very true. infra engineers will have generational run for years and beyond categories.
- model serving / inference infra
- gpu / distributed systems infra
- cloud + kubernetes + orchestration
- data infra
- eval + observability infra
- agent / runtime infra
i still think as ai advances there will be demand for areas which govern bottlenecks underneath like latency, throughput / cost / reliability / data quality and security.
Tbh the only AI policy we need is completely open research from all labs. Open sourcing everything would actually convince the public of any imminent danger, as opposed to vague headlines and fear-mongering
@aadilbuilds Seems like a hybrid approach could work, wherein we have a hierarchical structure, where a few orchestrator agents coordinate (through any consensus algos) and delegate tasks to sub-agents
@Westoncb Tbh starting with some form of hand-written code + basic template + asking the agent occasionally + changelog should be enough to keep track of our codebase
@vxnuaj Though i don't believe LLMs are conscious, tbh this a weak argument, cause the same can be said of your neural synapses. It's not the numbers, but the *act* of calculation and what emerges from it is the question.
I always thought it would be interesting to have a small-ish vocabulary for random but consistent llm behaviours, depending on models and their versions. Maybe like a behaviour guide or something
I've been calling this "thought-terminating flattery", and it's quite annoying:
Me: <adds random suggestion>
Claude: "that's a better frame than anything I've said." <proceeds to just explain what I suggested and stop>
@aadilbuilds Fair. My mind immediately went to either graph-CRDTs or some sort of self-reconciler graph structure (though both seem off somehow).
I did find some repos which have a Paxos/Raft like structure to get coordination down, but it seems a sub-agent setup is the best for now
@joodalooped@shapegrifter Yeah, the higher level shape of feelings is hard to grasp, not impossible, and maybe even easy in certain cases, but overall hard