Tencent Launches Hy3, a 295B-Parameter Mixture-of-Experts Model for Agentic Workflows

Tencent has unveiled Hy3, a 295 billion-parameter Mixture-of-Experts model designed for reasoning and agentic workflows, positioning itself as a strong competitor to trillion-scale models. The model features 21 billion active parameters with 192 experts and offers a 256K context window aimed at long-horizon tasks such as coding, document processing, and financial analysis. Hy3 is built to mitigate hallucination issues by providing grounded responses and is available for commercial use under the Apache 2.0 license. Users can access a free API for two weeks, enhancing its affordability for various applications.

Read Full Article

View All For This Day

AIEnterprise Adoption+2

AIEnterprise AdoptionXboxTech Industry

Microsoft Announces Layoffs of Nearly 5,000 Employees, Impacting Xbox and Commercial Sales

Microsoft has announced the layoff of approximately 4,800 employees, equating to 2.1% of its global workforce, with significant impacts on its Xbox division and commercial sales. Xbox will see around 1,600 job cuts, as detailed in internal memos from executives. Amy Coleman, EVP and chief people officer, highlighted that while these roles are not being replaced by AI, the evolving technology landscape does necessitate changes in how work is conducted. The layoffs coincide with Microsoft’s launch of its Frontier Company business unit, which aims to enhance enterprise AI deployments backed by a $2.5 billion investment. Asha Sharma, CEO of Xbox, described this as the most significant restructure in Xbox history, citing ongoing challenges within the industry and the need for a reset to ensure long-term success.

Tabular DataFoundation Models+2

Tabular DataFoundation ModelsIndependent EvaluationMachine Learning

Independent Evaluation of Google's TabFM Foundation Model Yields Insights

Google Research's TabFM, a zero-shot foundation model for tabular data, claims to outperform tuned gradient-boosted trees without any training. An independent evaluation was conducted to verify these assertions, utilizing three machines. The results confirmed that the model adheres to the scikit-learn contract, demonstrating bit-exact deterministic predictions. Initial tests showed that TabFM could learn from context effectively, achieving high accuracy on known-answer tasks. This evaluation underscores the need for rigorous independent reproduction in confirming the efficacy of new models.

Hacker NewsRead →

AISupply Chain+2

AISupply ChainAdvanced MaterialsProduct Launch

Honeywell Spin-Off Aims to Transform AI Supply Chain with $14.5 Billion Acquisition of Element

Honeywell's recent $14.5 billion acquisition of Element is set to establish a prominent advanced materials company, boasting a combined enterprise value of approximately $29 billion. This strategic move underscores Honeywell's commitment to enhancing its position in the AI supply chain sector, fostering innovation and growth in advanced materials technology.

Financial Times TechRead →

HealthcareAI+2

HealthcareAIMultimodalBreast Cancer

New Multimodal AI Test Enhances Breast Cancer Recurrence Risk Assessment

A recent study published in Nature Communications introduces a multimodal AI test that utilizes routine breast cancer H&E slides to predict the likelihood of cancer recurrence across various invasive breast cancer subtypes. The research, which analyzed data from 8,161 patients, demonstrates that the AI test provides a more precise method for predicting disease-free intervals, achieving a C-index of 0.71 compared to the standard 21-gene assay's C-index of 0.61. The AI test's performance was particularly notable in triple-negative breast cancer, where existing guidelines lack recommended assays. The findings indicate the potential of AI-driven pathology tests to significantly improve risk stratification and personalized treatment decisions in breast cancer care.

Nature CommunicationsRead →

Generative AINLP+2

Generative AINLPPrivacyAI Training

Google's Updated Privacy Settings Allow AI Training on User Data Unless Opted Out

Google has updated its privacy settings to allow the company to store more user data, including images, files, and audio recordings, to enhance its AI models. Users who upload media to Google Search services are automatically included in this data collection process unless they choose to opt out. The changes, introduced via a customer email in June, enable two new settings: Search Services History and Personalized Recommendations, which allow users to manage how their activity is used to personalize their Google experience. This update affects multiple Google services, including Maps, Shopping, and Translate, and reflects a broader industry trend of collecting user-generated data for AI training. Google has confirmed that user-uploaded media will be utilized to develop and improve its AI services, with users having the option to control their data preferences.

TechCrunch AIRead →

OpenAIGenerative AI+2

OpenAIGenerative AIStartup FundingRegulation

OpenAI CEO Proposes Government Stake to Share AI Wealth with Americans

OpenAI CEO Sam Altman is reportedly in discussions with President Trump to provide the U.S. government with a 5% stake in OpenAI, which could translate to approximately $42.6 billion based on the company's latest valuation of $852 billion. This proposal echoes Altman's previous suggestions for a broader initiative where companies contribute a percentage of their market value to fund annual disbursements to Americans. The rationale behind this initiative includes compensating creators of content used by AI and addressing public concerns over AI's impact on the labor market. While details remain limited, the proposal aims to enhance public trust in AI companies amidst widespread skepticism. Additionally, maintaining favorable relations with the Trump administration could be crucial for OpenAI and other tech firms navigating the current regulatory landscape.

MIT Technology ReviewRead →