Lifeboat Foundation

While autonomous robots have started to move out of the lab and into the real world, they remain fragile. Slight changes in the environment or lighting conditions can easily throw off the AI that controls them, and these models have to be extensively trained on specific hardware configurations before they can carry out useful tasks.

This lies in stark contrast to the latest LLMs, which have proven adept at generalizing their skills to a broad range of tasks, often in unfamiliar contexts. That’s prompted growing interest in seeing whether the underlying technology—an architecture known as a transformer—could lead to breakthroughs in robotics.

In new results, researchers at DeepMind showed that a transformer-based AI called RoboCat can not only learn a wide range of skills, it can also readily switch between different robotic bodies and pick up new skills much faster than normal. Perhaps most significantly, it’s able to accelerate its learning by generating its own training data.

When an animal takes notice of an approaching figure, it needs to determine what it is, and quickly. In nature, competition and survival dictate that it’s better to think fast—that is, for the brain to prioritize processing speed over accuracy. A new study shows that this survival principle may already be wired in the way the brain processes sensory information.

Kumar and fellow KTH neuroscientist Pawel Herman collaborated with KTH information theorists Movitz Lenninger and Mikael Skoglund to study input processing in the brain using information theory and computer models of the brain. Neuroscientist Arvind Kumar, an associate professor at KTH Royal Institute of Technology, says that the study offers a new view of neural coding of different types of inputs in the brain.

The new study surprisingly shows that initial visual processing is “quick but sloppy” in comparison to information processing in other parts of the brain’s vast neural network, where accuracy is prioritized over speed. The paper is published in the journal eLife.

Cofounder of Zeda.io—helping product leaders discover the right problems to solve and build around. And I love traveling, with my dog!

Artificial intelligence (AI) is transforming the world as we know it, and product management is no exception. It has the potential to revolutionize customer research, decision-making and much more, providing us with data-driven insights and paving the way for a future that is not only intelligent but intuitive.

With AI at our fingertips, we’re standing at the threshold of a new era in product management. However, integrating AI into product management also presents challenges that must be addressed. We will delve into how AI influences the world of product management and what it holds for the future.

Andreessen argues that thanks to A.I., “productivity growth throughout the economy will accelerate dramatically, driving economic growth, creation of new industries, creation of new jobs, and wage growth, and resulting in a new era of heightened material prosperity across the planet.”

This week, on the Lex Fridman Podcast, he offered advice to young people who want to stand out in what he describes in this “freeze-frame moment” with A.I.—where tools like ChatGPT and GPT-4 are suddenly available and “everybody is kind of staring at them wondering what to do.”

Continue reading “Marc Andreessen says we’re in a ‘freeze-frame moment’ with A.I.—and has advice for young people” »

Huge libraries of drug compounds may hold potential treatments for a variety of diseases, such as cancer or heart disease. Ideally, scientists would like to experimentally test each of these compounds against all possible targets, but doing that kind of screen is prohibitively time-consuming.

In recent years, researchers have begun using computational methods to screen those libraries in hopes of speeding up drug discovery. However, many of those methods also take a long time, as most of them calculate each target protein’s three-dimensional structure from its amino-acid.

Researchers in Japan and Australia have developed a new multicore optic fiber able to transmit a record-breaking 1.7 petabits per second, while maintaining compatibility with existing fiber infrastructure. The team–from Japan’s National Institute of Information and Communications Technology (NICT) and Sumitomo Electric Industries, and Macquarie University in Sydney, Australia—achieved the feat using a fiber with 19 cores. That’s the largest number of cores packed into a cable with a standard cladding diameter of 0.125 micrometers.

“We believe 19 cores is the highest practical number of cores or spatial channels you can have in standard cladding diameter fiber and still maintain good quality transmission,” says Georg Rademacher, who previously headed the project for NICT but who has recently returned to Germany to take up a directorship in optical communications at the University of Stuttgart.

Most fiber cables for long-distance transmission in use today are single core, single-mode glass fibers (SMF). But SMF is approaching its practical limit as network traffic rapidly increases because of AI, cloud computing, and IoT applications. Many researchers are therefore taking an interest in multicore fiber in conjunction with space-division multiplexing (SDM), a transmission technique for using multiple spatial channels in a cable.

Some of the world’s leading human and robot minds are heading to the United Nations.

At a UN summit in Geneva next week, tech luminaries ranging from futurist Ray Kurzweil to DeepMind COO Lila Ibrahim will discuss AI for good. It’s a stellar lineup of speakers, but the real stars in our eyes are the robots.

Over 50 of the beasts — the majority from Europe — will be in attendance. All of them merit places in your dreams and nightmares, but we’ve narrowed the roster down to a list of our 10 favourites.

I’d place Sigmund et al. as one of my favorite papers that I have read this year! They leverage protein engineering to create genetically encoded nanocages which accumulate metals and appear as concentric circles when imaged by electron microscopy. Six classes of distinct “EMcapsulins” could be differentiated by training a machine learning model (a convolutional neural network) to recognize and classify them within images. Fusion of fluorescent protein domains to the EMcapsulins also allowed correlative imaging between fluorescence microscopy and electron microscopy. The authors demonstrated 3D imaging of EMcapsulins via serial section transmission electron microscopy and focused ion beam… More.

Multiplexable barcodes for electron microscopy are applied to brain imaging.

TOKYO (Reuters) — What would society look like if cyborg body parts were freely available for use like roadside rental bicycles? Masahiko Inami’s team at the University of Tokyo have sought to find out by creating wearable robotic arms.

Inami’s team is developing a series of technologies rooted in the idea of “jizai”, an Japanese term that he says roughly denotes autonomy and the freedom to do as one pleases.

The aim is to foster something like the relationship between musician and instrument, “lying somewhere between a human and a tool, like how a musical instrument can become as if a part of your body.”