Steven Boada, PhD

Spent some time yesterday diving into RAG (Retrieval-Augmented Generation) and found these videos to be extremely helpful in understanding how to move from toy projects to production-grade pipelines. If you're working on RAG applications, these are must-watch: 1️⃣ Jerry Liu @ AI Engineer Summit – Talks about challenges in productionizing RAG, evaluation techniques, and how to make retrieval better. Link: https://lnkd.in/gRK8hi2R 2️⃣ LlamaIndex x Arize – Practical tips on fine-tuning embeddings, chunking, and query routing for more robust RAG systems. Link: https://lnkd.in/gdP7RZvW 3️⃣ Langchain RAG Course – End-to-end hands-on tutorial to build a multi-user chatbot using Langchain, FastAPI & Streamlit. Link: https://lnkd.in/gV_e7KGa Definitely worth a watch if you're building with LLMs in the real world.

52

1 Comment

Analyzing large-scale profiling data is easier thanks to new work from Alexandr Kalinin, Ph.D., Gregory Way (University of Colorado), Shantanu Singh, and others. To improve upon existing methods that make assumptions about distribution, linearity, or sample size of profiling data, they developed a statistical framework and open-source software for information-retrieval-based assessment of profile strength and similarity. Employing mean average precision as an evaluation metric, they demonstrated its utility on simulated and real-world datasets for image, protein, and mRNA profiles. Their versatile framework could help streamline hypothesis generation and improve hit prioritization from a range of high-dimensional biological profiling datasets. Read more in Nature Communications. 🔗: https://lnkd.in/eg7rqZ9D #BroadInstitute #Science #ScienceNews #Research #ScientificResearch

37

I'm frequently doing weird stuff in code that doesn't have many (if any) examples out on the internet. I've noticed that LLM code assist tools will do something annoying in these cases, especially if I'm trying to decode something that uses a custom encoding protocol, and its struggling to come up with a solution. It will try to have me make changes to my code which will "fake" a success. If its a decoder it will create a code that produces proper looking output but its not actually decoding anything, just making up cleartext that looks right. I was working with a fairly obscure and not commonly used quantum simulation library, which hasn't been well maintained and wouldn't run. Two different LLMs gave me similar patches which made the library run, but would produce false, totally made up simulation results. If you're doing something challenging, and the LLM is struggling to figure it out, watch carefully to make sure it doesn't provide code that fakes it. Sometimes its quite obvious that's what they are doing, other times they can be quite sneaky.

68

12 Comments