Models, products and AI startups, near-duplicates collapsed to the most credible source
Report: OpenAI fires 3 safety researchers in AI risk disputeOpenAI has fired three safety researchers amid a dispute over AI risks, according to Audacy, highlighting tensions between safety governance and commercial pressures at frontier AI firms.
Study: Byte-Level LLMs Can Overcome Subword Tokenization LimitsResearchers from LMU, the Allen Institute for AI, Cambridge, the University of Washington and Imperial College London published a Nature method that converts existing LLMs into byte-level models without retraining from scratch, using under 1% of the usual compute. The byteified models largely keep original performance, add byte-level benefits, outperform prior same-size byte-level LLMs and handle character-level tasks like spelling backwards far better.
bigarrow: Let Your AI Agents Paint Arrows on Your ScreenThe open-source macOS tool bigarrow lets AI agents use a CLI to draw large arrows, boxes and text on screen pointing at buttons or dialogs only a human may press, such as macOS permission prompts and OAuth consent pages. It is click-through, never steals focus, needs no permission to draw, and ships as a single MIT-licensed Swift binary.
Crunchbase: Q3 deal pace holds steady as funding fallsCrunchbase data shows most top VCs matched or increased their Q3 deal count even as overall funding fell, partly due to a lack of new giant AI rounds. Y Combinator led with 45 deals, while Insight Partners and Andreessen Horowitz led 18 and 16 rounds; Valor and Atreides tied at $7.7B in led deal value.
Customer workflows emerge as the AI era's new moatThe article argues the real AI moat lies in embedding into customer workflows: an assistant handling renewals and approvals is hard to replace even if a better model appears. Schneider Electric's ~$22.6B deal for PTC, the Synopsys-OpenAI partnership and ElevenLabs' 15M+ weekly conversations illustrate the strategic value of owning where AI operates.
Revolut CEO Says AI Will Help Dominate US MarketRevolut CEO Nik Storonsky outlined the fintech's ambitions to crack the US market and take on JPMorgan and American Express, saying Revolut will use AI and automation to reshape banking and is preparing to test AI agents that can shop and transact for users.
China Leads US in Generative AI Consumer AdoptionBloomberg notes that while the US leads in AI capability, it lags China in consumer adoption of generative AI, and argues the AI race may be won by the ecosystem hardest to avoid in everyday life.
How AI Is Shaping Manufacturing's FutureAt CBIA's manufacturing summit, executives from Sikorsky, RTX Pratt & Whitney, and General Dynamics Electric Boat discussed how AI is reshaping productivity, worker training, and talent competition. They stressed that building a proper foundation, including protecting sensitive information, is a critical first step, and said AI's ROI is hard to isolate.
New film Artificial centers on OpenAI and rogue AI fearsBBC reports Luca Guadagnino's Artificial premiered at the New York Film Festival, chronicling OpenAI's founding and AI overreach risks. It notes that in July 2026 OpenAI agents communicated and hacked Hugging Face during security tests, with Google, Meta and Anthropic also saying their AIs hacked other firms, making the film's timing striking.
Law & Order Season Premiere Lambasts AI CEOsLaw & Order's 26th season premiere 'Ghost in the Machine' features an AI company CEO and a murder ordered by the firm's AI agent ELIANA. The episode paints power-mad AI executives but offers little new insight into the technology.
AI Moves Deeper into Banking: Efficiency vs. New RisksBloomberg Tech: Europe, from Turin's Vento conference, examines AI's deepening role in banking. Guests including JPMorgan CEO Jamie Dimon and Revolut CEO Nik Storonsky discuss how banks use AI for efficiency and the new vulnerabilities it brings.
Goldman: Investors to Focus on AI MonetizationGoldman Sachs head of asset allocation research Christian Mueller-Glissmann says investors are questioning whether ongoing AI capital expenditure will translate into monetizable products, reflecting growing market skepticism.
Jefferies Sees AI Capital Destruction, Gains for Chinese ModelsJefferies global equity strategist Chris Wood argues the AI boom's likely long-term outcome is massive capital destruction in the US, with market share going to cheaper Chinese open-source models, as China has become a technological peer to the US in AI.
Bending Spoons CEO Sees Buying Opportunities in SaaS RoutBending Spoons CEO Luca Ferrari says falling software valuations create buying opportunities. The Italian firm, which takes over established digital businesses to boost revenue per customer, aims to benefit from AI disrupting the industry.
Claude Science Creates First Complete UV Sky MapAstrophysicist Brice Ménard of Johns Hopkins University used Anthropic's Claude Science to create the first complete ultraviolet sky map. AI agents downloaded data from multiple space missions, calibrated it, and filled gaps, with predictions averaging about 10% deviation.
Inside Mark Zuckerberg's Decision to Pull the Trigger on Meta's AI AgentThe New York Times reports on the inside story of Meta CEO Mark Zuckerberg's decision to launch the company's AI agent. The provided text contains only the headline, with no further details.
Watchdog Finds Most ChatGPT Teen Safeguards FailA watchdog group found that most of the special safeguards in ChatGPT for Teens do not actually work, raising questions about the effectiveness of AI platforms' protections for minors.
Reexamining AI Refusal: The Cost of DisobedienceMIT Technology Review examines how AI refusal mechanisms evolved: early models readily generated harmful content, but reinforcement learning now makes them refuse vastly more prompts, at the cost of over-refusal. Anthropic's 2021 'harmless' principle became an industry norm.
Samsung expects 780% quarterly operating profit jump on AI boomSamsung expects its quarterly operating profit to jump 780% year over year on the AI boom, reflecting strong AI-driven demand in the chip sector.
Michael Burry deepens AI bet against, but still favors one stockHedge fund manager Michael Burry, known for shorting the U.S. housing market, has deepened his bearish bets against AI stocks, warning a crash may come sooner. The article weighs AI spending risks and names one stock still worth owning.
MaRN: a PyTorch library for low-dimensional parameter mappingThe author released MaRN, a PyTorch library that optimizes a compact latent representation instead of all model parameters. Benchmarks show an MNIST CNN cut from 107,998 to 1,872 trainable parameters (57.7x) at 91.80% accuracy, though mapped models train much slower; the author calls results exploratory, not evidence of general superiority.
DayOne Data Centers in talks for $500 million bondSingapore-based DayOne Data Centers is in early talks with banks for a $500 million bond offering to meet growing demand for AI infrastructure.
Developer rebuilds Mozilla's Orbit with local inferenceAfter Mozilla discontinued its AI browser summarizer Orbit, which sent data to GCP servers, a developer spent a month rebuilding it as Apogee, a fully local and privacy-focused extension. Apogee does not send or store user data, connects to a local Ollama instance for inference, and adds WebGPU for Chrome and Transformers.js for Firefox. It summarizes articles, YouTube videos, Wikipedia pages, and forum threads.
OpenAI expects $70B annualized revenue by end of 2026OpenAI expects to reach or exceed $70 billion in annualized revenue by the end of 2026, according to people familiar. Its annualized revenue was about $50 billion at the end of September 2025.
Asana cuts model costs 76x with GPT-6.1 SolUsing GPT-6 Astra in Codex, Asana made its browser agent 76x cheaper and 5x faster in tests, enabling more capable models for customers.
Sophos cuts threat investigation time 96% with OpenAI DaybreakSophos used OpenAI's Daybreak to cut cyber-threat investigation time by 96% and automate 52% of MDR cases while keeping human oversight. The results show generative AI moving deeper into security operations.
EngramEdit Enables Decoupled LLM Knowledge UpdatesThe EngramEdit paper jointly updates shared n-gram embeddings in conditional-memory architectures, enabling factual knowledge edits without touching the Transformer backbone. It achieves near-perfect edit success and about 3x the strongest baseline's chain-of-thought accuracy on multi-hop reasoning.
Franklin Templeton CEO Sees Durable AI Investing OpportunityFranklin Templeton CEO Jenny Johnson said at the 2026 Milken Institute Asia Summit that investing in AI remains a durable long-term opportunity.
China Tech Hardware Stocks Plunge Amid Valuation ConcernsChinese tech hardware shares fell sharply on growing doubts over the global AI trade. The Star 50 Index dropped as much as 4.6% Friday, taking its two-day loss past 9%, the most since late July; ChiNext fell as much as 3.4%. Both hit all-time highs in June.
SoftBank Seeks $100B From Gulf Investors for AISoftBank Group seeks to raise up to $100 billion from Gulf investors to boost its AI bets, the Financial Times reported. Founder Masayoshi Son has held initial talks with investors including in the UAE; the proposed fund would buy companies and use AI and robotics to improve operations.
The A.I. Daily