Lifeboat Foundation

Backpropagation is the key algorithm that makes training deep models computationally tractable. For modern neural networks, it can make training with gradient descent as much as ten million times faster, relative to a naive implementation. That’s the difference between a model taking a week to train and taking 200,000 years.

Beyond its use in deep learning, backpropagation is a powerful computational tool in many other areas, ranging from weather forecasting to analyzing numerical stability – it just goes by different names. In fact, the algorithm has been reinvented at least dozens of times in different fields (see Griewank (2010)). The general, application independent, name is “reverse-mode differentiation.”

Fundamentally, it’s a technique for calculating derivatives quickly. And it’s an essential trick to have in your bag, not only in deep learning, but in a wide variety of numerical computing situations.

The cosmos may have existed forever, according to a revolutionary model that extends Einstein’s theory of general relativity using quantum correction terms. By taking into consideration dark matter and energy, the model can concurrently address a number of concerns.

Just after filming this video, Sam Altman, CEO of OpenAI published a blog post about the governance of superintelligence in which he, along with Greg Brockman and Ilya Sutskever, outline their thinking about how the world should prepare for a world with superintelligences. And just before filming Geoffrey Hinton quite his job at Google so that he could express more openly his concerns about the imminent arrival of an artificial general intelligence, an AGI that could soon get beyond our control if it became superintelligent. So, the basic idea is moving from sci-fi speculation into being a plausible scenario, but how powerful will they be and which of the concerns about superAI are reasonably founded? In this video I explore the ideas around superintelligence with Nick Bostrom’s 2014 book, Superintelligence, as one of our guides and Geoffrey Hinton’s interviews as another, to try to unpick which aspects are plausible and which are more like speculative sci-fi. I explore what are the dangers, such as Eliezer Yudkowsky’s notion of a rapid ‘foom’ take over of humanity, and also look briefly at the control problem and the alignment problem. At the end of the video I then make a suggestion for how we could maybe delay the arrival of superintelligence by withholding the ability of the algorithms to self-improve themselves, withholding what you could call, meta level agency.

▬▬ Chapters ▬▬

Continue reading “Dangers of superintelligence | Separating sci-fi from plausible speculation” »

This article introduces new approaches to develop early fault-tolerant quantum computing (early-FTQC) such as improving efficiency of quantum computation on encoded data, new circuit efficiency techniques for quantum algorithms, and combining error-mitigation techniques with fault-tolerant quantum computation.

Yuuki Tokunaga NTT Computer and Data Science Laboratories.

Noisy intermediate-scale quantum (NISQ) computers, which do not execute quantum error correction, do not require overhead for encoding. However, because errors inevitably accumulate, there is a limit to computation size. Fault-tolerant quantum computers (FTQCs) carry out computation on encoded qubits, so they have overhead for the encoding and require quantum computers of at least a certain size. The gap between NISQ computers and FTQCs due to the amount of overhead is shown in Fig. 1. Is this gap unavoidable? Decades ago, many researchers would consider the answer to be in the negative. However, our team has recently demonstrated a new, unprecedented method to overcome this gap. Motivation to overcome this gap has also led to a research trend that started at around the same time worldwide. These efforts, collectively called early fault-tolerant quantum computing “early-FTQC”, have become a worldwide research movement.

Winter in the northern hemisphere is always a brutal reminder for the shipping industry that routing vessels efficiently is a big challenge. Winter storms bring low visibility conditions, freezing spray, and sea ice, all of which can lead to catastrophic results if not appropriately navigated, including lost cargo, damaged hulls and even potentially toppling a ship in the most extreme weather. But this January adds additional pressures to the sector with new and enacted regulations around greenhouse emissions and carbon usages. The beneficial news is that in both scenarios, weather intelligence can help those navigating the open seas better plan and safely and efficiently navigate these waters.

While most of us know that weather impacts nearly every aspect of shipping, we likely think of it in terms of safety of people and cargo. According to The Swedish Club 2020 loss prevention report, heavy weather is cited in half of all claims and contributes to 80% of the financial losses. Weather optimized routing uses real-time weather forecasts, oceanic data, and the vessel’s current position to keep captains at sea and voyage managers on land about changing conditions. If there is hazardous weather, most voyage routing algorithms can make numerous calculations in real time and provide one or more alternatives for a ship operator to optimize a route. While ultimately this may not be the most efficient route, it will likely be the safest route for current conditions.

Weather intelligence is also critical in evaluating, and potentially adjusting, greenhouse gas emissions based on vessel performance and fuel usage. The Carbon Intensity Indicator (CII) introduced in 2023 is a rating framework that evaluates how efficiently a ship transports goods or passengers from a carbon emissions standpoint. This is the first year that ships will be assigned a rating. The data from the previous year is used in an efficiency conversion ratio. Each ship is assigned an individual CII rating from A to E, with A being the best possible rank.

The equations that describe physical systems often assume that measurable features of the system—temperature or chemical potential, for example—can be known exactly. But the real world is messier than that, and uncertainty is unavoidable. Temperatures fluctuate, instruments malfunction, the environment interferes, and systems evolve over time.

LimX Dyamics claims CL-1 is one of the few humanoid robots around the world that achieves dynamic stair climbing based on real-time terrain perception.

The Chinese company claims that CL-1 stands out as one of the few models capable of dynamic stair climbing through real-time terrain perception among the global array of humanoid robots. This is achieved through sophisticated motion control and AI algorithms developed by LimX Dynamics, complemented by their proprietary high-performance actuators and hardware systems.

Continue reading “LimX Dynamics’ first humanoid robot gains real-time terrain perception” »

Year 2014 Basically once the master qubit is found it could even lead to a sorta master algorithm. Also it could show who actually pulls the strings of reality.

Whatever the u-bit is, it rotates quickly (Image: Natalie Nicklin)

Our best theory of nature has imaginary numbers at its heart. Making quantum physics more real conjures up a monstrous entity pulling the universe’s strings

Continue reading “From i to u: Searching for the quantum master bit” »

A new, potentially revolutionary artificial intelligence framework called “Blackout Diffusion” generates images from a completely empty picture, meaning that the machine-learning algorithm, unlike other generative diffusion models, does not require initiating a “random seed” to get started. Blackout Diffusion, presented at the recent International Conference on Machine Learning (“Blackout Diffusion: Generative Diffusion Models in Discrete-State Spaces”), generates samples that are comparable to the current diffusion models such as DALL-E or Midjourney, but require fewer computational resources than these models.

“Generative modeling is bringing in the next industrial revolution with its capability to assist many tasks, such as generation of software code, legal documents and even art,” said Javier Santos, an AI researcher at Los Alamos National Laboratory and co-author of Blackout Diffusion. “Generative modeling could be leveraged for making scientific discoveries, and our team’s work laid down the foundation and practical algorithms for applying generative diffusion modeling to scientific problems that are not continuous in nature.”

A new generative AI model can create images from a blank frame. (Image: Los Alamos National Laboratory)