Labs Weekly Catchup #2

Our second weekly catchup. We organize an agenda the day beforehand and consolidate our findings, notable research progress, possibly new tech, insight or technique everyone should know.

Agenda:

• Kick-off topic - Self-hosted AI model in browsers with a quick demo by Thanh Phan and Toan Ho
• Elixir cluster problem: scalability, broadcast messaging, leader election by Hieu Phan
• Compare Rust 101 vs Go with experience in running Rust classes with the team by Hieu
• Rust and understanding monadic error handling by Khac Vy
• AI trending and understanding model and runtime architectures in LLMs by Tom Nguyen

Output:

• Understand the research progress and passions of the labs team
• Understand any shortcomings or things to improve for the next weekly catchup

Notes from the catchup

🧠 Self-hosted AI model in browsers:

• @thanh expressed concerns about the usage and download size of LLMs on weak internet connections.
• @toanho is exploring WebGPU and its application for our use case, although browser support is still limited.
• Action items for next week are being planned.

⚙️ Elixir Cluster Problem:

• @hieuthu1 proposed one possible solution in solving Elixir clustering and scalability through using Process Groups to group and orchestrate actions between processes, enabling message broadcasting across multiple instances/pods.
• Raft consensus (leader voting) is also being considered to help orchestrate state and work.

⚖️ Comparing Rust 101 vs Go:

• @hieuthu1 discussed the differences in concurrency between Rust and Go, highlighting Rust’s focus on functional style and managing references, borrowing, and closures.
• Rust may have a moderate barrier of entry for new learners due to the nuanced understanding required for references and borrowing.
• @hieuthu1 mentioned that a workshop/session on Rust will be organized soon.

🎮 Rust Game Server:

• @khacvy demoed a more interactive and flushed UI for the spaceship Rust game.
• He also mentioned the challenges he encountered, and gave a short course in running Rust

💡 Updates on AI:

• @Tom shared information about Mixtral 8x7B AI, based on Mixture of Experts.
• Mixture of Experts is extensively used in ChatGPT, but advancements in multi-agent LLMs have led to clearer architectures for scaling multi-modal and specialized infrastructure.