English Summary:

Apple has filed a motion in federal court seeking court-supervised settlement talks with Epic Games to resolve their long-running legal battle. This move came on the same day Apple proposed a new fee structure for purchases made through links to payment systems outside the App Store.

The proposed fees include a 15% commission for standard apps, 10% for specific programs and subscription renewals, and 5% for apps in its Small Business Program, with Apple arguing these are fair compensation for the tools and technology provided to developers. However, Epic quickly pushed back, stating these fees are "far outside of the bounds of the Ninth Circuit's guidance on permissible fees."

Concurrently, Google is facing scrutiny over its implementation of a separate court order requiring it to open Google Play to competing app stores. A U.S. District Judge ordered Google to fix issues where rival app stores were not easily discoverable or installable, citing several extra steps and poor search results.

The legal disputes date back to 2020 when Epic sued both tech giants after its game Fortnite was removed from their app stores for bypassing their payment systems. While Epic largely lost its case against Apple, a jury found Google violated antitrust law in 2023.

English Summary:

Anthropic has announced that future Claude models will generate text containing an undetectable watermark, a measure implemented to comply with the EU AI Act. This watermarking serves as a method to determine the likelihood of Claude's involvement in writing a given text.

The company emphasizes that this watermarking method has no practical impact on the quality, creativity, or readability of Claude's outputs, and is indistinguishable to readers. It does not add anything to the text, contains no hidden characters, requires no extra tokens, and does not increase costs.

Anthropic's chosen watermarking approach is a version of Google DeepMind's SynthID-Text, which leverages "low-stakes choices" made during the word generation process. By subtly altering the source of randomness in these choices, it embeds a pattern that is undetectable to humans but verifiable with a key.

Importantly, the watermark carries no identifying information and cannot be traced back to a specific person, organization, or chat, thus protecting user privacy. However, limitations exist, such as reduced effectiveness on small text samples and less applicability in highly factual passages or code where word choices are constrained.

Beyond text, when Claude produces supported file types like images, it attaches a content credential in the form of C2PA metadata. This open industry standard cryptographically signs a note in the file's metadata, indicating Claude's involvement in its creation or processing, without altering the file content or including identifying user information.

English Summary:

This URL was inaccessible, therefore a full summary of the article's content cannot be provided.

Based on the article's title, "Amazon Can Use Your Twitch Content to Train Its AI—Unless You Opt Out," it suggests that Amazon intends to utilize user content from its live-streaming platform, Twitch, for training its AI models. This raises significant concerns regarding user data privacy and content ownership.

The article likely details how users can opt out of having their content used for AI training. Such an opt-out mechanism provides users with a degree of control over their data usage, though the default settings and clarity of the opt-out process would be crucial considerations.