Here are the notable topics from the overseas AI news we confirmed as of November 22 (Japan time, morning). Cloud-side AI assistants are stepping into the practical domain of ad operations, while an AI-involved proof surrounding an unsolved mathematical problem appears to be reaching a major milestone.

AWS Quick Gets a Meta Ads Connector — Create and Analyze Campaigns in Natural Language

AWS announced on its official blog that it has added an AI connector for Meta ads to its AI assistant, AWS Quick. The connector hands control of Meta's advertising MCP (Model Context Protocol) server over to Quick, letting advertisers create, manage, and analyze campaigns entirely through natural language interactions.

In the official demo, tasks such as generating an eight-slide launch plan from past campaign results, or building a dashboard that compares spend, lead count, and cost per lead across two ad sets, can be carried out with prompts alone. It can also combine Meta ads data with first-party and third-party data to build dashboards, documents, slides, and automated workflows.

What deserves attention is how write operations are handled. While the connector can actually make changes to ad accounts, it is designed so that Quick always asks for the user's permission before executing write actions such as creating campaigns or changing budgets. This governance design — leaving analysis and planning to the AI while humans keep the decision to execute — could become a reference case for how such an approach is received in a mission-critical, practical domain like marketing.

The head of marketing at fal.ai, an AI image generation service, commented that "being able to go from planning to execution in one place reduces the friction of switching between tools." Quick comes with a three-month free trial, and for the first three months, inviting up to 10 team members is free.

Riemann Hypothesis: The Collaborative Proof with Claude Passes Lean Community Verification

The case in which Levent Alpöge, a mathematician at Anthropic, claimed to have discovered a proof of the Riemann hypothesis — regarded as mathematics' greatest unsolved problem — in collaboration with the AI Claude has entered a new stage. According to ITmedia, Alpöge released a 17-million-line formal proof written in the proof assistant Lean, and the Lean community issued a joint statement certifying that "the proof is correct" after multiple rounds of verification.

In our previous roundup (November 21), we reported that a separate, independent proof claim had emerged at the same time, describing the situation as "competing claims side by side." As a follow-up development, the collaborative proof with Claude has now passed community verification. That said, this is still at the stage of verification and certification by the Lean community — how the usual academic peer review process will unfold, and whether the mathematical community as a whole will reach a consensus, remains unknown at this point.

Anthropic Co-founder Set to Announce Donation to "Trump Accounts"

According to Bloomberg, Tom Brown, Anthropic's co-founder and Chief Compute Officer, is set to announce a donation to "Trump accounts." U.S. Treasury Secretary Scott Bessent mentioned it on CNBC on Friday, and sources requesting anonymity confirmed the details. The amount and timing of the donation have not been reported.

A prominent figure at an AI company that only recently completed its IPO announcing a political donation can be read as one indicator of the distance between the AI industry and politics. That said, this appears to be a donation made by personal choice, and any connection to decision-making at Anthropic as a company is unknown at this point.

Other Topics

Apple's camera chief, who opposed AI photo alteration, has resigned**. DPReview reported that the executive who led the iPhone camera division has left Apple. The departure of someone who took a critical stance on AI-driven photo alteration and generation has become a talking point on Hacker News.

An essay examining the discourse around "conscious AI"**. Noema Magazine has published an article titled "The Mythology of Conscious AI." It explores the mythological dimensions of the debate over AI consciousness and has also been picked up on Hacker News.

A tool that pinpoints the "one byte" that breaks LLM prompt caching**. Kvdiff, released by a developer on GitHub, runs as a local proxy and shows, down to the byte, the exact position where prompt caching stops working. It looks useful for debugging cost spikes caused by cache misses.