We present LaMDA: Language Models for Dialog Applications. LaMDA is a family of Transformerbased neural language models specialized for dialog, which have up to 137B parameters and are pre-trained on 1.56T words of public dialog data and web text. While model scaling alone can improve quality, it shows less improvements on safety and factual grounding. We demonstrate that fine-tuning with annotated data and enabling the model to consult external knowledge sources can lead to significant improvements towards the two key challenges of safety and factual grounding. The first challenge, safety, involves ensuring that the model's responses are consistent with a set of human values, such as preventing harmful suggestions and unfair bias. We quantify safety using a metric based on an illustrative set of human values, and we find that filtering candidate responses using a LaMDA classifier fine-tuned with a small amount of crowdworker-annotated data offers a promising approach to improving model safety. The second challenge, factual grounding, involves enabling the model to consult external knowledge sources, such as an information retrieval system, a language translator, and a calculator. We quantify factuality using a groundedness metric, and we find that our approach enables the model to generate responses grounded in known sources, rather than responses that merely sound plausible. Finally, we explore the use of LaMDA in the domains of education and content recommendations, and analyze their helpfulness and role consistency. * Work done while at Google.

Oxytocin neurons represent one of the major subsets of neurons in the paraventricular hypothalamus (PVH), a critical brain region for energy homeostasis. Despite substantial evidence supporting a role of oxytocin in body weight regulation, it remains controversial whether oxytocin neurons directly regulate body weight homeostasis, feeding or energy expenditure. Pharmacologic doses of oxytocin suppress feeding through a proposed melanocortin responsive projection from the PVH to the hindbrain. In contrast, deficiency in oxytocin or its receptor leads to reduced energy expenditure without feeding abnormalities. To test the physiological function of oxytocin neurons, we specifically ablated oxytocin neurons in adult mice. Our results show that oxytocin neuron ablation in adult animals has no effect on body weight, food intake or energy expenditure on a regular diet. Interestingly, male mice lacking oxytocin neurons are more sensitive to high fat diet-induced obesity due solely to reduced energy expenditure. In addition, despite a normal food intake, these mice exhibit a blunted food intake response to leptin administration. Thus, our study suggests that oxytocin neurons are required to resist the obesity associated with a high fat diet; but their role in feeding is permissive and can be compensated for by redundant pathways.

SUMMARY The melanocortin receptor 4 (MC4R) is a well-established mediator of body weight homeostasis. However, the neurotransmitter(s) that mediate MC4R function remain largely unknown and as a result, little is known about the second-order neurons of the MC4R neural pathway. Single minded 1 (Sim1)-expressing brain regions, which include the paraventricular nucleus of hypothalamus (PVH), represent key brain sites that mediate melanocortin action. We conditionally restored MC4R expression in Sim1 neurons in the background of Mc4r-null mice. The restoration dramatically reduced obesity in Mc4r-null mice. The anti-obesity effect was completely reversed by selective disruption of glutamate release from those same Sim1 neurons. The reversal was caused by lower energy expenditure and hyperphagia. Corroboratively, disruption of glutamate release selectively from adult PVH neurons led to rapid obesity development via reduced energy expenditure and hyperphagia. Thus, this study establishes glutamate as the primary neurotransmitter that mediates MC4Rs on Sim1 neurons in body weight regulation.

Summary Cre-loxP technology enables specific examination of the function and development of individual nuclei in the complex brain network. However, for most brain regions, the utilization of this technique has been hindered by the lack of mouse lines with Cre expression restricted to these regions. Here, we identified brain expressions of three transgenic Cre lines previously thought to be pancreas-specific. Cre expression driven by the rat-insulin promoter (Rip-Cre) was found mainly in the arcuate nucleus, and to a lesser degree in other hypothalamic regions. Cre expression driven by the neurogenin 3 promoter (Ngn3-Cre mice) was found in the ventromedial hypothalamus. Cre expression driven by the pancreas-duodenum homeobox 1 promoter (Pdx1-Cre) was found in several hypothalamic nuclei, the dorsal raphe and inferior olivary nuclei. Interestingly, Pdx1-Cre mediated deletion of vesicular GABA transporter led to postnatal growth retardation while Ngn3-Cre mediated deletion had no effects, suggesting a role for Pdx1-Cre neurons, but not pancreas, in the regulation of postnatal growth. These results demonstrate the potential for these Cre lines to study the function and development of brain neurons.