MLOps Community

@mlopscommunity.bsky.social

A community of folks looking to put AI and ML into Production

This is the third time in the last 2 months I have heard a guest talk about how our biggest bottleneck working with large models is the memory constraints. All from people I would consider to be preeettyyyy legit Here is a clip taken from my conversation with Krishna Sridhar from Qualcomm

is anyone building mixpanel for AI agent use? Like which agents are being used and effective, how they're doing on their tasks etc. I guess its a form of evaluation but more like evaluating the pmf of the agent and less the output. (out eval also good)

Creating AI awards of 2024. So far the categories I have are: - Most censored model: the original Gemini? - Best aquahire: Adept? Character AI? Inflection? - Best model thats bad at benchmarks: Claude Sonnet? - Best/Worst AI Product: Truth terminal? - Best open source model: llama 3.2? What else?

Agents are to tools what ML Models are to Jupyter Notebooks. Ok. I know. That doesn't make much sense. But stay with me. Tool Calling gives AI power. It allows LLMs to interact with external systems and perform tasks dynamically, which is kinda important for building agents.