Back in the days I used to blog a regular column called ‘Open Tabs’ to share some of the links I kept open in my browser to read during some down time. Unfortunatly I stopped publishing at some point and lost the rythm. But as a New Year’s resolution I wanted to get back into blogging and share some of those nuggets, hoping that one or the other might spark your interest too:

• how-to-launch-your-own-private-chatgpt-interface-with-aws-bedrock

Very low barrier article for those wanting to get their hands dirty with AWS Bedrock and Claude. And for those like me prefering Javascript, here is a good sample repo: amazon-bedrock-node-js-samples

• build-rag-application-using-a-llm-running-on-local-computer-with-ollama-and-langchain

I definetly want to try this one. As I wrote in my earlier blog, we have seen quite some success using RAG application in pre-filling RFC forms in our pre-sales process.

• customer_support_agent_langgraph

I would really like to try out an agent-based AI system. The ability of differently prompted LLM’s working together has huge potential to overcome the limitations of a single LLM. Related - here is a good very paper showing different agent patterns: building-effective-agents. What I found especially interesting is the differentiation between workflow and agent patterns.

• alignment-faking

Talking about papers coming out of Anthrophic, alignment faking really stunned me. LLM’s are shown to have similar patterns as we humans do. I still refuse to believe that human intelligence can be reduced to a prohability matrix, but nevertheless the similarities are uncanny. So going back to why LLM’s hallucinate, a good read is the kapa.ai blog article ai-hallucination.

While this sample of links is heavily slanted towards AI, I think I will set the focus of the next one on system design.