A common practice in knowledge-grounded dialogue generation is to explicitly utilize intermediate steps (e.g., web-search, memory retrieval) with modular approaches. However, data for such steps are often inaccessible compared to those of dialogue responses as they are unobservable in an ordinary dialogue. To fill in the absence of these data, we develop a self-improving method to improve the generative performances of intermediate steps without the ground truth data. In particular, we propose a novel bootstrapping scheme with a guided prompt and a modified loss function to enhance the diversity of appropriate self-generated responses. Through experiments on various benchmark datasets, we empirically demonstrate that our method successfully leverages a self-improving mechanism in generating intermediate and final responses and improves the performances on the task of knowledge-grounded dialogue generation. Along with the progress of Language Model (LM) pretraining, open-domain dialogue models have evolved to leverage the advantage of the transformer architecture's generalization ability (Zhang et al., 2019; Freitas et al., 2020; Roller et al., 2021; Xu et al., 2022a; Shuster et al., 2022b; Thoppilan et al., 2022). While model scaling also improves the dialogue quality (Freitas et al., 2020) as seen in large LMs, relying on sole LMs casts limitations such as hallucination and the lack of faithfulness by outdated training data (Brown et al., 2020; Thoppilan et al., 2022; Chowdhery et al., 2022). In order to overcome the limitations, prior works have adopted a modular design where multiple modules generate intermediate texts (e.g., to retrieve documents) before the final response (Lewis et al., 2020; Adolphs et al., 2021; Zhang et al., 2021; Shuster et al., 2022a). Among them, Komeili et al. (2022); Shuster et al. (2022b) have shown promising results in dialogue generation. Specifically, they adopted a modular design to integrate external knowledge (e.g., internet) and internal knowledge (e.g., memory) in dialogue models. For example, in Komeili et al. (2022), a LM first decides whether to access a knowledge in a form of text generation. Upon deciding to access knowledge, the LM generates an appropriate query for knowledge retrieval from external sources such as search engines. Then, the LM generates a response based on extracted knowledge from the accessed data. See Figure 2 of Appendix A for an illustrative example. Regarding each intermediate phase as a separate module, a convenient method of training these modules would be to apply supervised learning on each module using individual datasets (Dinan et al., 2019; Shuster et al., 2022a; Glass et al., 2022; Shuster et al., 2022b).

Vincross, the company behind the six-legged robot Hexa, said on Tuesday that it's picked up $10 million in a Series A funding round led by Lenovo Capital, the startup fund managed by Lenovo Group. Returning investor GGV Capital and newcomer Seekdource Capital also participated. The company declined to disclose its latest valuation but said the proceeds will go towards research and development as well as new product lines. Neuroscience and artificial intelligence researcher Tianqi Sun started Vincross in Beijing back in 2016 when he raised $220,000 for Hexa on Kickstarter. At the time the insectile, programmable robot had separated itself from the horde of humanoids on the market by billing itself as the first robot that can climb stairs, making it suitable for firefighting and other rescue tasks.

If Matt Chasen gets his way, there will be a time -- in the not-so-distant future -- when commuters are able to order an air taxi that whisks them across town in minutes, bypassing traffic-clogged streets below. For now, however, the chief executive of LIFT Aircraft will have to use his start-up's electric-powered vertical-takeoff-and-landing aircraft, the Hexa, for something else: 15-minute flights across a lake outside Austin, for $249 a pop. Though the flights will target a recreational crowd, Chasen sees them as a steppingstone to a new form of convenient urban transportation. "Today's regulatory environment does not allow for a transportation use of these aircraft -- yet," said Chasen, a former Boeing engineer with a background in mechanical and aerospace engineering. "We'll build public trust in the technology. Once that happens, it's inevitable that people will want to use it for certain types of commuting flights."

While the likes of Uber, Airbus and Porsche tinker away on their respective passenger and transportation drones, a lesser-known startup is taking an altogether different approach. Instead of getting mired in the logistics and regulatory frameworks of city-wide drone rides, Lift Aircraft wants you to use its 18-rotor "Hexa" aircraft for short recreational flights. The large drone -- which weighs 432 pounds and is capable of 10-15 minutes of continuous flight with a single passenger -- could be available to the public as early as next year. Lift is promising flight experiences at hubs located in "scenic, uncongested areas" in 25 cities across the US. Because the Hexa doesn't count as a "real" aircraft (it's a "powered ultralight") it doesn't require a pilot's license.

A startup has developed a robot that might make it so you'll never accidentally kill a plant again. Called Hexa, the intelligent planter is just big enough to hold a medium-sized succulent and is equipped with spider-like legs that can walk into the sunlight when its plant passenger needs some vitamin D. Beijing-based robotics company Vincross created Hexa with the idea that humans might become more acquainted with robots in their home. Hexa can walk toward the light when the plant is in need of some sun and rotate or spin to get the right amount of sunlight. The six-legged robot is equipped with several motors, infrared and 720p cameras, as well as a Linux operating system. Hexa uses its legs to walk toward the light when the plant is in need of some sun and rotate or spin to get the right amount of sunlight.

The uncanny valley teems with creepy humanoids--machines not quite perfect enough to be mistaken for people, but not quite comically robotic enough to be endearing. Lately, they've been joined by robo-animals, like the mechanical dog from Boston Dynamics that not-at-all-unsettlingly regains its balance if you kick it. Now, a new robot is scuttling into the uncanny valley. Hexa has six legs, looks like a bug, and moves with bizarre confidence. And it just might bring robot hacking to the masses.

The other day, I was in one of my student's homes. Noticing the lack of dust--really, the house is always clean--I joked to her, "Is it you or your parents vacuuming these floors so well?" She told me that they actually have a central vacuum system installed in the house. As someone who lives in a rather modest townhouse, she might as well have been speaking Latvian. Thus, as I usually do, I began the rabbit hole internet search about central vacuum systems.