Neural nets may remain at runtime when tackling judgments that can only be made dynamically at runtime. Every other LLM agent belongs at buildtime accelerating the production of executable software.
I like (and apply) this as a design principle. I do have to note that much of my operations RAM goes to embedding models nowadays; I simply have tons of stuff that I want to have in vector databases. This may be the nature of doing more consulting work nowadays, and when someone throws me a 2000 page PDF I am not going to read all that.
I like (and apply) this as a design principle. I do have to note that much of my operations RAM goes to embedding models nowadays; I simply have tons of stuff that I want to have in vector databases. This may be the nature of doing more consulting work nowadays, and when someone throws me a 2000 page PDF
I am not going to read all that.