Release Notes of Generative AI Platform

14 Aug 2024

The GenAI Platform now supports GPT-4o mini and Llama 3.1 70B, the latest AI models that can significantly enhance productivity and efficiency across various domains with the following benefits:
 

GPT-4o mini

  • High Performance & Cost-Efficiency: Despite its smaller size, GPT-4o mini delivers impressive performance across various benchmarks. It is also designed to be highly affordable than previous models.
  • Expanded Context Window: With a context window of up to 128K tokens, it can handle extensive conversations and complex tasks without losing context.
  • Multimodal Capabilities: GPT-4o mini supports text, audio, and image inputs, making it versatile for various applications.
  • Improved Latency and Speed: The model offers faster response times, making it suitable for real-time applications like customer support chatbots.


Llama 3.1 70B

  • Multilingual Capabilities: Supports multiple languages, including English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai.
  • Long Context Handling: Can process up to 128K tokens, making it ideal for tasks requiring long-form text understanding.
  • Content Creation: Excellent for generating high-quality text, making it useful for writers, marketers, and content creators.
  • Language Understanding: Great for tasks like text summarization, classification, and sentiment analysis2.
  • Nuanced Reasoning: Capable of understanding and generating complex and nuanced responses, making it suitable for research and development.
  • Code Generation: Useful for developers as it can assist in generating and understanding code.

Notes: Llama 3.1 70B is an open-source LLM from Meta, using HKUST's limited internal computing resources for inference.  The computational capacity will be improved gradually.  It is free for staff and students with fair use.

 

24 Jun 2024


Change of URL

  • The Platform is using a new URL, https://genai.ust.hk , while visits to https://chatgpt.ust.hk will be redirected to the new location automatically.
     

New and Upgraded Models

  • GPT-4o - A cutting-edge model that understands text and images, excels in complex tasks. With a knowledge cutoff of Oct 2023, it’s more accurate, safer, widely accessible and economical.
  • GPT-4 Turbo - A significant increase in multi-modal capability with an updated knowledge cutoff of Dec 2023 and a large context window that can handle the equivalent of 300 pages of text in a single prompt. It is more cost-effective than the previous gpt-4 model.
  • GPT-3.5 Turbo (version 0125) (16k) - An economical and enhanced model with performance improvement and a larger input window than the previous gpt-3.5 model. Its knowledge is up to Sep 2021.
  • Gemini-1.5 Flash - developed by Google, is a high-speed, efficient AI model designed for high volume tasks. It boasts multimodal reasoning capabilities, advanced problem-solving, code manipulation, and enhanced creativity. Knowledge cutoff in Nov 2023.
  • Gemini-1.5 Pro - a significant upgrade from its predecessor developed by Google and has advanced multimodal capabilities, being able to understand and analyze text, images, video, audio, and codes with a large context window. Knowledge cutoff in Nov 2023.
  • Gemini-1.0 Pro - An efficient model developed by Google, which is designed for a wide range of text-based tasks with training date up to Nov 2023
     

Support of Vision Models

  • Use image inputs for chatting with GPT-4o and GPT-4 Turbo
     

API

  • Supports new and upgraded Azure OpenAI models with multi-modal capabilities (except GPT-3.5 Turbo) and function calling