•
August 3, 2023
•
7 mins
Hi Niuralogists!
We’re back again with the most recent advancements from the thrilling landscape of artificial intelligence! As AI continues to evolve, we’re here to keep you ahead of the curve. Our goal is to summarize the complex world of AI into understandable insights focused on business, legislation, and individuals. This week we plan to go over Google Assistant getting updated with AI, Amazon’s expansion of their AI tool, Intel’s future plans, MIT’s AI editing prevention, and Stability AI’s new flagship model.
Google is planning for an AI makeover of its Google Assistant, with a focus on using generative AI technologies similar to OpenAI’s ChatGPT and Google’s Bard chatbots. This shift will change how Assistant functions for consumers, developers, and Google’s own employees.
In fact, the revamp has already begun with the mobile version being updated. As part of this transformation, Google is also reorganizing its teams that work on Assistant and making a small number of layoffs. This move is similar to what other companies are doing like Amazon which is working on an AI-powered reboot for its digital assistant, Alexa.
Amazon Web Services (AWS) is stepping up its game with the goal of becoming the preferred cloud platform for organizations looking to adopt AI. A key aspect of Amazon’s strategy is Bedrock, an AI platform that serves as a comprehensive toolbox for businesses. It provides a range of services that facilitate the creation, training, and deployment of new machine-learning models. During the AWS Summit in New York, AWS announced a series of new enhancements and updates to Bedrock. One of the key offerings is the addition of new models on the platform.
Bedrock now supports models from Cohere, Anthropic, and Stability AI, broadening the range of AI capabilities available for businesses. Another major feature is the Bedrock agent, which is designed to simplify the process of building services for users. An agent is a generative AI that can help execute multiple tasks on behalf of a developer. With the Bedrock agent, developers can enable generative AI to complete tasks in just a few clicks and help configure foundation models automatically. This makes the process of building AI models more efficient.
Intel CEO Pat Gelsinger has announced the company's ambitious plan to incorporate AI into every product they build. This initiative will kick off with the launch of Meteor Lake, Intel's first consumer chip equipped with a built-in neural processor for machine learning tasks. Gelsinger sees AI being integrated into all Intel products, not just premium offerings. He argues that while AI and cloud are often linked, they don't necessarily have to be.
He suggests that AI enablement will be client-centered and will also be at the edge, as latency, bandwidth, and cost issues make it impractical to always rely on the cloud. Gelsinger believes that AI will revolutionize everything from real-time language translation in video calls to content generation in gaming environments and productivity tools. This move by Intel is seen as a strategic step to carve out its own niche in the AI landscape which is currently dominated by companies like Nvidia.
Researchers at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) have developed a technique called "PhotoGuard" to protect images from unauthorized edits by AI. As generative AI systems become more prevalent and gain the ability to edit images, concerns about unauthorized manipulation or theft of online artwork and images have grown.
PhotoGuard addresses this issue by subtly altering certain pixels in an image in a way that disrupts an AI's ability to understand the image. These alterations, referred to as "perturbations," are invisible to the human eye but can be detected by machines. The technique employs two methods: the "encoder" attack, which targets the AI model's representation of the image, and the more advanced "diffusion" attack, which camouflages an image as a different image in the eyes of the AI. While the technique isn't foolproof, it represents a significant step towards protecting images from unauthorized AI manipulation. The researchers emphasize the need for a collaborative approach involving model developers, social media platforms, and policymakers to defend against unauthorized image manipulation.
Stability AI recently announced the launch of its SDXL 1.0 text-to-image generation models which the company describes as the "world’s best open image generation model”. SDXL 1.0 is the latest in a series of innovative models released by Stability AI and follows the Stable Diffusion and StableLM models. It is also Stability AI’s new flagship model.
The SDXL 1.0 model is designed to generate high-quality images in any art style with a particular focus on photorealism and can accurately depict complex concepts, creating vivid images. The model is capable of producing images with vibrant colors, superior contrast, lighting, and shadows, all in native 1024×1024 resolution. Despite its advanced capabilities, consumers don’t need supercomputers to run it and the model can run effectively on consumer-grade GPUs with 8GB VRAM and readily available cloud instances.
📬 Receive our amazing posts straight to your inbox. Get the latest news, company insights, and Niural updates.
👮♀️How can AI be used in crime prevention and law enforcement?
There are already various ways that AI is benefitting crime prevention and law enforcement. One of the key ways that AI is contributing is through its ability to efficiently process and analyze large amounts of data from different sources. These sources can be anything from surveillance footage and public records to social media posts. AI can identify and comprehend complex patterns in this data in ways that would usually be extremely challenging or even impossible with traditional methods. Additionally, AI aids in strategic resource allocation. By spotting crime trends, AI can guide the optimal distribution of resources such as police patrols or emergency services. Lastly, AI can evaluate the effectiveness of different crime prevention programs. With the number of different campaigns happening globally to target different problems such as human trafficking, drugs, and weapons, it used to be extremely difficult to determine which ones were successful. Now, AI can sift through the data and provide accurate assessments of different strategies, allowing law enforcement to prioritize the most effective ones. As AI grows and becomes more efficient, the potential to reshape crime prevention and law enforcement is becoming more apparent.
🧠Can AI be programmed to understand human emotions?
The feat of teaching AI to understand human emotions has seen considerable progress over the past few years. Several companies such as the company Affectiva, use multimodal emotion AI to assess various aspects of human emotion including facial expressions, speech, and body language, reaching success rates up to 90%. The creation of Artificial General Intelligence (AGI) represents a leap in the pursuit of AI emulating human-like qualities such as emotional and social intelligence. Despite AI technology being able to mimic and determine human emotions, it doesn’t genuinely possess any of its own emotions. Experts say that we are still many years away from an AI that can truly replicate and understand all the nuances of human emotion.
Will AI contribute to the discovery of new scientific principles or theories?
The impact of AI on science is expected to be transformative, paving the way for new discoveries and theories. Experts such as Professor Andrew Biggs envisions a future where AI is equipped with advanced techniques like reinforcement learning and Bayesian optimization in order to determine what data is necessary to collect which could lead to unprecedented discoveries. This shift would see AI systems move from being just tools to actually being active contributors in research. Some areas that AI can be applied to are quantum technologies, batteries, solid-state lighting, nanoelectronics, nanomechanics, and the optimization of data-rich manufacturing operations. However, the type of machine learning that would produce the best results in aiding scientific discovery is very complex and requires computing hardware that is much more efficient than what exists today. Thankfully, different companies are beginning to invest in these technologies such as Sony and DeepMind. Drug discovery and medical research will also benefit from new AI-driven techniques. Professor Biggs predicts new drugs that are able to directly reach the diseased tissues with fewer side effects. He also predicts that with sufficient investment, rapid progress can be made in 5 years leading to levels of research similar to those conducted by second-year graduate students. Additionally, new machine learning techniques will be able to conduct complex research that would not be possible with humans alone.
AI Tools:
.png)