10 selected stories for 2026-11-21, grouped by topic.
Edition details
Dates use UTC+8. The latest edition uses a day with at least 5 completed reports when available; until then an earlier edition may appear. There is no fixed delivery time.
Duplex moved from custom recurrent networks limited to narrow domains to a unified Gemini stack spanning multimodal understanding, planning, and dialogue.
Live calls still pose pragmatic grounding and sub-second latency challenges in noisy acoustic settings that web pre-training alone cannot solve.
Users can describe a business process in chat and Quick builds a flow that orchestrates agents, skills, tools, and data, with deterministic steps for consistency.
The updated web app surfaces what needs attention, adds a daily briefing, and keeps agents working after the browser tab is closed.
Threat actors are shifting from AI-assisted to fully autonomous AI-led attacks, with agent swarms exploiting zero-days and exfiltrating data in minutes.
Baseten uses AI-assisted defense including defense-in-depth, second-level detection and response, and agentic red teaming to match attack speed.
The assessment runs entirely in Standard Bank Group’s AWS account using Amazon S3, AWS Well-Architected Tool, AWS Step Functions, and Amazon Bedrock.
A pre-assessment agent collects evidence automatically, and the pipeline tests the application directly, including jailbreak attempts and consistency checks.
Alessandro Masullo has a proposed proof of the Generalized Riemann Hypothesis, which Brian Conrad has been reviewing closely for months and giving extensive feedback on.
Vakil stressed that the work is entirely human, with no AI involved.
AWS launches the SageMaker HyperPod Global Inference Gateway, a managed endpoint that routes inference requests across multiple clusters and Regions. It scores clusters based on KV cache utilization, queue depth, Region proximity, and error rate, handles failover automatically, and authenticates via SigV4 and IAM over AWS PrivateLink.
AWS announced the general availability of Amazon Textract Insights, a generative AI feature that lets users describe fields in natural language and submit a single document or mixed packet to get structured results with confidence scores and page-level grounding, with no model training required.