Lifeboat Foundation

With recent developments in language modeling (LM) research, machine-generated text applications have spread to a number of previously untapped domains. However, a significant issue remains that LM-generated text frequently contains factual errors or inconsistencies. This problem usually arises in any LM generation scenario, but it is particularly problematic when generation is performed in uncommon domains or when it requires up-to-date information that the LM was not trained on.

Retrieval-Augmented Language Modeling (RALM) methods, which display the LM pertinent documents from a grounded corpus during generation, offer a possible solution to this problem. Current RALM strategies concentrate on changing the LM architecture to include external data. However, this approach often makes deployment significantly complex. Working on this problem statement, AI21 Labs, an organization that develops artificial intelligence systems, introduced an alternative strategy called In-Context Retrieval-Augmented Language Modeling (In-Context RALM), which can supplement an existing language model with ready-made external information sources. The necessary files are added as input into the language model, which keeps the underlying LM architecture unaffected. The team published their findings in a research paper titled “In-Context Retrieval-Augmented Language Models.”

In the same publication, AI21 Labs also unveiled Wordtune Spices, an addition to their Wordtune text editor. Wordtune Spices is an artificial intelligence robot that helps authors swiftly generate text and create content, thereby accelerating the pace of the composition of academic papers, theses, and creative documents. Spices’ main principle is based on the In-context RALM technique. Users of Spices have access to 12 prompt alternatives, including explications, definitions, and even jokes. Users can select the prompt that best supports their use case and receive a string of supplemental sentences to bolster their case and provide further details.

Could the rise of advanced AI tools mean the end of some creative and knowledge jobs? Here’s what the experts had to say.

Anders Sandberg is “not technically a philosopher,” he tells IEEE Spectrum, although it is his job to think deeply about technological utopias and dystopias, the future of AI, and the possible consequences of human enhancement via genetic tweaks or implanted devices. In fact, he has a PhD in computational neuroscience. So who better to consult regarding the ethics of neurotech and brain enhancement?

Sandberg works as a senior research fellow at Oxford’s Future of Humanity Institute (which is helmed by Nick Bostrom, a leading AI scholar and author of the book Superintelligence that explores the AI threat). In a wide-ranging phone interview with Spectrum, Sandberg discussed today’s state-of-the-art neurotech, whether it will ever see widespread adoption, and how it could reshape society.

The question is no longer whether AI will change the workplace; it’s how companies can successfully use it in ways that enable – not replace – the human workforce. AI will help to make humans faster, more efficient, and more productive.

It’s true that AI will threaten some unskilled jobs through automation, but it will also potentially create different kinds of jobs that require new skill sets that will be developed through training.

AI can be used in manufacturing to make processes more efficient while also keeping human workers out of harm’s way. Opportunities to leverage AI and machine learning in manufacturing include product development, logistics optimization, predictive maintenance, and robotics.

In recent years, many computer scientists have been exploring the notion of metaverse, an online space in which users can access different virtual environments and immersive experiences, using VR and AR headsets. While navigating the metaverse, users might also share personal data, whether to purchase goods, connect with other users, or for other purposes.

Past studies have consistently highlighted the limitations of password authentication systems, as there are now many cyber-attacks and strategies for cracking them. To increase the security of users navigating the metaverse, therefore, password-based authentication would be far from ideal.

This inspired a team of researchers at VIT-AP University in India to create MetaSecure, a password-less authentication system for the metaverse. This system, introduced in a paper pre-published on arXiv, combines three different authentication techniques, namely device attestation, facial recognition and physical security keys.

A multi-disciplinary team of researchers has developed a way to monitor the progression of movement disorders using motion capture technology and AI.

In two ground-breaking studies, published in Nature Medicine, a cross-disciplinary team of AI and clinical researchers have shown that by combining human movement data gathered from wearable tech with a powerful new medical AI technology they are able to identify clear movement patterns, predict future disease progression and significantly increase the efficiency of clinical trials in two very different rare disorders, Duchenne muscular dystrophy (DMD) and Friedreich’s ataxia (FA).

DMD and FA are rare, degenerative, genetic diseases that affect movement and eventually lead to paralysis. There are currently no cures for either disease, but researchers hope that these results will significantly speed up the search for new treatments.

AI robots, with Elon Musk, Boston Dynamics. To learn AI, visit: https://brilliant.org/digitalengine where you’ll also find loads of fun courses on maths, science and computer science.

Sources:

Continue reading “Why these explosive AI robot birds will be everywhere, with Elon Musk, ChatGPT” »

Ray Kurzweil on the Law of accelerating returns:

Ray Kurzweil: Acceleration of technology is the implication of what I call the law of accelerating returns. The nature of technological progress is exponential. If I count linearly 30 steps: 1, 2, 3, 4, 5… I get to 30. If I count exponentially: 2, 4, 8, 16… 30 steps later I’m at a billion. It makes a dramatic difference.

Continue reading “Transcendent Man: The Law of Accelerating Returns” »

AI is being used to generate everything from images to text to artificial proteins, and now another thing has been added to the list: speech. Last week researchers from Microsoft released a paper on a new AI called VALL-E that can accurately simulate anyone’s voice based on a sample just three seconds long. VALL-E isn’t the first speech simulator to be created, but it’s built in a different way than its predecessors—and could carry a greater risk for potential misuse.

Most existing text-to-speech models use waveforms (graphical representations of sound waves as they move through a medium over time) to create fake voices, tweaking characteristics like tone or pitch to approximate a given voice. VALL-E, though, takes a sample of someone’s voice and breaks it down into components called tokens, then uses those tokens to create new sounds based on the “rules” it already learned about this voice. If a voice is particularly deep, or a speaker pronounces their A’s in a nasal-y way, or they’re more monotone than average, these are all traits the AI would pick up on and be able to replicate.

The model is based on a technology called EnCodec by Meta, which was just released this part October. The tool uses a three-part system to compress audio to 10 times smaller than MP3s with no loss in quality; its creators meant for one of its uses to be improving the quality of voice and music on calls made over low-bandwidth connections.