Lifeboat Foundation

In 2020, OpenAI introduced GPT-3 and, a year later, DALL.E, a 12 billion parameter model, built on GPT-3. DALL.E was trained to generate images from text descriptions, and the latest release, DALL.E 2, generates even more realistic and accurate images with 4x better resolution. The model takes natural language captions and uses a dataset of text-image pairings to create realistic images. Additionally, it can take an image and create different variations inspired by original images.

DALL.E leverages the ‘diffusion’ process to learn the relationship between images and text descriptions. In diffusion, it starts with a pattern of random dots and tracks it towards an image when it recognises aspects of it. Diffusion models have emerged as a promising generative modelling framework and push the state-of-the-art image and video generation tasks. The guidance technique is leveraged in diffusion to improve sample fidelity for images and photorealism. DALL.E is made up of two major parts: a discrete autoencoder that accurately represents images in compressed latent space and a transformer that learns the correlations between language and this discrete image representation. Evaluators were asked to compare 1,000 image generations from each model, and DALL·E 2 was preferred over DALL·E 1 for its caption matching and photorealism.

DALL-E is currently only a research project, and is not available in OpenAI’s API.

Summary: People with depression who responded to psilocybin therapy showed an increase in brain connectivity for up to three weeks following treatment. The increased brain connectivity was correlated with self-reported improvements in depression symptoms.

Source: Imperial College London.

Psilocybin, the psychedelic compound found in magic mushrooms, helps to “open up” depressed people’s brains, even after use, enabling brain regions to talk more freely to one another.

With this month’s release of the F-150 Lightning, Ford hopes to electrify new and traditional truck buyers alike, and—eventually—to replace its industry-defining gas-powered line.

A team of researchers at Stanford University, working with a colleague at the Chinese Academy of Sciences, has built an AI-based filtration system to remove noise from seismic sensor data in urban areas. In their paper published in the journal Science Advances, the group describes training their application and testing it against real data from a prior seismic event.

In order to provide advance warning when an earthquake is detected, scientists have placed seismometers in earthquake-prone areas, including urban areas where quakes do the most damage and harm or kill the most people. But seismologists have found it troublesome to sort out seismic data related to natural ground movements from data related to city life. They note that human activities in cities, such as vehicles and trains, produce a lot of seismic noise. In this new effort, the researchers developed a deep learning application that determines which seismic data is natural and which is man-made and filters out those that are non-natural.

The researchers call their new application UrbanDenoiser. It was built using a deep-learning application and trained on 80,000 samples of urban seismic noise along with 33,751 samples from recorded natural seismic activity. The team applied their filtering system to seismic data recorded in Long Beach, California, to see how well it worked. They found it improved the level of desired signals compared to background noise by approximately 15 decibels. Satisfied with the results, they used UrbanDenoiser to analyze data from an earthquake that struck a nearby area in 2014. They found the application was able to detect four times the amount of data compared to the sensors without the filtering.

The notion of self-driving vehicles is currently met with equal parts wonder and alarm. But a new study reveals how the pros may outweigh the cons as a business strategy.

An article titled “Impact of Autonomous Vehicle Assisted Last-Mile Delivery in Urban to Rural Settings” determines that this technology reduces the completion time of delivery tours and provides the most cost-effective business model. It appears in Transportation Science.

“The starting point of this paper involved the United States Postal Service announcing its idea to start using autonomous vehicles in rural routes,” said Sara Reed, assistant professor of business analytics at the University of Kansas.

In this one-hour public lecture Josh Frieman, director of the Dark Energy Survey, presents an overview of our current knowledge of the universe and describe new experiments and observatories. Over the last two decades cosmologists have made remarkable discoveries: Only 4 percent of our universe is made of ordinary matter — atoms, molecules, etc. The other 96 percent is dark, in forms unlike anything with which we are familiar. About 25 percent is dark matter, which holds galaxies and larger-scale structures together and may be a new elementary particle. And 70 percent is thought to be dark energy, an even more mysterious entity which speeds up the expansion of the universe. Josh Frieman is senior staff scientist at the Fermilab and Professor of Astronomy and Astrophysics and member of the Kavli Institute for Cosmological Physics at the University of Chicago. The Dark Energy Survey is a collaboration of 300 scientists from 25 institutions on 3 continents, which built and uses a powerful 570-Megapixel camera on a telescope in Chile to carry out a 5-year survey of 300 million galaxies and thousands of supernovae to probe dark energy and the origin of cosmic acceleration.

“This is the most expansive industrial control system attack tool that anyone has ever documented,” says Sergio Caltagirone, the vice president of threat intelligence at industrial-focused cybersecurity firm Dragos, which contributed research to the advisory and published its own report about the malware. Researchers at Mandiant, Palo Alto Networks, Microsoft, and Schneider Electric also contributed to the advisory. “It’s like a Swiss Army knife with a huge number of pieces to it.”

Dragos says the malware has the ability to hijack target devices, disrupt or prevent operators from accessing them, permanently brick them, or even use them as a foothold to give hackers access to other parts of an industrial control system network. He notes that while the toolkit, which Dragos calls “Pipedream,” appears to specifically target Schneider Electric and OMRON PLCs, it does so by exploiting underlying software in those PLCs known as Codesys, which is used far more broadly across hundreds of other types of PLCs. This means that the malware could easily be adapted to work in almost any industrial environment. “This toolset is so big that it’s basically a free-for-all,” Caltagirone says. “There’s enough in here for everyone to worry about.”

The CISA advisory refers to an unnamed “APT actor” that developed the malware toolkit, using the common acronym APT to mean advanced persistent threat, a term for state-sponsored hacker groups. It’s far from clear where the government agencies found the malware, or which country’s hackers created it—though the timing of the advisory follows warnings from the Biden administration about the Russian government making preparatory moves to carry out disruptive cyberattacks in the midst of its invasion of Ukraine.

Existing quantum devices can actually do things that we cannot compute with classical computers. The question is only can we harness this computational power that is apparently there,” van Bijnen says. “Maybe doing arbitrary computational problems is a bit much to ask, so we are now looking at whether we can match problems well to available quantum hardware.” Many current experiments involving Rydberg atoms would likely not require any radical changes in instrumentation that is already being used, he adds.