Starting June 12, 2023, our latest update focuses on recent advancements and announcements in the fields of data science, machine learning, and artificial intelligence. This week, we spotlight prominent companies such as Google, OpenAI, UC Berkeley, and AWS.
Generative AI support in Vertex AI is now generally available
Google has introduced the availability of generative AI support in Vertex AI, enabling customers to take advantage of the latest platform capabilities for creating and operating personalized generative AI applications. This update gives developers access to a variety of tools and resources, including text models powered by PaLM 2, embedding APIs for text, and base models in Model Garden. Additionally, Generative AI Studio provides easy-to-use tools for model fine-tuning and deployment. Backed by enterprise-level data governance, security, and safeguards, Vertex AI simplifies the process for customers to access base models, customize models with their own data, and rapidly develop generative AI applications.
DeepMind Introduces AlphaDev
Google Deepmind has announced AlphaDev, an artificial intelligence system that uses reinforcement learning to discover improved computer science algorithms that surpass those that scientists and engineers have developed over the years. AlphaDev has discovered a more efficient sorting algorithm used to organize the data. These algorithms play a fundamental role in many aspects of our daily lives, from ranking online search results and social media posts to processing data on our computers and smartphones. Harnessing AI to generate better algorithms will revolutionize computer programming and have a profound impact on all aspects of our growing digital society.
OpenAI Announces Function Calls
OpenAI has introduced an API update that includes a feature called Function Calls. This allows a developer to write functions in his GPT-4 and GPT-3.5 and have the model write code to execute those functions. Function calls facilitate the development of chatbots that can leverage external tools, translate natural language into database queries, or extract structured data from text. These models are fine-tuned to not only identify the instances where the function should be called, but also to provide a JSON response that matches the function’s signature.
University of California, Berkeley Develops Gorilla, API-Connected LLM
Researchers at the University of California, Berkeley have released Gorilla, a fine-tuned model based on LLaMA that outperforms GPT-4 in terms of performance when generating API calls. The integration with the documentation retrieval feature allows Gorilla to effectively adapt to documentation changes during testing, allowing for flexible API updates and version changes. Additionally, Gorilla greatly addresses the hallucination problem commonly encountered when directly prompting LLM. APIBench, a comprehensive dataset consisting of HuggingFace, TorchHub, and TensorHub APIs, was introduced to evaluate model capabilities. A good combination of the search system and Gorilla has shown the potential for LLM to leverage tools more accurately, stay up-to-date with frequently changing documents, and improve the reliability and usefulness of the output. increase.
AWS Introduces Falcon, Foundational LLM Built and Trained on Sagemaker
AWS and the UAE Technology Innovation Institute have launched Falcon LLM, a basic large-scale language model with 40 billion parameters. Falcon matches the performance of other high performance LLMs and is the top ranked open source model on the public Hugging Face Open LLM leaderboard. Available as open source in two different sizes, Falcon-40B and Falcon-7B, it was built from the ground up with data preprocessing and model training jobs built on Amazon SageMaker. Open source, his Falcon 40B allows users to build and customize AI tools that cater to unique user needs, facilitating seamless integration and ensuring long-term preservation of data assets. Model weights can be downloaded, inspected and deployed anywhere.
