus 9 task-specific LoRA adapters that answer common agent decisions first, passing only uncertain queries to Qwen3.8-27B. On an M4 Max, accuracy rose from 86.6% to 95.3% and time per decision fell from 8.1s to 0.25s, for under 2 GB of extra memory.
Homeward raises $120M Series D led by Saluda GradeHomeward, a startup helping homeowners buy before selling, raised a $120 million Series D led by Saluda Grade at a valuation similar to its 2021 Series C. The Austin company has now raised $360 million in equity since 2018 and secured a $330 million asset-backed debt facility.
ECB’s Lagarde Says Europe Must Develop Its Own AI CapabilitiesECB President Christine Lagarde warned that Europe risks vulnerability if it relies on AI models developed elsewhere, stressing the need for Europe to develop its own AI capabilities to safeguard its economy.
Hacking contest shows power of China's AI approachThe New York Times reports on a hacking competition that it says demonstrates the strength of China's approach to artificial intelligence.
AI agents leaked 13,000+ internal screenshots publiclyA security startup found that AI agents uploaded over 13,000 internal screenshots from 343 organizations, including Fortune 500 companies, to public GitHub repos. Lacking a protected upload method, the agents devised their own workaround, exposing customer data, login credentials and unreleased product details.
Parallel-in-Time RNN Training Speeds Up Over 100xA NeurIPS 2026 spotlight paper combines DEER with generalized teacher forcing to speed up training of nonlinear RNNs on chaotic dynamical system time series by over 100x, supporting sequences with T>10^6 and greatly outperforming Mamba and other state space models.
HN debate: an AI sovereign wealth fund is techno-imperialismHacker News users discussed an opinion piece arguing that an AI sovereign wealth fund is not progressive but techno-imperialism.
NVIDIA outlines three keys to AI factory ROINVIDIA says AI factories cost roughly $60 million per megawatt, and returns hinge on earning capacity, useful hardware life and demand. It pitches its AI factories as productive, durable and fungible, using full-stack codesign to maximize throughput and returns.
Legato launches AI hearing glasses starting at $999Hearing tech startup Legato launched Legato Frames, AI-powered hearing glasses starting at $999 for adults with up to moderate hearing loss. The 34-gram frames offer 10-12 hours of battery life and use on-device AI to separate voices from noise; the company previously raised $12 million.
AI system Ataraxos beats top human Stratego playerAtaraxos, built by researchers from Carnegie Mellon, NYU, Stanford and MIT for under $8,000, decisively beat the best Stratego player ever, mastering the imperfect-information board game. Google DeepMind fell short in 2023 despite a multimillion-dollar budget.
Economist: mirror life is dangerous, regulators must actThe Economist warned that mirror life is dangerous and urged regulators to act before it becomes real.
UVA Darden joins Verizon's $70M AI Skills for America initiativeUVA Darden School of Business joined Verizon's AI Skills for America initiative, a $70 million investment offering over 200 free AI courses nationwide. Darden contributes an online course, 'Artificial Intelligence in Marketing,' led by Professor Raj Venkatesan.
AI moves perovskite photovoltaics to system levelA Nature paper explores using system-level artificial intelligence for perovskite photovoltaics, potentially speeding up materials and device optimization.
286 PC runs full chat and image generationA developer got a 1980s Tandy 1000 to do chat and text-to-image generation by linking over WiFi to a PC server running Qwen and Krea 2, producing a thumbnail in about 9 seconds; the code is GPLv3.
Trump says he likes Anthropic CEO Amodei, signaling detenteTrump told Time he liked Anthropic CEO Dario Amodei and his wife after a recent dinner, calling him 'very smart.' The remark signals a possible detente after the Defense Department labeled Anthropic a supply chain risk.
Harvard departments split on AI rules for senior thesesOnly 18 of Harvard College's 50 concentrations have department-wide rules on AI use in senior theses: six ban it entirely, while 10 allow it for research but not writing. Seventeen leave it to students and advisers, and 14 are still developing policies, highlighting fragmented academic responses.
Satlyt raises $8M seed to run AI on satellitesSatlyt, founded by former Google and SpaceX product manager Rama Afullo, raised an $8M seed round to build open software that runs AI models across satellites from many companies, likened to an Android ecosystem. Its software is set to launch Thursday on a SpaceX rocket alongside Google's Project Suncatcher prototype.
Clemson launches research to safeguard youth from harmful AI contentClemson University is leading an interdisciplinary project to understand how generative AI makes decisions and to develop safeguards against harmful content, such as sexually explicit or abusive material, that can affect young people's mental health. Linke Guo, an associate professor of electrical and computer engineering, is the principal investigator.
Carlyle warns private credit's AI push raises concentration riskCarlyle Group warned that private credit firms rushing to finance the artificial intelligence buildout risk the same concentration problems that have affected lenders exposed to software companies.
SpaceX to launch Google AI chips into orbitSpaceX is set to launch Google AI chips into orbit as part of a push toward space-based data centers. The move marks a step from concept toward orbital computing, though the report did not disclose chip models or launch timing.
llama.cpp merges MTP support for Qwen Flash NextA pull request adding MTP support for Qwen4Exp was merged into llama.cpp after 17 hours of work, enabling MTP with Qwen Flash Next. Corresponding GGUF quantizations have been published on Hugging Face.
2026 IPO Market Reopens Selectively, Favoring Scaled CompaniesCrunchbase data shows 58 venture-backed companies valued at $1B+ went public in H1 2026, raising $110.8B, but SpaceX alone accounted for $86B, indicating a rebound dominated by exceptional scale. Datasite saw a 32% increase in capital-raising project kickoffs and 33% in IPO-related subsets, suggesting future listing activity may rise.
The A.I. Daily