July 2023 Summaries

Cohere For AI is a research lab and open science community dedicated to addressing complex machine learning challenges and advancing fundamental research in the field. Launched in June 2022, the lab is particularly focused on nurturing emerging talent through initiatives like the Cohere For AI Scholars Program, which offers aspiring NLP researchers the opportunity to work in a supportive, collaborative environment alongside leading experts from January to August 2024. This program, which does not require prior research degrees or experience, aims to broaden participation in machine learning research by providing access to large-scale experimental settings, world-class expertise, and a commitment to responsible open-source scientific practices. The initiative is part of Cohere's broader efforts to contribute to the ongoing development of AI technologies while emphasizing diversity and inclusivity in research opportunities.

The AI Tinkerers Summer Hackathon, organized by AI Tinkerers and Madrona Ventures, saw the triumph of a team comprising Reza Salehi, Ramsey Khadder, and Pratik Prakash, who developed Sherlock AI, a Discord bot designed to streamline information retrieval from company documentation and support threads. Leveraging Cohere's Rerank technology, which enhances search precision with minimal coding, the team created a functional demo within the hackathon timeframe, addressing the common challenge of navigating incomplete documentation and support channels for technical queries. The hackathon offered exclusive access to the Rerank private beta, which significantly expedited the development process for the winning team, emphasizing its utility for both large enterprises with extensive internal frameworks and smaller companies reliant on open-source projects. As the first-place winners, the team received prizes including cloud credits, Cohere credits, and a VIP dinner invitation, while all participants enjoyed community swag.

Cohere has unveiled a new programmatic fine-tuning feature for its Python SDK, allowing developers to integrate fine-tuning into their workflows more seamlessly than before. Previously limited to the Cohere Dashboard, this enhancement offers automation, version control, and reproducibility, streamlining the process of customizing pre-trained language models for specific tasks. By utilizing the SDK, developers can automate fine-tuning, manage code with version control systems like Git, and ensure reproducibility by defining, training, and fine-tuning models through code. The process involves using the `create_custom_model` method with essential and additional parameters, enabling efficient model customization. This development empowers users to harness Cohere's advanced language models more effectively, opening up possibilities for tailored applications across various industries.

Cohere's collaboration with Amazon Web Services (AWS) enables its advanced enterprise AI technology to be integrated into Amazon Bedrock, a fully managed service that simplifies the use of pre-trained foundation models through an API. This partnership will include Cohere's Command model, which is designed for business-oriented text generation tasks such as summarization and question answering, and the Embed model for text understanding tasks like search and classification across over 100 languages. This integration aims to provide AWS customers with greater flexibility and access to powerful AI models for diverse business applications, further extending Cohere's reach from its initial offering on Amazon SageMaker JumpStart. The move aligns with Cohere's strategy to offer enterprises cloud options and meet them where they store their data, while Amazon Bedrock continues to democratize access to generative AI applications with a variety of industry-leading models.

The text discusses the advantages and challenges of implementing intelligent chatbots for business use, emphasizing their potential to enhance customer support by providing quick and accurate product information, which can lead to faster case resolutions and improved customer sentiment. However, concerns about data privacy and trust in generative AI pose significant barriers to widespread adoption, as companies fear sensitive data might be leaked. Cohere is developing an enterprise-grade chatbot called Coral, designed to be conversational, customized, grounded with verifiable data, and secure within a company's data perimeter. This chatbot aims to support knowledge workers by integrating with various data sources and adapting to specific business needs. The text includes endorsements from industry leaders who highlight the potential of Coral to improve decision-making and customer experiences by leveraging businesses' own data securely. Coral is currently in private access with selected customers, and Cohere invites interested parties to participate in the program.

Cohere and McKinsey have partnered to help organizations integrate generative AI into their operations, aiming to redefine business processes, upskill workforces, and address enterprise challenges to accelerate business value. This collaboration allows McKinsey to offer enterprise AI capabilities that are cloud agnostic, providing flexibility for clients to choose their preferred cloud provider or opt for on-premises solutions, thereby ensuring secure and private data environments. This is especially beneficial for industries handling sensitive data like financial services, healthcare, and retail. The partnership seeks to enhance internal document search and discovery, improving employee productivity and efficiency in business analysis and knowledge work. Additionally, Cohere's technology will support automation by connecting models to third-party apps, expanding potential use cases. McKinsey's strategic and technological expertise positions it to implement Cohere's AI effectively, leveraging the potential of large language models in a time when AI is increasingly sought after in various sectors.

Collaborative research plays a pivotal role in advancing Generative AI, and there is a strong commitment to supporting the broader research community in developing safer and more capable technology. The initiative is particularly interested in requests for API access that facilitate beneficial applications and responsible use of large language models (LLMs). The program seeks to support diverse research questions and gain insights from academic partners' findings. Interested applicants are encouraged to fill out an application detailing their research objectives, the models they wish to study, and other relevant project information. For any inquiries, participants are invited to contact support via email.

The text discusses the use of Cohere's Rerank endpoint to improve the relevance of search results from the Wikipedia API by utilizing reranking techniques that go beyond basic relevance signals and leverage semantic similarity. This method enhances the overall search experience by delivering more accurate and meaningful results to users, as demonstrated through examples like searches for "Where are Monet’s water lilies?" and "David Bowie hits." The text highlights the initial steps of retrieving search results from Wikipedia, reranking them using Cohere’s model, and displaying the improved results. The integration of Cohere’s reranking capabilities into user interfaces helps developers provide more valuable search outcomes, ultimately increasing user satisfaction and uncovering valuable information that might otherwise be overlooked.