Lifeboat Foundation

Inspired by progress in large-scale language modelling, we apply a similar approach towards building a single generalist agent beyond the realm of text outputs. The agent, which we refer to as Gato, works as a multi-modal, multi-task, multi-embodiment generalist policy. The same network with the same weights can play Atari, caption images, chat, stack blocks with a real robot arm and much more, deciding based on its context whether to output text, joint torques, button presses, or other tokens.

During the training phase of Gato, data from different tasks and modalities are serialised into a flat sequence of tokens, batched, and processed by a transformer neural network similar to a large language model. The loss is masked so that Gato only predicts action and text targets.

When deploying Gato, a prompt, such as a demonstration, is tokenised, forming the initial sequence. Next, the environment yields the first observation, which is also tokenised and appended to the sequence. Gato samples the action vector autoregressively, one token at a time.

Recent text-to-image generation methods provide a simple yet exciting conversion capability between text and image domains. While these methods have incrementally improved the generated image fidelity and text relevancy, several pivotal gaps remain unanswered, limiting applicability and quality. We propose a novel text-to-image method that addresses these gaps by (i) enabling a simple control mechanism complementary to text in the form of a scene, (ii) introducing elements t… See more.

Recent text-to-image generation methods provide a simple yet exciting.

Conversion capability between text and image domains. While these methods have.

Continue reading “Make-A-Scene: Scene-Based Text-to-Image Generation with Human Priors” »

Existing pre-trained models are generally geared towards a particular class of problems. To date, there seems to be still no consensus on what the right architecture and pre-training setup should be. This paper presents a unified framework for pre-training models that are universally effective across datasets and setups. We begin by disentangling architectural archetypes with pre-training objectives — two concepts that are commonly conflated. Next, we present a generalized and unified perspective for self-supervision in NLP and show how different pre-training objectives can be cast as one another and how interpolating between different objectives can be effective. We then propose Mixture-of-Denoisers (MoD), a pre-training objective that combines diverse pre-training paradigms together.

Scientists have grown plants in soil from the moon collected by NASA’s Apollo astronauts.

A receptor that was first identified as necessary for insulin action, that also is located on the neural stem cells found deep in the brains of mice, is pivotal for brain stem cell longevity, according to a Rutgers study, a finding that has important implications for brain health and future therapies for brain disorders.

The study 0, appearing in the journal Stem Cell Reports, pinpoints a specific protein known as the insulin receptor (INSR), which is abundant on the neural stem cells that reside in the brain’s subventricular zone. During development, neural stem cells give rise to the entire nervous system, and they persist into adulthood. Over the lifespan these neural stem cells produce new neurons and non-neuronal cells that maintain the infrastructure and functioning of the brain.

Separately, the scientists made another finding when examining brain tumors: INSR plays a crucial role in sustaining and maintaining a population of specialized brain cancer cells known as glioblastoma (GBM) stem cells. When they inactivated the INSR in the GBM stem cells they inhibited the growth of those primitive tumor forming cells.

NASA and Lockheed Martin’s X-59 is nearing its first flight.

A new video from Lockheed Martin provides a new update on the X-59 aircraft it is developing in collaboration with NASA — and the development seems to be ticking along nicely.

Common supplements, such as St. John’s wort, calcium, and iron, can reduce the efficacy of certain drugs, like antibiotics and antiviral medication.

Geophysicists used remote sensing to see reservoirs beneath the surface. That water could speed up the loss of ice as the climate warms.

Geophysicists used remote sensing to see reservoirs beneath the surface. That water could speed up the loss of ice as the climate warms.

According to a new concept by LMU chemists led by Thomas Carell, it was a novel molecular species composed out of RNA and peptides that set in motion the evolution of life into more complex forms.

Investigating the question as to how life could emerge long ago on the early Earth is one of the most fascinating challenges for science. Which conditions must have prevailed for the basic building blocks of more complex life to form? One of the main answers is based upon the so-called RNA world idea, which molecular biology pioneer Walter Gilbert formulated in 1986. The hypothesis holds that nucleotides—the basic building blocks of the nucleic acids A, C, G, and U—emerged out of the primordial soup, and that short RNA molecules then formed out of the nucleotides. These so-called oligonucleotides were already capable of encoding small amounts of genetic information.

As such single-stranded RNA molecules could also combine into double strands, however, this gave rise to the theoretical possibility that the molecules could replicate themselves—i.e. reproduce. Only two nucleotides fit together in each case, meaning that one strand is the exact counterpart of another and thus forms the template for another strand.