MLR-Copilot: Autonomous Machine Learning Research based on Large Language Models Agents https://
github.com/du-nlp-lab/MLR
-Copilot
… https://
arxiv.org/pdf/2408.14033
AGENTS
-
MLR-Copilot: Autonomous Machine Learning Research with LLM Agents
By
–
-
Single Agent Systems Easier to Deploy Production
By
–
I too find single agent based systems much easier to reason about, tune, and deploy. As context windows get longer, seems like a better way to get things into production.
-
LLM-Weighted Function Selection for Advanced AI Systems
By
–
Log every function call, store success/error, rate result with LLM, use that to add weights to functions, use weights when calling functions. Advanced methods might include embedding user input and somehow calculate a weight specific to the embedding (is this the right tool for
-
User Interaction Drives AI Self-Improvement and Data Generation
By
–
Thanks! Yeah I think building in user interaction can generate helpful data to continue self-building better and better.
-
Function Creation Complexity and Multi-Agent Consensus Approaches
By
–
I tried to keep this high-level but the “Create Function” step to get past level 1 is really complicated and there are tons of approaches/tricks here. Curious about your multi agent consensus.
-
ReACT Paper Defined LLM Agents Era with LangChain
By
–
Our guest cohost @hwchase17 on How the ReACT paper defined the dawn of the LLM agents era
— Latent.Space (@latentspacepod) 26 septembre 2024
(and inspired @LangChainAI)
next on the pod with @ShunyuYao12! pic.twitter.com/MbHSKj6DrXOur guest cohost @hwchase17 on How the ReACT paper defined the dawn of the LLM agents era (and inspired @langchain
) next on the pod with @ShunyuYao12
! -
Auto-installing imports without user consent raises safety concerns
By
–
That would be smart, but my current system goes as far as auto installing imports without user consent…. no user feedback is baked in
-
Building Reusable Functions for AI Agents
By
–
Good call. I didn’t include it here but my “Create function” step breaks the task down into multiple reusable functions, and then a wrapper function that uses those. So I think with good reasoning (plus external search when helpful), the agent should be building tools you don’t
-
Level 3 Self-Build Agent Prototype with Query Processing
By
–
Now it's far working well (as you can see from the errors…), but here's the output of a prototype self-build agent (lvl 3): – User description as input – Generate queries – Process query – Generates and stores multiple reusable functions for that query excited to share
-
AI-Powered Function Generation: Request-Based Tool Creation
By
–
Level 1: Request-based Users can request to have functions generated by an AI, and when they query the system, the new function is available as an option for the AI to use. *There's lots that goes into this "Create Function Tool", but let's keep the chart simple (eg.