So far it’s been…
- unhelpful error messages that don’t really explain what’s going on
- lack of crucial access to API fields (afaict you can’t use logit bias in the ChatOpenAI model?)
- bloated architecture that seems more complicated as genAI APIs evolve so fast

I feel like we already have examples of this, except they’re under-publicized, caught early, and not super easy to understand to people that aren’t super tech-literate

This is how I ended up here…just had a handful of friends already here, and I’m sick of “influencers.” I know it’ll happen here too but for now it’s nice

I think there’s probably an art to being “dense” and not “cluttered”—maybe density is good when the information is all of the same type, and not a bunch of separate features? Might be interesting to see cultural differences wrt UX performance

LLMs can't do:
1. Causal inference: LLMs can't reason about cause and effect. They can't build models of the world and identify from first principles what interventions can bring change. They can't data mine datasets to make effective recommendations about what data to collect next.

Thanks