Equilibrium climate sensitivity (ECS) is a purely theoretical but useful metric of climate change, defined as the planet's long-term response to the abrupt doubling of atmospheric carbon dioxide.  The ECS range arising from CMIP models has remained stubbornly large despite many advances in climate modeling and computing power. 

 It has been suggested in recent years that satellite and in situ observations of the recent past may help to narrow this range.  However, the usefulness of this exercise depends on whether the present-day climate is predictive of the far future climate.  

In this talk, I'll discuss the reasons the present may be dissimilar to the future in the CMIP5 models.  I'll present work in which we argue that, at least in a perfect model framework, observations are highly likely to underestimate feedbacks at equilibrium.  

Additionally, the particular manifestation of internal variability and forced response that characterizes recent observations resulted in an even lower estimate.   Moreover, we find that not only do climate models indicate that current estimates of ECS are biased low, but models' sensitivities to observed conditions are not even predictive of their long-term sensitivities.  

Our results suggest a need for caution in any attempt to use current conditions to constrain long-term climate projections.