The Hal Times

Anthropic investors are discussing a $1.5 trillion valuation for the AI company, a figure that represents a $500 billion decline from earlier estimates. The valuation reflects the company's recent performance and market conditions, though investors remain cautious about the valuation's sustainability.

The shift in valuation underscores broader concerns within the AI sector, where high-profile companies are facing scrutiny over their financial metrics. Anthropic's valuation is now seen as a reflection of its growth trajectory and strategic positioning, rather than a sign of overvaluation.

The company has not yet commented publicly on the valuation, but the discussion highlights the challenges of maintaining investor confidence in the face of rapid technological and market changes. The $1.5 trillion figure remains a point of contention, with some investors questioning whether it adequately reflects Anthropic's long-term potential.

The valuation also comes amid growing interest in AI policy and regulation, particularly in regions like Singapore, where Anthropic has been active in shaping AI governance frameworks. The company's financial performance will be closely watched as it navigates both market pressures and regulatory developments. (Seeking Alpha)

Anthropic Opens Claude to METR Researchers

Anthropic has granted METR researchers employee-level access to its AI models, including Claude, allowing them to conduct independent assessments. METR will publish its own evaluations of Anthropic's risks, incidents, and practices, marking a significant step in AI transparency.

The move comes as part of a broader trend in the AI industry, where companies are increasingly opening their models to external scrutiny. Anthropic has given up the right to block METR's research, ensuring that its assessments are not subject to the company's influence. (TechCrunch)

Meta Launches Muse Spark Open on Apple Silicon

Meta has launched Muse Spark Open, an open-weight AI model designed to run locally on Apple silicon. The model is part of Meta's broader strategy to make AI more accessible and efficient, particularly for developers working on Macs.

Muse Spark Open is a multimodal mixture-of-experts model that combines chat, decision-making, and speech capabilities into a single endpoint. The launch reflects Meta's growing focus on open-source AI solutions, which are increasingly popular among developers and researchers. (TechCrunch; Hugging Face)

From the Laboratories

•  Anthropic has partnered with Singapore to offer AI policy expertise, as the city-state seeks to establish itself as a leader in AI governance.

•  UK small businesses are increasing their spending on Anthropic, though OpenAI remains a close competitor in the AI tools market. (TechRadar)

•  NVIDIA has released a 2.8 trillion-token dataset for Indic language pretraining, aiming to improve AI models' performance in regional languages. (Hugging Face)

•  OpenAI has introduced the Rosalind Academic Plan, offering sponsored access to GPT-Rosalind for academic research in life sciences. (OpenAI)

Judicial Oversight

Court Rules Congress Must Address AI Traffic Crisis

A recent court ruling has highlighted the growing friction between artificial intelligence search tools and digital publishers. The court determined that the crisis regarding web traffic necessitates action from Congress to resolve. This decision places the burden of policy adjustment directly on lawmakers rather than leaving the industry to self-regulate.

In response to these mounting pressures, Google has begun adding publisher links to its AI search features. The move aims to provide a clearer path for users to visit source material. It remains to be seen whether this adjustment will satisfy the legal requirements or quell publisher concerns.

The ruling underscores the lack of current framework for how AI search engines should attribute and direct traffic to original creators. Without clear legislative guidance, publishers face an uncertain future regarding their digital revenue models. The involvement of Congress will be watched closely by both technology firms and media companies to determine the scope of future compliance. (Tech Times)

Apple Ordered To Produce AI Hiring Records

A court has ordered Apple to produce hiring records related to its artificial intelligence division. This order emerges from an ongoing trade secrets dispute. The documents are intended to assist the court in evaluating whether the company inappropriately acquired sensitive information through its recruitment processes.

The case hinges on whether internal hiring practices crossed the line into misappropriation of intellectual property. Apple must now submit internal files for judicial review. The outcome could set a precedent for how technology companies manage the movement of personnel between competitors while protecting proprietary development techniques and trade secrets. (Yahoo Finance Singapore)

Anthropic Partners With Singapore On Policy

Anthropic has entered into an agreement with Singapore to provide expertise on artificial intelligence policy. The partnership aims to assist the nation in refining its regulatory approach to emerging technologies. Representatives from the company will work alongside local officials to identify safe implementation strategies.

This collaboration reflects a broader trend of AI developers engaging directly with governments to shape legislative standards. While the specific nature of the advisory role remains broad, it suggests a move toward shared responsibility for safety and ethics. Both parties intend to use this pact to ensure that AI integration respects national security and public interests. (World Business Outlook)

In Brief

•  Hong Kong has inaugurated an Embodied AI hub in an attempt to attract global robotics firms to the region. (Crypto Briefing)

•  The Supreme Court has flagged the presence of AI-generated fake judgments within a legal compilation while upholding an insolvency admission. (LawChakra)

•  Experts have called for Australia to increase the resources and authority allocated to its AI Safety Institute. (The Nightly)

•  Turkmenistan has formally requested the establishment of international rules to govern AI regulation. (vestiabad.ru)

•  Sci-Fi Valley Con in Altoona has introduced a policy prohibiting AI-generated artwork from its vendors. (WPSU)

New Open Model

Meta And Apple Release Muse Spark Open

Meta and Apple have unveiled Muse Spark Open, a new multimodal model based on a mixture-of-experts architecture. The model is currently available as an open-weight release, marking a significant entry from the two companies into the open-weights landscape. It is positioned as a capable model for developers and researchers looking to integrate advanced multimodal capabilities into their own infrastructure.

The release is accessible through the official Meta research blog and the Hugging Face platform. Detailed documentation on the model's architecture and training methodology has been provided alongside the open-weight files. These resources aim to assist the developer community in evaluating the performance of the model across various tasks.

While the release offers new possibilities for self-hosting and custom applications, observers note that the broader regulatory environment remains uncertain. Recent commentary suggests that future administrations may scrutinise open-source models for potential intellectual property concerns. The long-term implications for open-weight development, and the extent to which these models will be subject to state-level sanctions, are not yet clear. (Hugging Face; Research Meta; Stocktwits)

IBM Updates Granite Switch Model

IBM has introduced Granite Switch 4.1, a model designed to handle chat, decision-making, and speech tasks from a single endpoint. This approach aims to streamline the development process for those relying on IBM's Granite architecture.

The update is now listed on the Hugging Face blog. It is intended for engineers who require a consolidated interface for multiple artificial intelligence functions, potentially reducing the overhead associated with managing disparate models for speech and text generation. (Hugging Face)

Research Details Developer Adoption Patterns

A recent presentation by Brendan O'Leary highlights how developers adopt agentic workflows, based on an analysis of 40 trillion tokens of usage data. The research suggests that building trust is more critical to adoption than chasing the highest model benchmarks.

The analysis identifies four distinct rungs on the AI adoption ladder. It also examines common failure points, including multi-model routing and context construction. These findings are intended to assist engineering leaders in scaling their agentic systems by addressing the practical hurdles encountered during real-world implementation. (InfoQ)

In Brief

•  The latest version of the image generation model, FLUX.3 Image Klein, is now available for use within Diffusers. (Hugging Face)

•  New documentation details a pipeline for data generation and pretraining on 2.8 trillion tokens to support Indic languages. (Hugging Face)

New Industry Standards

YouTube Music Introduces Labels For Artificial Intelligence Songs

YouTube Music has commenced the deployment of labels intended to identify songs created with the assistance of artificial intelligence. These notifications are currently appearing within the music player and are expected to extend to the broader YouTube platform in the near future. The move follows growing concern regarding the transparency of machine-generated content in the music industry.

By clarifying the origin of a track, the service aims to inform listeners whether a composition is the work of a human artist or an automated system. While the labels provide a measure of clarity for the audience, the company has not yet detailed how it verifies the presence of artificial intelligence in submitted files.

The implementation arrives alongside broader efforts by industry participants to distinguish between human and machine-made creative works. As streaming services grapple with an influx of algorithmic content, such labelling remains the primary method for maintaining a distinction between professional studio recordings and synthetic music.

The effect of these labels on consumer behaviour remains to be seen. Whether listeners will treat AI-generated tracks with the same regard as traditional music is an open question. For now, the move serves as a recognition that the provenance of creative media has become a significant matter for the modern listener. (Ars Technica; Notebookcheck)

Sega Withdraws Total War Shogun 2 Remaster

Sega has removed the Total War, Shogun 2, Complete Edition from digital storefronts following a poor reception from players. The remaster, which launched on Tuesday, bundled the original 2011 strategy game with its expansions. Users reported significant technical difficulties, including software bugs and poor visual quality.

The negative response centred on elements of the game that appeared to have been enhanced by artificial intelligence. Upscaling features often associated with these tools failed to meet expected standards, leading to the decision to pull the release from sale entirely. (The Verge)

Adobe Collaborates With Nvidia On RTX Spark

Adobe has confirmed that it is optimising its software suite for RTX Spark. This is the new AI-focused Arm chipset released by Nvidia, designed specifically for laptops and compact computing devices. The collaboration suggests a deeper integration between Adobe’s professional tools and Nvidia’s current hardware architecture.

This update aims to improve performance for creative professionals who rely on mobile workstations. While the partnership highlights a commitment to the personal computer market, the long-term impact on the speed and capability of Adobe’s generative features in a mobile environment remains a subject of active development. (PetaPixel)

In Brief

•  Filipino filmmaker Vincent Soberano has received awards for his film Secret Apartment, which utilises artificial intelligence production techniques. (GMA Network)

•  Pixieset has introduced an on-device AI photo editor to assist photographers with culling and batch-editing tasks. (PetaPixel)

•  The American Film Market is convening a Global Screen Economy Summit in November to discuss the influence of artificial intelligence on the industry. (Hollywood Reporter)

•  A Korean startup is developing tools for copyright tracking to assist in the creation of legal AI music. (The Korea Times)