- Recent technological advancements have made it possible to run AI models on devices, known as edge AI.
- Engineers from Carnegie Mellon University, University of Washington, Shanghai Jiao Tong University, and OctoML have collaborated to run large language models on smartphones, PCs, and browsers.
- If this research can be implemented in everyday use, it could greatly expand the applications of generative AI.
- Currently, most AI models run in the cloud, requiring users to rent servers from providers like Azure, AWS, and Google.
- The ability to run AI models on devices could potentially reduce costs and increase accessibility for users.
The main topic is SoftBank's launch of SB Intuitions, a new company focused on developing Large Language Models (LLMs) specialized for the Japanese language and selling generative AI services based on Japanese. The key points are:
1. SB Intuitions will be 100% owned by SoftBank and will use data housed in Japan-based data centers.
2. SoftBank plans to tap into its extensive consumer and enterprise operations in Japan to support SB Intuitions.
3. The company will utilize a computing platform built on NVIDIA GPUs for developing generative AI and other applications.
4. Hironobu Tamba, a long-time SoftBank employee, will lead the new business.
5. SoftBank has not disclosed the total investment in SB Intuitions but recently issued a bond for AI investments.
6. SoftBank has had a mixed track record with AI, both in its in-house services and as an AI investor.
7. SoftBank aims to address the lack of domestically-produced generative AI and its importance in Japanese business practice and culture.
8. SoftBank has a strategic alliance with Microsoft and will provide a secure data environment for enterprises interested in AI initiatives.
9. SoftBank plans to establish a multi-generative AI system by selecting the most appropriate model from companies like OpenAI, Microsoft, and Google.
Main topic: DynamoFL raises $15.1 million in funding to expand its software offerings for developing private and compliant large language models (LLMs) in enterprises.
Key points:
1. DynamoFL offers software to bring LLMs to enterprises and fine-tune them on sensitive data.
2. The funding will be used to expand DynamoFL's product offerings and grow its team of privacy researchers.
3. DynamoFL's solutions focus on addressing data security vulnerabilities in AI models and helping enterprises meet regulatory requirements for LLM data security.
Hint on Elon Musk: There is no mention of Elon Musk in the given text.
The struggle between open-source and proprietary artificial intelligence (AI) systems is intensifying as large language models (LLMs) become a battleground for tech giants like Microsoft and Google, who are defending their proprietary technology against open-source alternatives like ChatGPT from OpenAI; while open-source AI advocates believe it will democratize access to AI tools, analysts express concern that commoditization of LLMs could erode the competitive advantage of proprietary models and impact the return on investment for companies like Microsoft.
Cloud computing vendor ServiceNow is taking a unique approach to AI by developing generative AI models tailored to address specific enterprise problems, focusing on selling productivity rather than language models directly. They have introduced case summarization and text-to-code capabilities powered by their generative AI models, while also partnering with Nvidia and Accenture to help enterprises develop their own generative AI capabilities. ServiceNow's strategy addresses concerns about data governance and aims to provide customized solutions for customers. However, cost remains a challenge for enterprises considering the adoption of generative AI models.
Enterprises need to find a way to leverage the power of generative AI without risking the security, privacy, and governance of their sensitive data, and one solution is to bring the large language models (LLMs) to their data within their existing security perimeter, allowing for customization and interaction while maintaining control over their proprietary information.
Microsoft and Datadog are well positioned to benefit from the fast-growing demand for generative artificial intelligence (AI) software, with Microsoft's exclusive partnership with OpenAI and access to the GPT models on Azure and Datadog's leadership in observability software verticals and recent innovations in generative AI.
IBM has introduced new generative AI models and capabilities on its Watsonx data science platform, including the Granite series models, which are large language models capable of summarizing, analyzing, and generating text, and Tuning Studio, a tool that allows users to tailor generative AI models to their data. IBM is also launching new generative AI capabilities in Watsonx.data and embarking on the technical preview for Watsonx.governance, aiming to support clients through the entire AI lifecycle and scale AI in a secure and trustworthy way.
Generative AI's "poison pill" of derivatives poses a cloud of uncertainty over legal issues like IP ownership and copyright, as the lack of precedents and regulations for data derivatives become more prevalent with open source large language models (LLMs). This creates risks for enterprise technology leaders who must navigate the scope of claims and potential harms caused by LLMs.
Generative AI's intersection with Web3 has sparked debates about decentralization, with the philosophical case supporting decentralizing AI due to its control by dominant providers and lack of transparency, raising the need for a decentralized network model that allows collaboration, knowledge sharing, and democratic access to models and benefits. The lack of success in decentralized AI until now was largely due to questionable value propositions and different architectural paradigms. Decentralization in generative AI can be considered across various dimensions, including compute, data, optimization, evaluation, and model execution, with each phase of the lifecycle presenting different opportunities for decentralization. Although achieving decentralized AI will require technical breakthroughs, it is the right approach in the era of foundation models to ensure knowledge is not concentrated in the hands of a centralized entity.
