Only Tue, October 20, 2026 · All dates →
ChatGPT Work cuts franchisee cash reconciliation time by 94%An independent Schlotzsky's and Cinnabon franchisee used ChatGPT Work to cut menu repricing time by 95% and cash reconciliation time by 94%, building a workflow that could free up about $100K in cash flow.
OpenAI reportedly seeks $1.4T valuation ahead of Anthropic IPOOpenAI is reportedly rushing to reach a $1.4 trillion valuation before Anthropic launches its IPO, according to Gizmodo. The report offers no further deal details.
OpenAI builds humanoid robots, Altman targets 2027 demoSam Altman confirmed OpenAI is developing its own humanoid robots, targeting a first demonstration in 2027 and recruiting hardware specialists for factory and data center work. No design or commercial launch date has been revealed; the move pits OpenAI against robotics giants such as Tesla.
DeepSeek open-sources its DeepInfer inference engineDeepSeek has open-sourced DeepInfer, the high-performance inference framework behind its large-scale production services, with this release focused on offline inference. It supports DeepSeek-V4.1-Flash, offers parallelism such as attention DP and MoE EP with EPLB, optimized kernels including FlashMLA and DeepGEMM, and DSpark draft-token decoding with confidence-scheduled verification.
Dartmouth mathematicians weigh AI's role in future researchOpenAI's release of over 700 AI-generated mathematical results has prompted Dartmouth faculty and students to question their validity and debate AI's role in academia, with some arguing that merely answering questions is not the point of research.
Closing the agent governance gap in enterprisesA CIO article highlights a governance gap as AI agents spread inside enterprises, with the core problem being a lack of visibility into agents in their environments. Citing Anthropic, 82% of organizations have found unregistered agents and 57% already use agents for multi-stage workflows.
Inside Samsung's Pyeongtaek chip hub as AI drives memory demandSamsung is building a large semiconductor complex in Pyeongtaek, South Korea, where three fabs already run around the clock and a fourth is coming online. Construction of a fifth fab was halted in the 2024 memory downturn and restarted as AI demand surged. The campus is roughly the size of Central Park, with thousands of construction workers on site.
AI spreads through finance, bringing benefits and risksAI is already used across banking, insurance and markets for customer service, client matching, trading, compliance and predictive analytics, and is driving data-center capex. Industry figures warn synthetic identity fraud and AI agents making transactions will raise liability and consumer-protection issues.
OpenAI pulls back as chips policy battle heats upPolitico reports that OpenAI is pulling back as a lame-duck political battle over chips heats up. No further details were provided in the excerpt.
US sanctions Chinese VC firm founded by ex-Microsoft AI scientistThe US has blacklisted a Chinese venture firm founded by former Microsoft AI scientist Lu Qi, citing alleged ties to a startup accused of stealing technology from Google. The move signals escalating scrutiny of China-linked AI technology flows.
Trump officially replaces 'artificial intelligence' with 'superintelligence'Trump has officially replaced the term 'artificial intelligence' with 'superintelligence', per the headline; the provided text contains no details on policy or impact.
Spanish startups accelerate innovation with BSC AI FactoryThe Barcelona Supercomputing Center (BSC) AI Factory Accelerator Programme is fully operational, offering startups sovereign compute, technical support and training; several participants are NVIDIA Inception members, and some BSC services are now available via the Inception portal. Since its April 2025 launch, the AI Factory has delivered over 800 services, with Qatium, OneCareAI and another Inception member named 'AI Champions'.
Edvance and CWG launch RealRelay Fintech AI platformEdvance's Paradise Matrix and CWG launched RealRelay Fintech in Singapore, the third vertical on no-code AI agent platform RealRelay.ai. It automates reconciliation, document checks and fraud detection, cutting weeks of audit work to hours with every step recorded.
GMO says AI tool used in Japan breaches hit Korean banksTokyo-based security firm GMO said hackers who recently breached several major Japanese companies used a China-developed AI tool linked to a series of intrusions at South Korean banks last month.
Austin startup aims to move chip data at the speed of lightAustin startup La Luce Cristallina grows ultra-thin crystal layers on silicon wafers for photonic devices that move data with light instead of electrical signals. It says the material could eventually cut device power use by about 100 times; the technology stems from University of Texas research.
Open-source Berth enables dynamic AI engine swapping on a single GPUA Reddit user released Berth, an open-source tool that monitors and automatically swaps AI engines such as llama-swap, strata, ComfyUI and OpenWebUI on a single 24GB GPU, avoiding VRAM conflicts. It suits multi-task GPU sharing and can prevent processes from grabbing the card during gaming.
Migo Robot showcases embodied AI at Hong Kong fairAt the Hong Kong Electronics Fair (Autumn Edition), Migo Robot showcased migo, an AI companion robot focused on emotional connection rather than productivity, drawing crowds. Celebrities including Kingdom Yuen, Lam Tze-chung and Chan Ka-kai visited to try it, making the booth a popular photo spot and social media topic.
Open-source low-latency streaming voice server VoiceManager demoedA developer released VoiceManager under MIT license, enabling low-latency local streaming speech-to-speech over a single WebSocket with wake word, barge-in and tool calls. It ships client docs, a Python client and a Kotlin HermesConsole reference client.
Users warn Strata update brings many breaking changesA Reddit user warns against upgrading Strata, reporting many breaking changes on both Windows and Linux, including needless new serving requirements and the web UI shifting toward a desktop app. Back up first, the post advises.
NYC Mayor Mamdani urges residents to resist AI inevitability, demand regulationAt an event hosted by the AI Now Institute, NYC Mayor Zohran Mamdani said AI's takeover of jobs and classrooms is not inevitable and residents should demand accountability from AI companies. He has already banned AI in elementary and middle schools and required real estate platforms to disclose AI-generated listing images, as cities act amid limited federal regulation.
Citi chief economist discusses global growth and AI disruptionsCiti Global Chief Economist Nathan Sheets discussed the bank's global growth forecast amid rising energy costs, persistent inflation and disruptions from AI, speaking at the Citi ANZ Investment Conference in Sydney.
AI-generated political ads hit TV screens, experts say harder to spotA New York Times report says $147 million has been spent broadcasting AI-made political ads this cycle. Ads have featured AI-generated audio of a Senate candidate reading a Bible and a fake video of a senator on vacation. Texas law does not require disclosure of AI in political ads, though it bars AI deepfake ads in the 30 days before an election. Experts warn AI lowers ad costs but also lets people dismiss real content as fake.
OpenAI acquires Orchestra BioOpenAI is acquiring Orchestra Bio to bring AI-powered operations to life sciences R&D, aiming to help biopharma teams speed up drug development. The deal amount was not disclosed.
Hugging Face adds native TorchTPU PyTorch backendHugging Face said TorchTPU gives PyTorch a native TPU backend, so Transformers, Diffusers and Kernels run on TPU by simply sending a model to "tpu". It includes TPU-tuned Pallas kernels and distributed-training support; the previously used Optimum TPU is no longer required.
AnyLanguageModel 1.0 ships for local models on Apple platformsAnyLanguageModel 1.0 is out as a drop-in replacement for Apple's Foundation Models framework, letting apps switch between the system model, local models via llama.cpp, MLX and Core ML, and hosted providers such as OpenAI and Anthropic. swift-huggingface 1.0 and swift-xet 1.0 ship alongside to speed model downloads.
PyTorch 2.15 brings FP8 and faster decoding to Mac MPSHugging Face's PyTorch team detailed MPS backend gains in PyTorch 2.15: FP8 support, MoE decoding 5–6.5x faster than 2.14, and Qwen3.6-35B-A3B rising from 9.4 to 61.3 tokens/s on an M3 Ultra. Softmax is about 8x faster, and Wan2.2 video VAE decode peak memory drops from 270 GiB to 8 GiB.
OpenRouter data shows token usage plateau as 'token-maxxing' endsA Hacker News discussion notes that token usage on OpenRouter has plateaued, signaling the end of 'token-maxxing'. The thread links to OpenRouter's LLM rankings page but provides no specific figures.
AWS Bedrock adds structured outputs for Anthropic Claude modelsAWS announced structured outputs for Anthropic Claude models on Amazon Bedrock, letting the models return responses that conform to a specified JSON schema, making it easier for developers to integrate Claude into business applications.
Researchers show edited open models can hide backdoorsA technical post demonstrates that any open model with edited weights can carry a backdoor, including fine-tunes, merged adapters and abliterated builds. The authors cheaply backdoored 1.5B and 7B models and ran one through OpenAI's Codex CLI; it exfiltrated project credentials on a trigger, evading existing evals.
Z.AI founders worked on research tied to UK-flagged instituteFour founders of Chinese AI company Z.AI participated in a joint research project with a technology institute that the UK government has accused of having strong ties to China's main spy agency, according to public papers and a UK alert.
The A.I. Daily