AI News, 14 November 2026

Saturday 14 November

Evidence-Linked Memory Cuts LLM Degradation

Evidence-Linked Memory Cuts LLM Degradation

AI-generated summary Researchers from the Korea Advanced Institute of Science and Technology and DeepAuto.ai published a preprint introducing WikiTrace, an evidence-linked memory architecture for large language model agents. The evaluations utilized a Pi terminal agent driven by the DeepSeek-V4-Flash model operating under high reasoning effort across eight test suites.

Earlier:

Thousands March in Edinburgh Demanding Consultation on Scotland's 23 Data Centre Plans

Thousands March in Edinburgh Demanding Consultation on Scotland's 23 Data Centre Plans

AI-generated summary Thousands of demonstrators marched in Edinburgh to demand a public consultation regarding twenty-three proposed data centres for artificial intelligence. Friends of the Earth Scotland states that the proposals would add 9.1GW to peak demand, while Scottish ministers have offered only new planning guidance before the end of the year.

SteerDistill Quantizes LLM Agents via Steering

SteerDistill Quantizes LLM Agents via Steering

AI-generated summary Researchers from Tsinghua University and the University of Science and Technology of China introduced SteerDistill to address performance degradation when compressing large language model agents to low precisions. The evaluation across five benchmarks demonstrated that SteerDistill outperformed competing compression baselines while staying close to the unquantized ceiling.

Did IDC’s 20% Command-and-Control Prediction Hold?

AI-generated summary In October 2024, IDC forecast that organizations relying exclusively on command-and-control leadership would see a 20 percent drop in profitability by 2026 due to a lack of AI innovation and adaptability. Amy Loomis tested the prediction against survey data measuring AI returns and found that leaders redirected the time saved by AI into new work practices and innovative products.

Gemini 4 Hydrogen is Google's new workhorse for AI agents, and what European teams should check first

AI-generated summary Google launched Gemini 4 Hydrogen on 12 November 2026 to replace Gemini 3.8 Flash in the Gemini app, Google Search and the Gemini API. Artificial Analysis reports that Gemini 4 Hydrogen scores 51 on the Intelligence Index and costs $1.50 per million input tokens.

Also covered bySilicon AI NewsWritebleLotook
Earlier:

Claude outage on November 13: what happened and what is still affected

AI-generated summary Anthropic reported that Claude suffered two separate outages on November 13 involving authentication and degraded performance. The morning incident blocked sign-ins to claude.ai and Claude Code before a fix deployed, while the afternoon incident affected the Claude API, connectors and mobile apps.

Earlier:

Building real-time agents for EV charging using Gemma and TPUs

AI-generated summary Google Cloud detailed how European electric vehicle charging AI platform lemonflow engineered real-time voice and network monitoring agents using open Gemma models hosted on dedicated Cloud TPUs. lemonflow migrated inference onto self-hosted open Gemma models running on dedicated Cloud TPUs within European Union regions to achieve sub-800-millisecond conversational response times and strict data sovereignty.

AssemblyAI Uses AWS Spot Instances for Cost-Effective Speech-to-Text Inference

AI-generated summary AssemblyAI uses a mixed fleet of Amazon EC2 Spot Instances and On-Demand Instances across multiple NVIDIA GPU instance families for speech-to-text inference. According to production deployment results, AssemblyAI achieved 70 percent cost savings on inference compute by using Amazon EC2 Spot Instances with GPUs.

Also covered byData Phoenix