Gordon Beacham, MBA-IT, BSc. Computer Science’s Post

Cool technique to enhance LLM reasoning ability: Simple cause LLM read the question twice https://lnkd.in/ecU-Z8Vv

Long-Term Memory for LLM - https://lnkd.in/ecBZ8YUm

A huge step toward LLM interpretability https://lnkd.in/g3Vv3ZTQ

A big step forward in the challenge of LLM explainability https://lnkd.in/eAPp9kUN

Long-Term Memory for LLM - https://lnkd.in/ecBZ8YUm

We don't LLM (not yet anyway!) But this is a useful guide to all things LLM and the arc they're on. https://lnkd.in/eePFGaDS

Here is the guide to get started with LLM journey!! https://lnkd.in/gVZq98kv

Interesting steps towards explainability of LLM output https://lnkd.in/gbfkitDK

7 LLM Parameters to Enhance Model Performance (With Practical Implementation) https://lnkd.in/d8HEZWqR

LLM is designed to generate next word(s) with probability rather than giving deterministic answers. The probabilistic nature of LLM is its biggest feature (e.g. creativity) and also its biggest "bug" at the same time (e.g. hallucination). Most of enterprise use cases prefers deterministic answers with high accuracy and explainability, especially when stakes are high. There will always be such tension between the probabilistic nature of LLM and the deterministic requirement of enterprise use cases. We can try to tame the LLM's behavior with RAG etc., but there will always be "wrong" answer due to the probabilistic nature of LLM, and people also perceive the correctness of an answer with subjectivity. To increase the adoption of LLM, it is crucial to strategically select appropriate use cases for LLM deployment, setting appropriate expectation to harness their capabilities while being able to tolerate/accept their limitations.