Introduction
Researchers from the University of Hong Kong Xlang lab, Salesforce Research C lab, University of Washington, and MIT CSL have recently unveiled two extraordinary AI models: Lemur and Lemur Chat. These models have the ability to harmonize natural language and code, making them ideal for advanced language agents. In this blog, we will explore the capabilities of these models and discuss BYU's latest AI model, Ernie 4.0.
Lemur and Lemur Chat: Bridging the Gap Between Natural Language and Code
Language agents are software programs that communicate using natural language to interact with humans or other agents. However, using language agents for more advanced tasks that involve code becomes challenging. Current language agents lack the coding expertise required for tasks such as searching for code snippets based on a description or generating code from a description. This is where Lemur and Lemur Chat come in.
Lemur and Lemur Chat are state-of-the-art models developed by a team of researchers. These models are based on Llama 270b, a 70 billion parameter llm that was trained on 2 trillion tokens of text data. To ensure the models have enough exposure to both natural language and code, the researchers used a code-centric corpus called the Stacks, which contains 90 billion tokens of text and code data with a 101 ratio.
Lemur is a general-purpose model that can handle various tasks involving text and code. Lemur Chat, on the other hand, is a specialized model optimized for dialogue use cases. The researchers fine-tuned Lemur using 100K instances from both text and code data and used reinforcement learning with human feedback to align the model with human preferences for helpfulness and safety.
According to the research paper, Lemur and Lemur Chat outperformed all other open-source models on eight text and code benchmarks. These benchmarks covered different scenarios such as code search, code summarization, code translation, text to code generation, and documentation translation. In comparison to popular closed-source models like Chat GPT and Palm, Lemur Chat significantly narrowed the gap on 13 agent benchmarks involving human communication tool usage and interaction under different environments.
These advanced models have the potential to assist with tasks related to natural language and code. They can help develop smarter language tools for everyday use and allow exploration of how natural language and code work together.
BYU's Ernie 4.0: On Par with GPT 4
BYU, the company behind China's largest internet search engine, has introduced its latest generative AI model, Ernie 4.0. This model, which has been in development since 2019, learns from different knowledge sources to better understand the world. According to BYU's CEO Robin Lee, Ernie 4.0 has greatly improved in understanding generation, reasoning, and memory, making it comparable to OpenAI's GPT 4.
During the BYO World 2023 event, Lee demonstrated Ernie 4.0's capabilities through live examples. The model quickly provided a detailed plan for a family trip to Japan during cherry blossom season, including flight details and tips for enjoying the blossoms. It also showcased its artistic abilities by creating an impressive artwork of a dragon flying over mountains based on a text prompt and a picture. Ernie 4.0 not only solves geometry problems but also explains its answers. Additionally, it seamlessly incorporates new details into its narrative when challenged to write a martial arts story.
While some experts remain skeptical, noting that Ernie 4.0 doesn't seem significantly different from its predecessor, Ernie 3.0, BYU's CT O Wang Haun mentioned that the model's performance has improved by almost 30% during beta testing. Ernie 4.0 already has over 45 million users and has received positive feedback.
BYU plans to incorporate generative AI across its offerings, including BYO search, BYO drive, and BYU Maps. This will result in more personalized user experiences, tailored answers, and improved file organization using natural language commands. BYU is also working on AI-powered tools such as Buu Wenu Smart Writer for content creation and Buu Infoflow Smart Video Maker for easy video production.
Challenges and Regulations for Generative AI in China
BYU's generative AI strategy, especially with Ernie 4.0, aims to compete with OpenAI and other big AI companies. However, as generative AI offers benefits and opportunities for innovation and creativity, it also poses challenges and risks for security and regulation.
China has recently proposed new rules and guidelines for managing generative AI services in the country. According to the interim measures for the management of generative artificial intelligence services issued by China's internet regulator, generative AI service providers must register their services with the authorities before launching them to the public. They are also required to conduct a security assessment and ensure compliance with Chinese laws and regulations.
China has a blacklist of banned training sources for AI models, which includes sources with illegal or harmful information. The goal is to prevent AI from generating harmful content and ensure a healthy AI sector in China. Additionally, China has guidelines like the Beijing AI principles, emphasizing respect for human rights, fairness, transparency, and more in AI development.
For Ernie 4.0 and BYU, this means following these guidelines and ensuring their AI services are ethical and beneficial. It also provides an opportunity for them to showcase their AI skills globally and offer value across different areas.
Conclusion
The unveiling of Lemur and Lemur Chat showcases the potential of AI models that can harmonize natural language and code. These models have the ability to handle advanced tasks and assist in the development of smarter language tools. BYU's Ernie 4.0, on the other hand, introduces a generative AI model that competes with OpenAI's GPT 4. The capabilities of Ernie 4.0 open doors for new innovations and personalized user experiences.
While these advanced AI models offer great opportunities, they also come with challenges and the need for regulations. China's proposed rules and guidelines aim to ensure the ethical and beneficial use of generative AI services. BYU's generative AI strategy aligns with these guidelines and focuses on leveraging the power of AI to bring innovative solutions to users.
What are your thoughts on BYU's generative AI strategy, especially Ernie 4.0? Do you think they can compete with GPT 4 and OpenAI? Share your thoughts in the comments!
0 Comments