It's a little off-topic for this thread maybe, but since you asked, and since I've been thinking about it...
I know a lot of people put their temperature sensors in gel packs, water, or other buffers, but I actually think this is unnecessary and causes the temperature of the liquid to vary more than if you leave it in the air.
My logic is as follows:
There is a lag time between a temperature change in the air and the corresponding temp change in the liquid, since it takes time for heat to transfer into or out of the liquid. If you raise the temp of the air in the chamber, say 10 degrees, the temp of the fluid will gradually start to rise until it's sitting at the same temp. Because of this, the air temp can fluctuate relatively rapidly up and down while the temp of the liquid changes only very slightly. There are some threads on this board where people have measured air temps varying in the 5-10 degree range, and the liquid in the carboy varying less than a degree.
Let's use a cooling-only case as an example. When the compressor is off, the air temp in the chamber is gradually rising. If you leave the temp probe exposed to the air, the compressor kicks in pretty much immediately as soon as the air temp hits the set point and works to bring the air temp back down to below the threshold. As such, the max temp of the air in the chamber using this system is very close to the "compressor on" set point.
Now consider the case where you buffer the temp probe. The compressor is off. Remember the air temp in the chamber is gradually rising. It hits the "compressor on" set point, but because the temp probe is buffered, it keeps rising until the heat can work its way through the buffering solution and eventually to the probe itself. By the time the compressor kicks on, the temp in the chamber may be much higher than the set point. The temp of the liquid in the chamber (both the carboy and your temp probe buffer) has actually moved enough to trigger the compressor.
The only reason I could see to buffer a thermostat temp probe is if you want to slow down the response time of the system. It's bad for compressors to cycle too quickly, but most temp controllers allow you to explicitly set a minimum cycle time (2-3 minutes is fine, but you can do more if you want) to avoid this. This is actually more reliable, since it guarantees that the compressor won't cycle more frequently, where the buffered temp probe system still could if the temp differentials were high enough.
The other reason I could see to put a thermometer in a gel pack is to act as a proxy for the actual liquid temp if you don't have a thermowell for your carboy. But note that this is not to act as a temp sensor to control the system, it would just be to give a better estimate of liquid temp than you'd get by measuring the air.
Sorry that was long. Hope it's helpful.