Ontology-driven observability
Running a large-scale system like Netflix requires diagnosing issues across disparate systems in real time. LLMs can help, but only if they’re all on the same page. To do this, the systems providing them data must include a semantic layer that ties everything together. Enter ontology.
LLMs aren’t good confidence scorers
Little by little, we’re figuring out what LLMs are good at and not good at. Stating confidence scores on event likelihoods is one of the latter, and this post explains why. When architecting AI-powered processes, use LLMs for what they do well. Giving confidence scores isn’t one of them.
Relationslipping
Individual AI adoption in an organization can erode team performance. How? By undermining human relationships in the team. The key is using AI to augment — not replace — human interactions. (HBR gift link)
Self-obsolescence
From-the-field notes from OpenAI forward-deployed engineers, who face an interesting question: what do you do when the pace of AI development makes your solutions obsolete? A: you embrace it: “If we make ourselves obsolete every couple of months, our FDE team is doing its job.”
Unified Context Layer
An explanation of how Zoom architects context, fusing long-term information about organizations with more dynamic information coming from their chats and meeting transcripts. Not a lot of detail in this product announcement, but worth sharing because all orgs will eventually operate something like this.