Lifeboat Foundation

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.

Researchers at North Carolina State University have now identified a welding technique that can be used to join composite metal foam (CMF) components together without impairing the properties that make CMF desirable. CMFs hold promise for a wide array of applications because the pockets of air they contain make them light, strong and effective at insulating against high temperatures.

CMFs are foams that consist of hollow, metallic spheres—made of materials such as stainless steel or titanium—embedded in a metallic matrix made of steel, titanium, aluminum or other metallic alloys. The resulting material is both lightweight and remarkably strong, with potential applications ranging from aircraft wings to vehicle armor and body armor.

In addition, CMF is better at insulating against high heat than conventional metals and alloys, such as steel. The combination of weight, strength and thermal insulation means that CMF also holds promise for use in storing and transporting nuclear material, hazardous materials, explosives and other heat-sensitive materials.

Despite facing opposition and controversy, Tesla and Elon Musk continue to revolutionize the industry and gain support, with the company’s valuation and Musk’s outspoken personality driving outsized interest.

Questions to inspire discussion.

Continue reading “Tesla’s $1 Trillion Projection in 2024: Analysts Clash” »

Experiments performed by psychologists indicate that many people can tell whether their tea is hot or cold by listening to it being poured into a mug. Pouring sounds can also contain other information about a liquid—a connection that has been thoroughly investigated for natural fluids, ranging from rain drops to ocean waves. Now Ho-Young Kim from Seoul National University, South Korea, and colleagues have studied an artificial system—a fluid jet hitting a liquid bath—revealing that the volume of the sound that is produced depends on the jet’s shape prior to impact [1]. The results indicate that sound could be used as a noninvasive probe of oxygen levels in bodies of water, says Kim.

The researchers sent a few-millimeters-diameter jet of water down a nozzle that hung above a 10-cm-diameter water-filled cylinder. Using an underwater microphone they recorded the amplitude of the sounds produced when the jet hit the surface as a function of the nozzle’s height. Using a camera, they monitored the pattern of bubbles that formed in the cylinder after jet impact and the outline of the jet, which for sufficiently high nozzles broke up into a train of droplets prior to hitting the water surface.

The team found that as the jet developed more ripples and then broke into droplets, it produced louder sounds—a consequence of the trapping of more air bubbles in the water bath. The result implies that, since thin jets break up more easily than thick ones, thin jets should splash more loudly for a given pouring height. Trying to silently serve tea by carefully pouring a thin stream of liquid may thus be unsuccessful, says Kim.

A Chinese research team claims to have developed a new device called ‘Golden Veil’ that can disguise cruise missiles as commercial aircraft.

Let’s kick off the Holiday Season in style, shall we? Like, say, with a single few people even know exist, from a massive pop culture phenomenon of the early ’80s? YES PLEASE! This really needs no introduction. It’s KITT, the Amazing Car of Tomorrow, in full hero mode, Saving Santa from bad weather, and then making the rounds to bring joy and cheer on Christmas Morning — All told in an early ’80s rap, over an electro groove lifted from the Knight Rider theme song. Hopefully you’ve already hit play, but if you haven’t yet, HIT PLAY NOW! And if you have, then PLAY IT AGAIN! And of course, I’d like to wish you all a Verry Merry Christmas! Like it? Subscribe! Follow me at / djmikebrady I don’t own any of the rights associated with this music, I simply share so that it will be heard. #PlayingRecords #NeedleDrop #RecordCollection

In machine learning, larger networks with increasing parameters are being trained. However, training such networks has become prohibitively expensive. Despite the success of this approach, there needs to be a greater understanding of why overparameterized models are necessary. The costs associated with training these models continue to rise exponentially.

A team of researchers from the University of Massachusetts Lowell, Eleuther AI, and Amazon developed a method known as ReLoRA, which uses low-rank updates to train high-rank networks. ReLoRA accomplishes a high-rank update, delivering a performance akin to conventional neural network training.

Scaling laws have been identified, demonstrating a strong power-law dependence between network size and performance across different modalities, supporting overparameterization and resource-intensive neural networks. The Lottery Ticket Hypothesis suggests that overparameterization can be minimized, providing an alternative perspective. Low-rank fine-tuning methods, such as LoRA and Compacter, have been developed to address the limitations of low-rank matrix factorization approaches.

BEIJING/SHANGHAI, Dec 22 (Reuters) — Tesla (TSLA.O) has acquired land in Shanghai for a megapack battery manufacturing plant with production expected to start in the fourth quarter of 2024, Chinese state media reported on Friday.

Tesla paid 222.42 million yuan ($31.13 million) for use rights to a 19.7-hectare (48.7 acres) plot, a separate government statement said on Thursday. The site is near an existing Tesla plant producing Model 3 and Model Y cars.

Tesla did not immediately respond to a request for comment.

By bringing together disparate approaches, machines can reach a new level of creative problem-solving.