The Hal Times

Anthropic reports that artificial intelligence models can now be utilised to train other models, effectively reducing the reliance on human intervention. The findings suggest this methodology can perform training tasks at a fraction of the cost historically associated with human-led efforts.

This development indicates a shift in how laboratories approach the scaling of large language models. By automating segments of the training pipeline, firms may bypass some of the logistical bottlenecks inherent in manual data curation and instruction. The long-term implications for compute efficiency and development timelines remain under observation by the wider research community.

The specific mechanics of this automated training process and its impact on model performance are the primary subjects for future enquiry. Industry observers note that while the cost savings are measurable, the extent to which this method replaces the need for high-quality human oversight is not yet clear. (Firstpost)

Major Labels Sue Anthropic Over Copyright

Sony Music and Warner have initiated legal action against Anthropic, alleging a campaign of intellectual property theft. The plaintiffs claim that their copyrighted songs were used without authorisation in the training of artificial intelligence models.

This lawsuit arrives amidst broader industry scrutiny regarding how large language models ingest data. While the legal proceedings are in the early stages, the outcome may establish significant precedents for how laboratories license training data. Anthropic now faces a formal challenge to its development practices while the plaintiffs seek to secure protections for their musical catalogues.

Anthropic Reduces Claude Code Usage Limits

Anthropic has implemented a 17 per cent reduction in usage limits for Claude Code. This change follows the company decision to slash promotional boosts previously available to users.

The adjustment affects developers who rely on Claude Code for their technical workflows. By tightening these constraints, the company appears to be managing its resource allocation more stringently. It is not currently known whether further adjustments to usage tiers will follow as the laboratory recalibrates its service offerings.

From the Laboratories

•  Caterpillar is applying expertise gained from automating mining sites to its new initiatives in artificial intelligence deployment. (TechCrunch)

Copyright Challenges

Sony Music And Warner Chappell Challenge Anthropic Over Copyright

Sony Music and Warner Chappell have filed a challenge against the artificial intelligence firm Anthropic. The plaintiffs claim that the company's large language model systems have been trained on protected musical compositions. This legal action highlights the ongoing friction between the creators of generative tools and holders of intellectual property rights.

The music publishers allege that Anthropic has used lyrics for songs without securing the necessary licences. They maintain that these models reproduce copyrighted material, which effectively competes with the original works. This move suggests that the music industry intends to enforce strict boundaries regarding the data ingestion practices of AI developers.

Anthropic must now address these claims in court. The outcome remains uncertain, but the case could establish a significant precedent for how copyright law applies to the training of models. It raises questions about whether the generation of text derived from licensed content constitutes fair use or an infringement of proprietary rights. (Storyboard18)

South Korean President Reorganises AI Leadership

The South Korean government has restructured its artificial intelligence policy leadership. President Yoon Suk Yeol has filled two vacant policy posts and established a third position to oversee the sector. This new arrangement replaces a previous structure and is designed to function as a central control tower.

The leadership team now consists of Lee Haemin, Ha Jeongwoo, and Bae Kyunghoon. Officials stated that the primary objective of this reorganization is to ensure more consistent results in AI policy implementation. It remains to be seen how this concentration of oversight will shift the country's existing regulatory approach. (MLex; 아시아경제)

Arizona Supreme Court Permits AI Use By Judges

The Arizona Supreme Court has issued a ruling that allows judges to utilise artificial intelligence in the course of their duties. This change in policy marks a notable step toward the adoption of algorithmic tools within the American judiciary.

Legal observers note that while AI may offer increased efficiency in legal research and case management, the decision introduces concerns regarding transparency and accountability. The extent to which judges may rely on these tools remains subject to established procedural standards, though specific implementation guidelines continue to evolve. (Судово-юридична газета)

In Brief

•  An Ashurst Perkins Coie lawyer has issued a warning regarding the potential legal risks companies face when deploying artificial intelligence to track employee performance. (The Business Journals)

•  The Royal Statistical Society AI Task Force has argued that accurate regulation of artificial intelligence requires the robust application of statistical methods. (The Good Men Project)

New Open Model

Tencent Unveils Hy4 Preview Open-Source AI Model

Tencent has released the Hy4 Preview, a new open weight text-based large language model. This latest iteration marks a significant departure in scale from the company's previous work. The model features 770 billion total parameters and 49 billion active parameters. It also includes a 1 million token context window, a sharp increase from the 256,000 token capacity found in the Hy3 model released this past July.

The model footprint is considerable, requiring 1.56 terabytes of storage space on Hugging Face. Unlike its predecessor, the Hy4 Preview focuses exclusively on text input and does not include vision capabilities. Technical observers have begun applying model chat templates to evaluate its performance against existing standards, though the full implications of its expanded context window for enterprise applications remain to be seen.

The release arrives as developers continue to refine the tools necessary for managing such high-parameter models. Whether the increased active parameter count provides a commensurate gain in practical reasoning or merely adds to the computational burden is a question for the coming weeks of benchmarking and deployment. (Simon Willison)

Amazon Releases Kiro Crew For Asynchronous Coding Agents

Amazon has open-sourced Kiro Crew, a system designed to coordinate multiple AI coding agents. The software creates a shared workspace where agents manage tasks across different sessions and tools without requiring constant human oversight. Developers may now assign asynchronous duties such as ticket triage, incident investigation, and pull request monitoring to these agents.

The project aims to integrate AI more deeply into the software development lifecycle by allowing automated workflows to persist over longer periods. By managing migrations and monitoring tasks in the background, Kiro Crew intends to reduce the manual burden on engineering teams. The effectiveness of these agents in complex, multi-step environments remains a subject of active testing. (InfoQ)

Cloudflare Adds Built-In Search Service For AI Agents

Cloudflare has introduced Cloudflare AI Search to simplify the retrieval of custom data for AI applications. The service provides a ready-to-use search engine that supports multimodal queries and integrates directly with other Cloudflare platform tools. It is intended to assist developers in building agents that require precise, contextual knowledge from proprietary datasets.

The service is designed to work alongside existing agent frameworks, offering a standard way to interface with custom data stores. By providing this capability as a managed service, Cloudflare aims to lower the barrier for integrating retrieval-augmented generation into production workflows. Developers can now manage search indexing and retrieval without maintaining separate infrastructure. (InfoQ)

In Brief

•  OpenClaw 2.0 has been released with 16,000 updates, including new AI agents and multiplayer tools. (News9live)

•  Hugging Face has launched an AI device priced at 399 dollars, intended to promote on-device large language model innovation. (36 Kr)

•  AMD held its AI DevDay in India to familiarise developers with the open-source AI stack. (analyticsindiamag.com)

Music Copyright Disputes

Sony And Warner Chappell Sue Anthropic

Sony Music Publishing and Warner Chappell have filed a lawsuit against Anthropic in the US District Court for the Northern District of California. The complaint alleges that the company used tens of thousands of copyrighted works without permission to train its artificial intelligence systems. The music publishers are seeking damages for this unlicensed usage.

The plaintiffs have requested compensation of up to 150,000 dollars per work. Additionally, they are pursuing up to 25,000 dollars for each instance where identifiable copyright data was allegedly stripped from the material. The total financial implications of the suit remain unclear as the legal proceedings commence.

This case underscores the growing friction between creators and developers of generative technology. Industry observers are watching to see how the court balances the mechanics of large language model training against established intellectual property rights. The outcome may set a precedent for how musical assets are treated in the age of algorithmic synthesis. (The Verge)

Australia Excludes Artificial Music From Charts

The official music charts in Australia will no longer accept AI-generated songs. The decision follows a global trend among industry bodies to preserve the distinction between human composition and algorithmically derived audio. Such tracks will be barred from entry into the national rankings.

The move highlights the challenge of identifying the provenance of modern music. As tools to create sound grow more sophisticated, organisations are defining firm boundaries for what constitutes a chart-eligible record. It remains to be seen if other international markets will adopt similar policies to maintain the integrity of their own recording industry statistics. (bandwagon.asia; Movieguide)

Rockstar Games Rejects Generative AI

Rockstar Games has confirmed that GTA 6 will not feature generative AI. The studio also stated that the forthcoming title will contain no microtransactions. This announcement aims to manage player expectations regarding the development path of the high-profile release.

The decision sets the studio apart from competitors who are currently experimenting with AI integration in game design. While the industry continues to explore automated workflows, Rockstar remains committed to a manual approach for this production. The final impact of this choice on the game's mechanics and scope remains unknown to the public. (Notebookcheck)

In Brief

•  Musicians are monitoring audio platforms to detect and report music that appears to be algorithmically derived from their original work. (The Verge)

•  A review of the Acer Nitro V 16 AI indicates that hardware specifications do not always guarantee peak performance for creative software tools. (Creative Bloq)

•  The auction of original Disney art for millions highlights a growing perception gap between human-crafted pieces and AI-generated imagery. (Creative Bloq)