Search for: All records

The strength of modern generative models lies in their ability to be controlled through text-based prompts. Typical "hard" prompts are made from interpretable words and tokens, and must be hand-crafted by humans. There are also "soft" prompts, which consist of continuous feature vectors. These can be discovered using powerful optimization methods, but they cannot be easily interpreted, re-used across models, or plugged into a text-based interface. We describe an approach to robustly optimize hard text prompts through efficient gradient-based optimization. Our approach automatically generates hard text-based prompts for both text-to-image and text-to-text applications. In the text-to-image setting, the method creates hard prompts for diffusion models, allowing API users to easily generate, discover, and mix and match image concepts without prior knowledge on how to prompt the model. In the text-to-text setting, we show that hard prompts can be automatically discovered that are effective in tuning LMs for classification. 

The objective of this paper is to demonstrate the flexure properties of ABS plastic in a 3D printed object as a process to enable embedded pressure sensing capabilities. Developing the potential for non-static 3D parts broadens the scope of the fused deposition modeling (FDM) process to include printing ‘smart’ objects that utilize intrinsic material properties to act as microphones, load sensors, accelerometers, etc. In order to demonstrate a strain-based pressure transducer, strain gauges were embedded either directly on top or in the middle of a flexible ABS diaphragm. Securing a strain gage directly on top of the diaphragm traced a reference pressure more closely than diaphragms with the strain gage embedded halfway into the diaphragm. To prevent temperature-related drift, an additional strain gage was suspended above the secured gage, inside the 3D printed cavity. The additional gage allowed for a half-bridge circuit in lieu of a quarter-bridge circuit, which minimized drift due to temperature change. The ABS diaphragm showed no significant signs of elastic hysteresis or nonlinear buckling. When sealed with 100% acetone, the diaphragm leaked ∼50x slower than as-printed sensors. After pressurizing and depressurizing the devices multiple times, they output pressure readouts that were consistent and repeatable for any given pressure within the operational range of 0 to 7psi. The repeatability of each of the final generation sensors indicates that ‘smart’ objects printed using an FDM process could be individually calibrated to make repeatable recordings. This work demonstrates a concept overlooked previous to now — FDM printed objects are not limited to static models, which lack dynamic motion of the part as an element of design. Altering FDM’s bottom-up process can allow for easily embedding sensing elements that result in printed objects which are functional on the mesoscale.