Newsletter
Authors from Letta/UC Berkeley presented their work on "Sleep-time Compute" for LLM reasoning, detailing cost reductions and accuracy gains, in an interactive session hosted by AI Tinkerers.
Researchers from Ohio State presented their paper on non-parametric continual learning for large language models during this virtual meetup.
Duke researchers presented HippoMM, a hippocampal-inspired multimodal memory for audiovisual understanding, featuring a Q&A with the presenter, supported by the Human Feedback Foundation.
MIT postdoc Pradyumna Chari presented on a secure architecture for AI agents, the NANDA index and Verified AgentFacts, during a virtual Paper Club session.
The AI Tinkerers Paper Club featured Google Research's Ali Behrouz discussing "Titans," a neural memory module, with a live Q&A, sponsored by ellamind.
Boxuan Li presented TheAgentCompany, an extensible benchmark evaluating LLM agents on real-world tasks, followed by Q&A, supported by the Human Feedback Foundation.
The virtual paper club session featured a researcher-led dive into AI agent benchmarks, offering practical guidance on optimization and avoiding overfitting. Ludus Labs, Human Feedback Foundation, and AI Tinkerers supported the event.
The Paper Club featured Lewis Hammond from Oxford University discussing multi-agent risks, failure modes, and mitigation directions for advanced artificial intelligence.
The Paper Club, with UC Berkeley researchers, explored multi-agent LLM system failures and intervention strategies, supported by CopilotKit.
The community engaged with an MIT PhD student discussing The AI Agent Index, a database of deployed agentic AI systems, with support from Composio.
