@chriswithans You can avoid the penalty because it’s a penalty for not doing something. You cannot avoid the 100k tax on the H1B it’s mandatory. Different things are different.
@robinebers It actually saves them compute in the long run. When you stop a task mid stream, when the user resumes the task most likely the entire context is no longer in the cache and has to be recomputed from scratch. So it's better to let it finish on your weekly limit than do a hard 5hr.
@COUnfiltered@9NEWS The biggest issue is the CO GOP is known for putting forward candidates who say lots of crazy things... They get those questions because that's the type of candidate we get on endless repeat.
@Alicia_Smith19 California is plagued by "local action" democracy where they let 1,000s of micro communities organize their own systems and plans. Pure top down planning and orchestration works better for this type of stuff-- but the Dems are averse to that.
@yacineMTB I think the underlying model is fine. I think the insane amount of attention they place on "alignment & safety" is crippling the models in post-training.
@dlwiest The 5 hour limits, mainly. Even if you don't hit your weeklies the fact that you can blow your 5 hour in 30 minutes and now you have no AI for the rest of the workday.
A lot of these LLM interfaces would be better if you could have a cheap local model that you could ask questions on the local context, without disturbing the broader context. I find myself often having the equivalent of /btw with claude code, but not wanting to have the token spend.
@xwanyex That should be a free market decision, not a mandated decision. There are a lot of obvious beneficial things that when COMBINED cause big negative externalities; like increasing the cost of housing everywhere by 300k more because no one can make tradeoffs on the whole set.