EN · International edition · 19/10/2026
Inferama daily · 19/10/2026
The most important AI developments since the previous newsletter, selected for editorial relevance.
News coverage period: 18/10/2026 08:30 → 19/10/2026 08:30 (Europe/Madrid).
Assessment Rubrics: Definition, Design, and Limits When Scoring AI Outputs
A rubric organizes criteria and performance levels to make evaluations easier to interpret. This guide explains its components, distinguishes analytic from holistic rubrics, and examines how rubrics can be used—and where they fall short—in evaluating AI outputs, academic work, and machine-learning systems.
Read article →
Eleven v4 vs. v4 Turbo: How to Choose for a Voice Agent
ElevenLabs reports an average time to first audio of about 150 ms for v4 Turbo over WebSocket, excluding network latency. That is a provider-reported figure, not a guarantee of end-to-end response time. This guide explains how to measure the full interaction and choose based on the trade-offs between speed, expressive control, and consistency.
Read article →
Four Gemini 4 Argon agents beat Minecraft’s Ender Dragon in a 27-hour autonomous run
A Google DeepMind preprint describes four agents completing a survival game after writing 2,783 programs and dying 605 times. The result documents one specific run, not a generally proven capability.
Read article →
Anthropic consults law firms on negotiation norms for AI agents
Anthropic is consulting with law firms on how AI agents should negotiate on behalf of people. Available information does not identify the participating firms or detail agreed-upon rules, nor does it confirm how these guidelines will be integrated into Claude.
Read article →
Anthropic allocates 5% of its shares to a Public Benefit Fund: structure, incentives, and governance limits
The company has confirmed the creation of a fund receiving 5% of its shares, bringing the total commitment to 20% when combined with prior pledges from co-founders and employees. While the mechanism scales with valuation, experts warn that its actual veto power over technical decisions remains untested in the face of direct commercial conflicts.
Read article →
AesCode Infographics Layout: Visual Composition Metrics and Their Isolation in the Evaluation Suite
A functional analysis of the 'Layout' metric within the AesCode architecture. Unlike previous audits on 'Content' or 'Chart' that focused on the lack of access to the formula, this piece addresses the structural function of the metric: how separating layout evaluation (composition, hierarchy, spacing) from content evaluation (text, data) allows for diagnosing specific failures in infographic generation. It explains how an evaluation harness must isolate the 'design' variable from the 'informatio...
Read article →
Article 14 of the AI Act: Continuous Risk Management and Its Practical Application in High-Risk Systems
A legal-operational analysis of Article 14 of Regulation (EU) 2024/1689, focusing on the obligation to establish a continuous risk management system for high-risk AI systems, distinguishing between initial assessment and permanent monitoring, and providing a compliance checklist.
Read article →
More Parameters Are Not More Progress: Separating AI Release Timelines from Measurable Advancement
Release dates and model size are weak indicators of technical progress. This analysis examines three dimensions where timelines mislead: token efficiency versus raw scale, effective retention in long tasks versus maximum context window, and reliability in multi-step flows versus mere tool-calling capability. It proposes an evidence-based framework to distinguish real advancements from marginal iterations.
Read article →
Auditing the Evidence Chain in AI Papers: From Observation to Claim
A five-step protocol to evaluate the internal validity of AI research articles, distinguishing between observed data, author interpretations, and unsupported conclusions.
Read article →
From Model to Network: How to Expose a Local Inference Server Without Opening Random Doors
A practical guide to configuring network access for local inference servers such as Ollama, vLLM, and LM Studio. Learn how to manage bind addresses, firewalls, and authentication so you can share access over a local network or VPN without exposing the service to the internet.
Read article →