But at least that use case fully acknowledges it as a glorified statistical probability machine. Anything that gets even close to considering it for any sort of decision-making or information retrieval should be seen as immediately ridiculous at best imo, unethical when it can cause harm at worst.
Check out @ewanmurray.bsky.social#cognition#psychscisky#neuroskyence#cogscisci#edusky
I have been doing my best to retrieve my 15yo Facebook account that has been disabled and has high security alert on it for days. Every thing that looks like a legitimate fb page to deal with account retrieval has been another hack. I called out the woman on the current call and she hung up on me.
the big thing I've felt was missing from OpenAI's beginner RAG was metadata support, doesn't seem to have been added among the retrieval upgrades. If I upload a bunch of docs, I can't tag them with the URL or local path or other metadata to use when responding.
- 📊 LlamaIndex: Connecting custom data to large language models - 🧠 DSPy: Optimizing language model prompts and weights - 📈 RAGAS: Evaluating Retrieval Augmented Generation (RAG) systems - 🔄 LiteLLM: Unified interface for multiple LLM providers
Siqi Li, Danni Liu, Jan Niehues Optimizing Rare Word Accuracy in Direct Speech Translation with a Retrieval-and-Demonstration Approach https://arxiv.org/abs/2409.09009
It's enjoyable for what it is and it has some of that sparkling Howard Hawks-ian dialogue. I appreciate these mugs absolutely accidentally blowing up an alien spaceship on a retrieval mission.
Data retrieval - a week, perhaps (evenings). Presentation - unknown, can be little or much. Mainly depends on factors we haven't discussed yet, like platform, design, etc. Establish routine (making fellas use "the signal"). Unknown, can take off in a second, it can also fail and burn miserably.
Shuting Wang, Xin Yu, Mang Wang, Weipeng Chen, Yutao Zhu, Zhicheng Dou RichRAG: Crafting Rich Responses for Multi-faceted Queries in Retrieval-Augmented Generation https://arxiv.org/abs/2406.12566