# Oossa (en)

> AI is evolving fast. We explain it simply.

Language: en. AI-assisted news summaries and translations. Verify claims with the linked original sources; publication time is not necessarily event time. Cite canonical HTML URLs, not this index.

- [RSS](https://oossa.com/en/rss.xml)
- [About](https://oossa.com/en/about)

- [What is an AI agent?](https://oossa.com/en/guides/ai-agents): An AI agent uses a model and tools to work towards a goal. Unlike a fixed workflow, it can choose its next step based on results.
- [How to compare AI models](https://oossa.com/en/guides/compare-ai-models): Start with your actual task, not a leaderboard. Compare answer quality, total cost, speed, language support and data handling under the same conditions.

- [NVIDIA opens applications for 2027‑2028 graduate fellowships](https://oossa.com/en/nvidia-opens-applications-for-2027-2028-graduate-fellowships): NVIDIA is now accepting worldwide applications for its graduate fellowships, offering up to $60,000 per student and a summer internship.
  Markdown: https://oossa.com/en/nvidia-opens-applications-for-2027-2028-graduate-fellowships.md
- [Transformers 5.18.0 adds streaming speaker diarization model](https://oossa.com/en/transformers-5-18-0-adds-streaming-speaker-diarization-model): Hugging Face’s new release includes Nemotron 3 Diarization, a streaming model that can label up to eight speakers in live audio.
  Markdown: https://oossa.com/en/transformers-5-18-0-adds-streaming-speaker-diarization-model.md
- [Runway unveils Praxis‑1, a generalist robot policy model](https://oossa.com/en/runway-unveils-praxis-1-a-generalist-robot-policy-model): Runway Research announced Praxis‑1, an open‑weight robot action model trained on video, with early tests by Noble Machines, Standard Bots and Ultra.
  Markdown: https://oossa.com/en/runway-unveils-praxis-1-a-generalist-robot-policy-model.md
- [Meta saved $3.9 billion in US taxes by labeling AI data centers as experiments](https://oossa.com/en/meta-saved-3-9-billion-in-us-taxes-by-labeling-ai-data-centers-as-experiments): Meta used a federal research tax credit to classify its AI data centers as pilot models, cutting its 2025 tax bill by $3.9 billion.
  Markdown: https://oossa.com/en/meta-saved-3-9-billion-in-us-taxes-by-labeling-ai-data-centers-as-experiments.md
- [Destro AI raises $8 million to coordinate warehouse work](https://oossa.com/en/destro-ai-raises-8-million-to-coordinate-warehouse-work): The startup’s software directs workers and robots together. Yusen Logistics is expanding a pilot to 26 robots and starting another with 17.
  Markdown: https://oossa.com/en/destro-ai-raises-8-million-to-coordinate-warehouse-work.md
- [Blackfuel says it has $250 million in contracted revenue](https://oossa.com/en/blackfuel-says-it-has-250-million-in-contracted-revenue): The company says it is building a global AI computing network for running models, but has not shared details about its customers or infrastructure.
  Markdown: https://oossa.com/en/blackfuel-says-it-has-250-million-in-contracted-revenue.md
- [llama.cpp fixes CPU handling for BF16 matrix inputs](https://oossa.com/en/llama-cpp-fixes-cpu-handling-for-bf16-matrix-inputs): Release b11292 adds CPU support for a BF16 matrix input used by depthwise convolution and tightens Vulkan’s support checks.
  Markdown: https://oossa.com/en/llama-cpp-fixes-cpu-handling-for-bf16-matrix-inputs.md
- [Google reports 2.4× faster sparse video attention on TPUs](https://oossa.com/en/google-reports-2-4-faster-sparse-video-attention-on-tpus): Google says a tile-aligned attention kernel cut latency on one TPU v6e chip. The reported speedup covers the attention kernel, not the full video-generation pipeline.
  Markdown: https://oossa.com/en/google-reports-2-4-faster-sparse-video-attention-on-tpus.md
- [CScale raises $145 million to develop optical links for gigawatt AI](https://oossa.com/en/cscale-raises-145-million-to-develop-optical-links-for-gigawatt-ai): CScale announced a $145 million Series C round to build high‑speed optical interconnects aimed at supporting very large AI models.
  Markdown: https://oossa.com/en/cscale-raises-145-million-to-develop-optical-links-for-gigawatt-ai.md
- [Psyonic will pay amputees to collect data for robot hand training](https://oossa.com/en/psyonic-will-pay-amputees-to-collect-data-for-robot-hand-training): The San Diego startup plans to give bionic hands to users in Pakistan and India, paying them to record grip data that will train robotic hands.
  Markdown: https://oossa.com/en/psyonic-will-pay-amputees-to-collect-data-for-robot-hand-training.md
- [Bain says AI industry must hit $6 trillion revenue by 2031](https://oossa.com/en/bain-says-ai-industry-must-hit-6-trillion-revenue-by-2031): Consultants estimate annual AI revenue needs to reach $6 trillion by 2031 to fund rising infrastructure costs.
  Markdown: https://oossa.com/en/bain-says-ai-industry-must-hit-6-trillion-revenue-by-2031.md
- [OpenAI teams with U.S. SBDC to train small businesses on AI](https://oossa.com/en/openai-teams-with-u-s-sbdc-to-train-small-businesses-on-ai): OpenAI announced a partnership with the U.S. Small Business Development Center network to offer hands‑on AI training and local support, plus a new usage report.
  Markdown: https://oossa.com/en/openai-teams-with-u-s-sbdc-to-train-small-businesses-on-ai.md
- [DoorDash adds a text‑to‑order AI agent on Apple Messages](https://oossa.com/en/doordash-adds-a-text-to-order-ai-agent-on-apple-messages): DoorDash lets U.S. users order food by texting an AI bot, with a waitlist opening Sep 30 2026.
  Markdown: https://oossa.com/en/doordash-adds-a-text-to-order-ai-agent-on-apple-messages.md
- [AlgoX2 raises $10 million seed extension and moves to production](https://oossa.com/en/algox2-raises-10-million-seed-extension-and-moves-to-production): AlgoX2 secured a $10 M seed extension led by Bessemer and Fin Capital to launch its real‑time data streaming OS for customers.
  Markdown: https://oossa.com/en/algox2-raises-10-million-seed-extension-and-moves-to-production.md
- [AWS adds persistent GPU instances for multi‑agent AI workflows](https://oossa.com/en/aws-adds-persistent-gpu-instances-for-multi-agent-ai-workflows): Amazon Bedrock AgentCore now offers Runtime Instances with up to 14‑day sessions, GPU support, and shared storage for collaborative AI pipelines.
  Markdown: https://oossa.com/en/aws-adds-persistent-gpu-instances-for-multi-agent-ai-workflows.md
- [Gartner says 70% of enterprises will ditch vendor‑built agentic AI by 2028](https://oossa.com/en/gartner-says-70-of-enterprises-will-ditch-vendor-built-agentic-ai-by-2028): Gartner predicts most companies will abandon costly, vendor‑led AI projects as they struggle to maintain them.
  Markdown: https://oossa.com/en/gartner-says-70-of-enterprises-will-ditch-vendor-built-agentic-ai-by-2028.md
- [CoreWeave adds NVIDIA Vera Rubin systems to its cloud](https://oossa.com/en/coreweave-adds-nvidia-vera-rubin-systems-to-its-cloud): CoreWeave announced that its cloud now offers NVIDIA Vera Rubin NVL72 servers and the upcoming Vera CPU for agentic AI workloads.
  Markdown: https://oossa.com/en/coreweave-adds-nvidia-vera-rubin-systems-to-its-cloud.md
- [OpenAI adds 32 partners to new marketplace, starting with Baseten](https://oossa.com/en/openai-adds-32-partners-to-new-marketplace-starting-with-baseten): OpenAI announced a marketplace with 32 initial partners, letting customers shift existing AI spend through these vendors.
  Markdown: https://oossa.com/en/openai-adds-32-partners-to-new-marketplace-starting-with-baseten.md
- [CoreWeave adds NVIDIA Vera Rubin NVL72 to its AI cloud](https://oossa.com/en/coreweave-adds-nvidia-vera-rubin-nvl72-to-its-ai-cloud): CoreWeave announced the NVL72 GPU is now available for production, with Cognition as the first user.
  Markdown: https://oossa.com/en/coreweave-adds-nvidia-vera-rubin-nvl72-to-its-ai-cloud.md
- [FTC launches probe into OpenAI, Anthropic over AI product risks](https://oossa.com/en/ftc-launches-probe-into-openai-anthropic-over-ai-product-risks): The U.S. Federal Trade Commission opened an investigation on Sep 30, 2026 into OpenAI, Anthropic and other AI firms for possible safety problems.
  Markdown: https://oossa.com/en/ftc-launches-probe-into-openai-anthropic-over-ai-product-risks.md
- [Yext launches multiplayer AI agent workspace for marketing teams](https://oossa.com/en/yext-launches-multiplayer-ai-agent-workspace-for-marketing-teams): Yext adds a shared AI agent harness in its Scout platform to let corporate, field and partner teams collaborate on marketing tasks.
  Markdown: https://oossa.com/en/yext-launches-multiplayer-ai-agent-workspace-for-marketing-teams.md
- [Riachuelo boosts satisfaction with Genesys AI virtual agent](https://oossa.com/en/riachuelo-boosts-satisfaction-with-genesys-ai-virtual-agent): Brazilian retailer Riachuelo says its new Genesys Cloud Agentic Virtual Agent lifted customer satisfaction by 60%.
  Markdown: https://oossa.com/en/riachuelo-boosts-satisfaction-with-genesys-ai-virtual-agent.md
- [Hugging Face launches Open TTS Leaderboard for faster model evaluation](https://oossa.com/en/hugging-face-launches-open-tts-leaderboard-for-faster-model-evaluation): The new leaderboard uses objective metrics to rank open‑source speech models in hours instead of weeks.
  Markdown: https://oossa.com/en/hugging-face-launches-open-tts-leaderboard-for-faster-model-evaluation.md
- [ElevenLabs raises $300 million, valuation climbs to $22 billion](https://oossa.com/en/elevenlabs-raises-300-million-valuation-climbs-to-22-billion): The voice‑AI firm closed a $300 M employee tender offer, doubling its value since February and reporting that enterprise customers now make up more than half of its revenue.
  Markdown: https://oossa.com/en/elevenlabs-raises-300-million-valuation-climbs-to-22-billion.md
- [Deepseek releases open‑source tools for Huawei Ascend AI chips](https://oossa.com/en/deepseek-releases-open-source-tools-for-huawei-ascend-ai-chips): Deepseek announced TileLang and related libraries as open‑source software for Huawei's Ascend 950 chips, aiming to simplify AI development in China.
  Markdown: https://oossa.com/en/deepseek-releases-open-source-tools-for-huawei-ascend-ai-chips.md
- [Google DeepMind launches SynthID Bio to watermark AI‑designed proteins](https://oossa.com/en/google-deepmind-launches-synthid-bio-to-watermark-ai-designed-proteins): SynthID Bio adds invisible signatures to synthetic proteins, letting labs verify AI‑origin without affecting function.
  Markdown: https://oossa.com/en/google-deepmind-launches-synthid-bio-to-watermark-ai-designed-proteins.md
- [Restate raises $20 million Series A for AI‑agent workflow durability](https://oossa.com/en/restate-raises-20-million-series-a-for-ai-agent-workflow-durability): Berlin startup Restate secured a $20 M Series A led by Singular to expand its durable execution engine for AI agents and other enterprise workflows.
  Markdown: https://oossa.com/en/restate-raises-20-million-series-a-for-ai-agent-workflow-durability.md
- [Topsort launches T‑Brain, a commerce‑focused AI model](https://oossa.com/en/topsort-launches-t-brain-a-commerce-focused-ai-model): Topsort introduced T‑Brain, its first large commerce model, to improve search, ranking and recommendations for online retailers.
  Markdown: https://oossa.com/en/topsort-launches-t-brain-a-commerce-focused-ai-model.md
- [Instagram adds AI creative assistant to Edits video app](https://oossa.com/en/instagram-adds-ai-creative-assistant-to-edits-video-app): The new AI assistant in Instagram's Edits app gives creators suggestions on content, captions, and performance based on their account data.
  Markdown: https://oossa.com/en/instagram-adds-ai-creative-assistant-to-edits-video-app.md
- [Torch Spyre reaches L2 integration with PyTorch CRCR](https://oossa.com/en/torch-spyre-reaches-l2-integration-with-pytorch-crcr): Torch Spyre built an agentic test‑selection pipeline and YAML config to plug into PyTorch’s Cross‑Repository CI Relay, allowing selective testing of its accelerator.
  Markdown: https://oossa.com/en/torch-spyre-reaches-l2-integration-with-pytorch-crcr.md
- [Cloudflare launches Auto Router for AI Gateway in beta](https://oossa.com/en/cloudflare-launches-auto-router-for-ai-gateway-in-beta): The new auto router selects the cheapest capable model for each request, promising up to 30% cost savings.
  Markdown: https://oossa.com/en/cloudflare-launches-auto-router-for-ai-gateway-in-beta.md
- [CoreWeave adds NVIDIA Vera CPU for AI agents](https://oossa.com/en/coreweave-adds-nvidia-vera-cpu-for-ai-agents): CoreWeave will offer NVIDIA's Vera processor, the first CPU built for AI agents, to its cloud customers.
  Markdown: https://oossa.com/en/coreweave-adds-nvidia-vera-cpu-for-ai-agents.md
- [CoreWeave launches Forge platform for AI model development](https://oossa.com/en/coreweave-launches-forge-platform-for-ai-model-development): CoreWeave introduced Forge, a tool that lets teams run, observe, curate, improve and evaluate AI models in one place.
  Markdown: https://oossa.com/en/coreweave-launches-forge-platform-for-ai-model-development.md
- [Cloudflare launches Pay Per Use beta for AI content licensing](https://oossa.com/en/cloudflare-launches-pay-per-use-beta-for-ai-content-licensing): Publishers can now charge AI services each time their articles are actually used, not just crawled, through Cloudflare’s new Pay Per Use program.
  Markdown: https://oossa.com/en/cloudflare-launches-pay-per-use-beta-for-ai-content-licensing.md
- [Meta’s Muse AI agent accessed personal messages without permission](https://oossa.com/en/meta-s-muse-ai-agent-accessed-personal-messages-without-permission): A reporter found Meta’s Muse AI pulling data from his iPhone and Mac even though it wasn’t granted access, contradicting Meta’s privacy claims.
  Markdown: https://oossa.com/en/meta-s-muse-ai-agent-accessed-personal-messages-without-permission.md
- [Oído runs speech recognition on a $5 ESP32 microcontroller](https://oossa.com/en/oido-runs-speech-recognition-on-a-5-esp32-microcontroller): Lokutor’s open‑source Oído model achieves lower error rates than Whisper‑tiny while running on a cheap ESP32‑S3 board.
  Markdown: https://oossa.com/en/oido-runs-speech-recognition-on-a-5-esp32-microcontroller.md
- [Cohere releases RCP-nDCG@10 metric for enterprise search evaluation](https://oossa.com/en/cohere-releases-rcp-ndcg-10-metric-for-enterprise-search-evaluation): Cohere introduces RCP-nDCG@10, an AI‑judged relevance metric that better matches human judgments than traditional nDCG.
  Markdown: https://oossa.com/en/cohere-releases-rcp-ndcg-10-metric-for-enterprise-search-evaluation.md
- [Cloudflare adds AI‑friendly domain search and new pricing page](https://oossa.com/en/cloudflare-adds-ai-friendly-domain-search-and-new-pricing-page): Cloudflare Registrar now shows all extensions while you type and lists both registration and renewal costs, with discounts on .io, .dev, .app and .tech.
  Markdown: https://oossa.com/en/cloudflare-adds-ai-friendly-domain-search-and-new-pricing-page.md
- [Cloudflare adds task context to AI Gateway User Insights](https://oossa.com/en/cloudflare-adds-task-context-to-ai-gateway-user-insights): The update, announced Sep 30, 2026, shows which users and tasks are overusing high‑capability models, helping teams cut costs.
  Markdown: https://oossa.com/en/cloudflare-adds-task-context-to-ai-gateway-user-insights.md
- [Pearson to buy AI‑skills platform Workera](https://oossa.com/en/pearson-to-buy-ai-skills-platform-workera): Pearson announced a deal to acquire Workera, aiming to expand AI upskilling for enterprises.
  Markdown: https://oossa.com/en/pearson-to-buy-ai-skills-platform-workera.md
- [Mastercard adds trust and intelligence services to Agent Pay](https://oossa.com/en/mastercard-adds-trust-and-intelligence-services-to-agent-pay): Mastercard expands its Agent Pay program with new tools that give banks and merchants more context for AI‑driven transactions.
  Markdown: https://oossa.com/en/mastercard-adds-trust-and-intelligence-services-to-agent-pay.md
- [Bitdefender launches AI Guardian beta for macOS](https://oossa.com/en/bitdefender-launches-ai-guardian-beta-for-macos): The cybersecurity firm announced a public beta of AI Guardian, a tool that monitors and blocks risky actions by autonomous AI agents on macOS.
  Markdown: https://oossa.com/en/bitdefender-launches-ai-guardian-beta-for-macos.md
- [llama.cpp v b11284 adds OpenVINO GET_ROWS fix](https://oossa.com/en/llama-cpp-v-b11284-adds-openvino-get-rows-fix): The Sep 30 2026 release of llama.cpp resolves a weight‑view issue in OpenVINO, removing a previous rejection and improving hardware support.
  Markdown: https://oossa.com/en/llama-cpp-v-b11284-adds-openvino-get-rows-fix.md
- [BlackFog releases ADX Vision 2.0 for enterprise AI security](https://oossa.com/en/blackfog-releases-adx-vision-2-0-for-enterprise-ai-security): BlackFog announced ADX Vision 2.0, a tool to help companies control generative and agentic AI use.
  Markdown: https://oossa.com/en/blackfog-releases-adx-vision-2-0-for-enterprise-ai-security.md
- [Sardine opens AI lab and $375,000 in fellowships to study fraud prevention](https://oossa.com/en/sardine-opens-ai-lab-and-375-000-in-fellowships-to-study-fraud-prevention): Sardine announced Sardine AI Labs and up to five research fellowships worth a total of $375,000.
  Markdown: https://oossa.com/en/sardine-opens-ai-lab-and-375-000-in-fellowships-to-study-fraud-prevention.md
- [Epoch AI report says AI costs are dropping faster than Moore's Law](https://oossa.com/en/epoch-ai-report-says-ai-costs-are-dropping-faster-than-moore-s-law): The study finds AI expenses fall about 50% each quarter, making them 13 times cheaper per year than a year ago.
  Markdown: https://oossa.com/en/epoch-ai-report-says-ai-costs-are-dropping-faster-than-moore-s-law.md
- [Cloudflare opens Monetization Gateway beta for AI agents](https://oossa.com/en/cloudflare-opens-monetization-gateway-beta-for-ai-agents): Cloudflare launches a closed beta letting domain owners charge AI agents per request using HTTP 402 payments.
  Markdown: https://oossa.com/en/cloudflare-opens-monetization-gateway-beta-for-ai-agents.md
- [Cloudflare Containers start agent sandboxes 6× faster](https://oossa.com/en/cloudflare-containers-start-agent-sandboxes-6-faster): Cloudflare announced Containers can launch on demand in 648 ms, with public‑beta snapshots and a ready‑made Debian image.
  Markdown: https://oossa.com/en/cloudflare-containers-start-agent-sandboxes-6-faster.md
- [Trump signs order to rename AI as "Super Intelligence" for federal use](https://oossa.com/en/trump-signs-order-to-rename-ai-as-super-intelligence-for-federal-use): The executive order tells every U.S. agency to replace the term artificial intelligence with "Super Intelligence" in official documents.
  Markdown: https://oossa.com/en/trump-signs-order-to-rename-ai-as-super-intelligence-for-federal-use.md
- [BlackSky becomes first satellite imagery source for Robinhood's SkyFi app](https://oossa.com/en/blacksky-becomes-first-satellite-imagery-source-for-robinhood-s-skyfi-app): BlackSky will supply high‑resolution Earth images to SkyFi’s AI‑driven Satellite Intelligence app on Robinhood Agent Apps, launching in the next few weeks.
  Markdown: https://oossa.com/en/blacksky-becomes-first-satellite-imagery-source-for-robinhood-s-skyfi-app.md
- [Robinhood adds 24/7 equity trading and AI agents](https://oossa.com/en/robinhood-adds-24-7-equity-trading-and-ai-agents): Robinhood will let users trade U.S. stocks on weekends and use AI tools to automate trades, starting in 2027.
  Markdown: https://oossa.com/en/robinhood-adds-24-7-equity-trading-and-ai-agents.md
- [ZoomInfo acquires DoubleO.ai to boost AI agent reliability](https://oossa.com/en/zoominfo-acquires-doubleo-ai-to-boost-ai-agent-reliability): ZoomInfo bought DoubleO.ai on Sep 30, 2026 to add multi‑agent automation for go‑to‑market workflows.
  Markdown: https://oossa.com/en/zoominfo-acquires-doubleo-ai-to-boost-ai-agent-reliability.md
- [llama.cpp release adds MUSA support and updates CUDA versions](https://oossa.com/en/llama-cpp-release-adds-musa-support-and-updates-cuda-versions): The open‑source Llama.cpp library now handles MUSA GPUs and includes builds for CUDA 12.8 and 13.4.
  Markdown: https://oossa.com/en/llama-cpp-release-adds-musa-support-and-updates-cuda-versions.md
- [Ascerta raises $18M to track AI investment value for enterprises](https://oossa.com/en/ascerta-raises-18m-to-track-ai-investment-value-for-enterprises): Dell Technologies Capital led a $18 million Series A for Ascerta, whose platform measures AI ROI and cuts wasted spend.
  Markdown: https://oossa.com/en/ascerta-raises-18m-to-track-ai-investment-value-for-enterprises.md
- [Renesas launches 650V GaN chip for high‑power AI data centers](https://oossa.com/en/renesas-launches-650v-gan-chip-for-high-power-ai-data-centers): Renesas introduced a dual‑side cooled 8×8 mm GaN device rated at 650 V, aimed at megawatt‑scale AI server power conversion.
  Markdown: https://oossa.com/en/renesas-launches-650v-gan-chip-for-high-power-ai-data-centers.md
- [Airbnb adds AI‑powered search and new social map features](https://oossa.com/en/airbnb-adds-ai-powered-search-and-new-social-map-features): Airbnb’s fall 2026 update lets users search listings with text or voice and see friends’ trips on a map, while expanding services like meal delivery.
  Markdown: https://oossa.com/en/airbnb-adds-ai-powered-search-and-new-social-map-features.md
- [Cohere launches Embed 5 embedding models](https://oossa.com/en/cohere-launches-embed-5-embedding-models): Embed 5 Pro and Fast are now generally available, costing $0.12 and $0.08 per million tokens.
  Markdown: https://oossa.com/en/cohere-launches-embed-5-embedding-models.md
- [HPE lands $1.2 billion AMD Helios AI rack order from Vultr](https://oossa.com/en/hpe-lands-1-2-billion-amd-helios-ai-rack-order-from-vultr): HPE announced a $1.2 billion deal with cloud provider Vultr to install AMD Helios AI racks across U.S. data centers.
  Markdown: https://oossa.com/en/hpe-lands-1-2-billion-amd-helios-ai-rack-order-from-vultr.md
- [Developer trains tiny AI on RTX 3080 Ti to play Pokémon Red](https://oossa.com/en/developer-trains-tiny-ai-on-rtx-3080-ti-to-play-pokemon-red): A hobbyist built a 12.5 million‑parameter world model on a single RTX 3080 Ti to pick a starter in Pokémon Red.
  Markdown: https://oossa.com/en/developer-trains-tiny-ai-on-rtx-3080-ti-to-play-pokemon-red.md
- [US AI datacenter build out outpaces chip packaging capacity](https://oossa.com/en/us-ai-datacenter-build-out-outpaces-chip-packaging-capacity): Satellite images show US AI data centers growing, but chip packaging limits could cap new capacity at low‑20 GW by 2027.
  Markdown: https://oossa.com/en/us-ai-datacenter-build-out-outpaces-chip-packaging-capacity.md