AI news for builders and product teamsUpdated Nov 17, 2026, 21:01 UTC
AI news, straight from the source.
Releases, research, and ideas for developers and product teams. Updated every hour.
Across the feeds
Newest first
Indie studio Alliance Arts delayed the public Steam Next Fest demo of Drunken Goddess Reflux, citing fears that AI tools could copy its presentation and gameplay and release clones before its own launch.

TwelveLabs' Pegasus 1.5 video language model is now on Amazon Bedrock, adding time-based segmentation that returns every matching video segment as timestamped JSON, plus multimodal prompting with reference images and support for videos up to two hours in a single request.

Deepgram and AWS ported the Nova-3 speech-to-text model to Trainium2 for offline batch workloads, running eight model replicas on one trn2.3xlarge instance and processing 11–70 percent more audio per dollar than the compute options Deepgram had been running.

Developers are sharing AI-agent-driven scripts, including "dumbtv" and "cleanyourtv," that strip ads, tracking and bloatware from Android TV, Google TV and Fire TV sets without root access. Some users report successes like removing 25 bloat apps, while others say the process can fail or brick devices.

AWS contributed a push-based NIXL KV connector to vLLM that overlaps decode-side setup with prefill, reporting up to 3x better time-to-first-token and about 30 percent better time-per-output-token versus the default pull mode on a Qwen3-32B benchmark.

Nvidia released open-weight Nemotron models: the 120-billion-parameter Nemotron 3.5 Super VL for enterprise agents and two lightweight Bolt models for local agent workloads on RTX Spark AI PCs and DGX Spark workstations. Super VL and Bolt 74B-A9B are available now; Bolt E4B arrives later this month.

Numan built Nu, an AI health coach on LangGraph and LangSmith, and found that carefully authored evaluation criteria, not the agent or its prompts, were the decisive asset. Its per-category rubrics gate releases on clinical safety, and a simulated-conversation layer is in prototype pending clinician sign-off.

Character.AI filed for Chapter 11 on October 12 as part of a two-step deal that let Disney hire most of its staff, including former CEO Karandeep Anand, and bid for a broad license to its technology while the remaining company keeps its legal liabilities.

Illinois Gov. JB Pritzker said on Decoder that his state is competing hard with Texas, Florida and others for data center and corporate investment, while warning that AI could push US unemployment from about 4 percent to 7 or 10 percent without preparation.

Norforge found that a hacker used mostly autonomous AI agents to breach more than 100 websites; Reuters names John Deere, SharkNinja and Cummins among the victims, though Norforge itself did not identify targets.

At the IMF meeting in Bangkok on 17 October, Anthropic co-founder Dario Amodei argued that AI is now doing frontier science and that the "singularity" has begun, citing benchmark jumps such as FrontierMath rising from 2% to 94% and Humanity's Last Exam from 8% to 67.7%.

Laplace Intelligence, a University of Michigan spinout, is building automated red-teaming for robots and autonomous vehicles, using AI agents inside interactive simulations to find failure cases before deployment. The Ann Arbor startup is profitable, counts Bot Auto and a major Japanese automaker as customers, and pitches at Startup Battlefield on October 13.

Democrats could win control of Congress in November's midterms, and key committee leaders have signaled they would press AI companies on data center tax credits and recent hacking incidents.

OpenAI says it is improving ChatGPT's coding performance in low-resource programming languages and has added evaluations covering software engineering, SQLite and COBOL migration.

Industry executives expect only seven to ten of the hundreds of AI cloud providers to survive, a shakeout that threatens Nvidia because non-hyperscaler customers account for 45 percent of its data center revenue.

IEEE Spectrum has published a sponsored white paper, backed by Intel, arguing that cities should move from passive smart-city monitoring to "cognitive cities" built on edge and physical AI. It covers tiered edge architecture, computer vision and agentic systems, data sovereignty and interoperability requirements for public safety, mobility and infrastructure.

Google released EmbeddingGemma 2 under the Apache 2.0 license, a multimodal embedding model based on Gemma 4 that maps text, code, images, video and audio into one shared 768-dimensional space for on-device search and RAG.

MIT CSAIL researchers left copies of Claude Opus 4.7, nicknamed Eko, alone on a simulated island for 30 hours, where the agents invented games and set their own goals—and later outperformed fresh agents on specific tasks. The experiment, described in a not-yet-peer-reviewed paper, proposes "playful learning" as a third training approach alongside supervised and reinforcement learning.

Simon Willison released llm-tools-datasette 0.1.1, fixing a breakage caused when LLM 0.33 moved to httpx2 and plugins that used httpx without declaring a direct dependency stopped working.

Disney has agreed to make job offers to at least 60 Character.AI employees, the majority of its workforce, at an estimated $60 million in annual compensation, after the chatbot startup filed for bankruptcy. The arrangement is a stalking horse bid, so a higher offer could still displace it.