AI Headlines:
November 20, 2025: Nvidia ushers in the era of applications, NPCs gain real memory, first “hallucination insurance” launches

Nvidia released its quarterly results yesterday evening after the market closed and surprised the markets this morning with a new strategic direction. CEO Jensen Huang announced that the market for pure training chips (used to create AI) is normalizing. The massive growth is now focused on inference (executing AI in everyday life).

The new product: Nvidia unveiled its new “Infero X” chip. This chip is not designed to train models, but rather to run billions of AI agents (like those announced by Microsoft, SAP, and Google) with extreme energy efficiency and lightning speed.

This confirms the week’s trend: away from building ever-larger models and toward the widespread use of specialized agents in business.

A consortium of game studios (including Ubisoft and EA) today unveiled “Project Mnemosyne” at a tech demo. This is considered the “holy grail” of role-playing games.

What’s new? Previous AI characters (NPCs) in games often “forgot” interactions as soon as the player left the room. The new technology gives NPCs a permanent, evolving memory.

If a player insults a merchant in the game, that merchant will remember it weeks later (in-game time), raising prices or refusing to sell. NPCs can build relationships with each other based on the player’s actions. The system uses small, local language models (SLMs) that run directly on the console/PC (similar to Apple’s on-device strategy) to avoid latency and save costs.

In direct response to the security risks posed by “shadow AI” (a topic of discussion yesterday) and the legal uncertainties (GEMA ruling last week), two of the world’s largest reinsurers today unveiled new policies.

The product: An “AI Performance Liability Insurance.”

What is insured? The policies cover financial losses resulting from “AI hallucinations” (errors/lies) or copyright infringements in business operations. For example, an AI purchasing agent might accidentally order 10,000 units instead of 1,000, or a marketing bot might use copyrighted music. To obtain insurance, companies must undergo rigorous audits and demonstrate that they are not using uncontrolled “shadow AI” and are adhering to official security standards (such as the UAP protocol introduced the day before yesterday).

November 20th illustrates the professionalization of the industry. Nvidia is supplying the hardware for continuous operation (Story 1), the gaming industry is using AI for deeper immersion rather than just graphics (Story 2), and the insurance industry is building the necessary safety net for companies (Story 3) so that AI becomes a calculable business risk.