Lifeboat Foundation

Generative multi modal models are context dependent learners.

Shared with Dropbox.

Gravity is the reason things with mass or energy are attracted to each other. It is why apples fall toward the ground and planets orbit stars.

Magnets attract some types of metals, but they can also push other magnets away. So how come you feel only the pull of gravity?

Continue reading “Einstein’s Insight: Why Does Gravity Pull Us Down and Not Up?” »

Immigration to and living on Mars have often been themes in science fiction. Before these dreams can become reality, humanity faces significant challenges, such as the scarcity of vital resources like oxygen needed for long-term survival on the Red Planet. Yet, recent discoveries of water activity on Mars have sparked new hope for overcoming these obstacles.

Scientists are now exploring the possibility of decomposing water to produce oxygen through electrochemical water oxidation driven by solar power with the help of oxygen evolution reaction (OER) catalysts. The challenge is to find a way to synthesize these catalysts in situ using materials on Mars, instead of transporting them from the Earth, which is of high cost.

Artificial intelligence (AI) is poised to revolutionise the way we diagnose and treat illness. It could be particularly helpful for depression because it could make more accurate diagnoses and determine which treatments are more likely to work.

Some 20% of us will have depression at least once in our lifetimes. Around the world, 300 million people are currently experiencing depression, with 1.5 million Australians likely to be depressed at any one time.

Because of this, depression has been described by the World Health Organization as the single biggest contributor to ill health around the world.

China’s cyber-space regulator has unveiled draft measures that would make companies responsible for the data used to train generative artificial intelligence (AI) models, such as Midjourney and ChatGPT.

The tool — dubbed ‘AI-Descartes’ by the researchers — aims to speed up scientific discovery by leveraging symbolic regression, which finds equations to fit data.

Given basic operators, such as addition, multiplication, and division, the systems can generate hundreds to millions of candidate equations, searching for the ones that most accurately describe the relationships in the data.

Using this technique, the AI tool has been able to re-discover, by itself, fundamental equations, including Kepler’s third law of planetary motion; Einstein’s relativistic time-dilation law, and Langmuir’s equation of gas adsorption.

A team led by researchers at Osaka University and University of California, San Diego has conducted simulations of creating matter solely from collisions of light particles. Their method circumvents what would otherwise be the intensity limitations of modern lasers and can be readily implemented by using presently available technology. This work might help experimentally test long-standing theories such as the Standard Model of particle physics, and possibly the need to revise them.

One of the most striking predictions of quantum physics is that matter can be generated solely from light (i.e., photons), and in fact, the astronomical bodies known as pulsars achieve this feat. Directly generating matter in this manner has not been achieved in a laboratory, but it would enable further testing of the theories of basic quantum physics and the fundamental composition of the universe.

In a study published in Physical Review Letters, a team led by researchers at Osaka University has simulated conditions that enable photon–photon collisions, solely by using lasers. The simplicity of the setup and ease of implementation at presently available laser intensities make it a promising candidate for near-future experimental implementation.

Any activity that requires comprehension and production in one or more modalities is considered a multimodal task; these activities can be extremely varied and lengthy. It is challenging to scale previous multimodal systems because they rely heavily on gathering a large supervised training set and developing task-specific architecture, which must be repeated for every new task. In contrast, present multimodal models have not mastered people’s ability to learn new tasks in context, meaning that they can do so with minimal demonstrations or instructions. Generative pretrained language models have recently shown impressive skills in learning from context.

New research by researchers from Beijing Academy of Artificial Intelligence, Tsinghua University, and Peking University introduces Emu2, a 37-billion-parameter model, trained and evaluated on several multimodal tasks. Their findings show that when scaled up, a multimodal generative pretrained model can learn similarly in context and generalize well to new multimodal tasks. The objective of the predict-the-next-multimodal-element (textual tokens or visual embeddings) is the only one used during Emu2’s training. This unified generative pretraining technique trains models by utilizing large-scale multimodal sequences, such as text, image-text pairs, and interleaved image-text video.

The Emu2 model is generative and multimodal; it learns in a multimodal setting to predict the next element. Visual Encoder, Multimodal Modeling, and Visual Decoder are the three main parts of Emu2’s design. To prepare for autoregressive multimodal modeling, the Visual Encoder tokenizes all input images into continuous embeddings, subsequently interleaved with text tokens. The Visual Decoder turns the regressed visual embeddings into a movie or image.

Quanta Magazine’s coverage of biology in 2023, including important research progress into the nature of consciousness, the origins of our microbiomes and the timekeeping mechanisms that govern our lives and development.\
\
Read about more breakthroughs from 2023 at Quanta Magazine: https://www.quantamagazine.org/the-bi…\
\
00:05 The Investigation of Consciousness\
Our minds are constantly taking in new external information while also creating their own internal imagery and narratives. How do we distinguish reality from fantasy? This year, researchers discovered that the brain has a “reality threshold” against which it constantly evaluates processed signals. \
- Original story with links to research papers can be found here: https://www.quantamagazine.org/is-it–…\
\
04:30 Microbiomes Evolve With Us\
This year, scientists provided clear evidence that the organisms in our microbiome —the collection of bacteria and other cells that live in our guts and elsewhere on our body — spread between people, especially those with whom we spend the most time. This raises the intriguing possibility that some illnesses that aren’t usually considered communicable might be.\
– Original story with links to research papers can be found here: https://www.quantamagazine.org/global…\
\
08:43 How Life Keeps Time\
The rate at which an embryo develops and the timing of when its tissues mature vary dramatically between species. What controls the ticking of this developmental clock that determines an animal’s final form? This year, a series of careful experiments suggest that mitochondria may very well serve dual roles as both the timekeeper and power source for complex cells.\
- Original story with links to research papers can be found here: https://www.quantamagazine.org/what-m…\
\
- VISIT our Website: https://www.quantamagazine.org \
- LIKE us on Facebook: / quantanews \
- FOLLOW us Twitter: / quantamagazine \
\
Quanta Magazine is an editorially independent publication supported by the Simons Foundation: https://www.simonsfoundation.org/