We are excited to announce the launch of the CBORG AI Portal, part of a comprehensive suite of AI services created to empower Berkeley Lab researchers and staff with the latest AI technologies to enhance research and productivity. CBORG is especially useful for software code development and troubleshooting.
Developed by the IT Division’s Science IT Group, the CBORG AI Portal offers a unique combination of commercial cloud-based models and open-source models hosted on Berkeley Lab infrastructure, all available at no cost to staff. This approach ensures that users have access to the latest AI advancements while maintaining flexibility and control over data privacy.
Key features of the CBORG AI Portal include:
- Access to ChatGPT 4o, Google Gemini Pro and Anthropic Claude
- Lab-hosted open models for free, on-premises use: Meta Llama 3, Command R+
- Support for web-based interactive chat, vision and vector embedding models
- API service for users developing their own LLM-powered applications
- Documentation and example code to facilitate API integration
Frequently Asked Questions
Q: Who can access the CBORG AI Portal?
A: CBORG is free to use for all Lab employees with an @lbl.gov, @es.net or @nersc.gov email address. Costs associated with use of commercial AI models are paid for by the IT Division.
Q: Why use Lab-hosted versus Commercial Models?
A: Lab-hosted models, such as the powerful, Meta developed open source Llama 3 LLM, offer data retention within Berkeley Lab systems, are free to use, and support open research. However, they may be offline during maintenance periods. Commercial models provide higher quality outputs, faster generation, and better multi-modal capabilities, but at a higher cost and with potential data sharing considerations.
Q: How do CBORG’s Lab-hosted Models compare to Self-Hosted Options?
A: CBORG utilizes advanced hardware such as Nvidia DGX H100 GPUs, large memory capabilities for high-resolution models and a scalable inference infrastructure, offering better performance and accuracy compared to typical self-hosted setups on consumer-grade hardware.
Q: Where does CBORG store data?
A: To support the chat user interface, user’s chat history transcripts are stored on Lab-hosted servers within the IT Division’s data center in Building 50. When interacting with Lab-hosted open models, your data stays entirely within the Berkeley Lab network. When using commercial cloud-hosted models, your data is sent to external services for processing, but will not be saved or used for training purposes.
Q: Can CBORG be used with PII or CUI data?
A: No – please contact Science IT at scienceit@lbl.gov for guidance regarding use of AI models with PII or CUI data.
Q: Can I host my own model on IT Division infrastructure?
A: Yes, we welcome discussions about hosting custom models or LLM-powered applications. Please contact scienceit@lbl.gov to explore this option.
in the bottom right or navigate to