OpenAI expands its custom model training program

OpenAI has introduced new enhancements to its fine-tuning API, offering developers increased control and more options to build custom models tailored to specific needs. The updated API now features epoch-based checkpoint creation, a comparative Playground UI, third-party integration, comprehensive validation metrics, and improved hyperparameter configuration. These advancements aim to boost model performance, reduce latency, and lower costs, exemplified by Indeed's application of the API to enhance job recommendation quality, achieving significant improvements in efficiency and scale.

Expanding its Custom Models Program, OpenAI has unveiled an assisted fine-tuning service, enabling closer collaboration between organizations and OpenAI's technical teams. This service leverages advanced fine-tuning methods and hyperparameters to optimize models for specific domains, enhancing performance and accuracy. SK Telecom's partnership with OpenAI to refine a model for telecommunications is a prime example, demonstrating substantial improvements in customer service efficiency and user satisfaction.

Furthermore, OpenAI now supports the development of fully custom-trained models, allowing organizations to infuse their models with unique domain-specific knowledge. This approach is ideal for entities with extensive proprietary data aiming to achieve highly specialized model behavior. The collaboration with Harvey, an AI-powered legal tool, illustrates the potential of custom-trained models, significantly improving the model's performance in legal contexts. OpenAI envisions widespread adoption of customized AI models across industries, offering tailored solutions to meet diverse organizational needs.

Introducing Command R+: A Scalable LLM Built for Business

Command R+, a new large language model (LLM) optimized for enterprise applications, is now available on Microsoft Azure. Command R+ is designed with a 128k-token context window and excels in advanced Retrieval Augmented Generation (RAG) to minimize hallucinations, multilingual support for ten crucial languages, and sophisticated Tool Use for automating business processes. This model builds on the successes of the Command R model, offering enhanced performance for business-critical functions while maintaining a commitment to data privacy and security.

Command R+ boasts significant advancements in RAG, providing enterprises with reliable, verifiable solutions for a variety of functions across sectors. It includes new features like Multi-Step Tool Use, improving task success rates by enabling the model to iterate over tool usage. Moreover, its multilingual capabilities ensure effective operation across global markets, supporting essential languages to cater to a diverse user base. Cohere has also partnered with Microsoft Azure and plans to expand availability to Oracle Cloud Infrastructure and other platforms, emphasizing its utility in real-world enterprise scenarios.

Pricing for Command R+ is set at $3.00 per million input tokens and $15.00 per million output tokens, reflecting its advanced capabilities and the value it offers to businesses seeking to integrate AI into their operations. Cohere is committed to data privacy and security, providing options for private deployments and data sharing opt-outs. With these advancements, Command R+ is poised to redefine the standards for enterprise-grade AI applications, offering powerful tools for businesses to enhance efficiency and productivity.

OpenAI wants Tesla staff, so Musk is poaching them first

Elon Musk is embroiled in a talent war, with OpenAI allegedly trying to recruit engineers from Tesla's Autopilot team, prompting him to increase salaries to retain his staff. Amidst this, Tesla's Ethan Knight defected to Musk's own startup, xAI, rather than OpenAI, indicating a fierce competition for AI talent. Concerns arise over Musk's management style and the potential conflict of interest as he moves talent between his companies, affecting Tesla's reputation and investor confidence.

Musk, who co-founded OpenAI, is now in a legal tussle with Sam Altman, OpenAI's head, and is actively trying to boost xAI's profile against OpenAI by portraying Tesla as a leading AI entity. This move is seen as an effort to justify his internal talent shifts and to maintain Tesla's edge in the AI space, despite criticisms over his overlapping interests across his ventures.

Investors and Tesla's small shareholders are increasingly alarmed by the ongoing situation, questioning the board's oversight and Musk's dual roles across his companies. The battle for AI talent is intensifying, with Musk emphasizing the significance of keeping top engineers to stay competitive in the rapidly evolving AI landscape, even as he navigates the complexities of managing multiple innovative companies.

Google is considering charging for AI-powered search in a big change to the business model

Google is reportedly contemplating introducing a paywall for premium AI-generated content, adding to its subscription services which include the AI assistant Gemini. This marks a potential shift in Google's traditionally free access model, though the basic search engine would remain free but ad-supported. The idea is still under consideration, with no firm decision made on implementing the paywall.

The tech giant faced some backlash with its AI assistant Gemini, which generated controversial images, prompting a temporary halt of the tool. Despite these hiccups, Google remains a dominant force in the internet search arena, with a significant share of the desktop search engine market and a broad user base relying on its various services.

Google's revenue is primarily derived from advertising, and while the company is exploring new revenue streams through subscription-based services, it has denied plans for an ad-free search experience. Instead, Google aims to enhance its subscription offerings, maintaining its position as a key player in the digital world with a vast portfolio of products and services.

Other stuff

• Fake AI law firms are sending fake DMCA threats to generate fake SEO gains

• Waymo is expanding its partnership with Uber, to make electric, fully autonomous Uber Eats delivery 'a reality'. (Does this mean no Tip 👀)

• Meta’s AI image generator can’t imagine an Asian man with a white woman

• Tool use is now available in beta to all customers in the Anthropic Messages API.

How to use Prompt Chains in ChatGPT

Hope you enjoyed today's newsletter

⚡️ Join over 200,000 people using the Superpower ChatGPT extension on Chrome and Firefox.