Last Week in AI #231: AI brain implant restores movement in quadriplegic man, probe into how investment firms use AI, studios replacing movie extras with AI, Meta's AI chatbot, and more!
AI brain implant restores control and sensing in quadriplegic man, investment firms not properly disclosing AI usage, studios and unions disagree on generative AI consent, Meta's AI chatbot, and more!
In a groundbreaking medical advancement, AI-driven brain implants have enabled a quadriplegic man, Keith Thomas, to regain movement and sensation. After a devastating diving accident in 2020 that damaged his C4 and C5 vertebrae, scientists from Northwell Health’s Feinstein Institutes successfully performed a "double neural bypass" on Thomas. Surgeons implanted microchips in his brain that interface with AI algorithms to re-establish the connection between his brain, body, and spinal cord. When Thomas thinks about moving, the chip signals electrodes on his spine and arm muscles, facilitating movement. Additionally, sensors on his fingers send touch data back to his brain. Remarkably, Thomas' arm strength has more than doubled in just four months post-implantation. This innovative procedure builds upon prior single neural bypasses, with the potential to offer increased mobility and independence to many.
Massachusetts securities regulators, led by Secretary of State Bill Galvin, have initiated an investigation into the use of AI by investment firms in their dealings with investors. The probe, which includes firms such as JPMorgan Chase, Morgan Stanley, Tradier Brokerage, US Tiger Securities, E*Trade, Savvy Advisors, and Hearsay Systems, is driven by concerns about the potential misuse of AI technology without proper disclosure and consideration of conflicts. This move follows a proposal by the U.S. Securities and Exchange Commission to mandate broker-dealers to eliminate possible conflicts of interest arising from the use of AI on trading platforms. The investigation will focus on the supervisory procedures firms have regarding AI and the disclosure process for those who have already deployed AI.
Background actors in Hollywood are expressing concerns about being replaced by AI as the industry increasingly uses body scans to create digital replicas. The Screen Actors Guild‐American Federation of Television and Radio Artists (SAG-AFTRA) union is currently in a labor dispute with studios over the issue, with actors worried about the potential misuse of their digital likenesses and the lack of compensation for their use. The practice of body scanning is becoming more common, with actors often feeling they have no choice but to comply due to fear of retaliation. The advent of generative AI, which can create new conversations, images, and videos, has further fueled these concerns. The union and studios disagree over the meaning of consent, with studios proposing to ask for permission once, while the union insists on separate negotiations each time a digital likeness is used.
Meta, the parent company of Facebook and Instagram, is reportedly developing AI-powered chatbots with distinct personas, set to launch as early as next month. These chatbots, designed for human-like interactions, will adopt various personas, such as a surfer advising on travel plans or a chatbot speaking like Abraham Lincoln. The initiative aims to enhance user engagement on Meta's platforms, with the chatbots serving as a search tool and recommendation system. However, the move is likely to raise privacy concerns due to the potential for increased data collection for targeted content and ads. This development follows Snapchat's launch of its AI chatbot, "My AI", earlier this year. Meta's CEO, Mark Zuckerberg, has indicated that AI will play a significant role in the company's future products, improving interactions across mobile apps and the metaverse. The company's AI roadmap will be further detailed at the Connect developer event in September.
AI-Supported Breast Screens May Reduce Radiologist Workload - AI-supported breast cancer screening is safe, accurate, and reduces radiologist workload, according to early results from a large randomized study.
Large Language Models and Nearest Neighbors - The article discusses a recent application of nearest-neighbor methods in the context of large language models (LLMs) and explores the use of a parameter-free classification method with compressors for text classification.
nhancing CLIP with GPT-4: Harnessing Visual Descriptions as Prompts - GPT-4 can be used to generate visually descriptive text, enhancing CLIP's performance on downstream tasks and improving transfer accuracy on specialized fine-grained datasets.
owards Generalist Biomedical AI - Generalist biomedical AI systems that encode and interpret multimodal biomedical data can enable impactful applications, as demonstrated by the development of Med-PaLM M, a large multimodal generative model that outperforms specialist models on diverse tasks and shows potential clinical utility.
Robotic grippers offer unprecedented combo of strength and delicacy - Researchers at North Carolina State University have developed a robotic gripping device that is gentle enough to pick up a drop of water, strong enough to pick up a 6.4 kilogram (14.1 pound) weight, and precise enough to pick up microfilms that are 20 times thinner than a human hair, with potential applications in manufacturing and robotic prosthetics.
Using AI to protect against AI image manipulation - Researchers from MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) have developed a technique called "PhotoGuard" that uses perturbations to disrupt an AI model's ability to manipulate images, protecting them from unauthorized edits.
The History of Open-Source LLMs: Better Base Models (Part Two) - Open-source research on large language models (LLMs) has led to the development of high-performing base models that are pre-trained over massive datasets, narrowing the performance gap between open-source and proprietary models.
LLM-grounded Diffusion: Enhancing Prompt Understanding of Text-to-Image Diffusion Models with Large Language Models - Enhancing prompt understanding of text-to-image diffusion models by incorporating off-the-shelf frozen LLMs in a two-stage generation process.
Unified Model for Image, Video, Audio and Language Tasks - UnIVAL is a ~0.25B parameter unified model that efficiently pretrains on multiple tasks and combines text, images, video, and audio into a single model, achieving competitive performance across modalities and demonstrating the benefits of multimodal model merging.
ool Documentation Enables Zero-Shot Tool-Usage with Large Language Models - Tool documentation is proposed as an alternative to demonstrations for teaching large language models how to use new tools, with empirical findings showing that zero-shot prompts with only tool documentation can achieve performance on par with few-shot prompts.
HierVST: Hierarchical Adaptive Zero-shot Voice Style Transfer - HierVST is a hierarchical adaptive end-to-end zero-shot voice style transfer model that outperforms other VST models in zero-shot scenarios, allowing for the transfer of voice styles of novel speakers without the need for text transcripts.
The patent-pending innovation sees texture and depth and perceives physical attributes of people and environments - Researchers at Purdue University have developed a patent-pending method called HADAR, or heat-assisted detection and ranging, which combines thermal physics, infrared imaging, and machine learning to improve machine perception in low-light conditions and overcome the limitations of traditional active sensors.
Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback - The article discusses the challenges and limitations of reinforcement learning from human feedback in AI.
AI Zoology: Why are there so many animal models? - Researchers have developed various animal-named models, such as Llama, Alpaca, Vicuna, Falcon, and Orca, which are descendants of the original Llama model and have been fine-tuned for different purposes, including instruction-following and conversation data.
IBM and NASA teamed up to build the GPT of Earth sciences - IBM and NASA have collaborated to build an open-source geospatial foundation model that will serve as the basis for a new class of climate and Earth science AIs that can track deforestation, predict crop yields, and track greenhouse gas emissions.
Steg.AI puts deep learning on the job in a clever evolution of watermarking - Steg.AI has developed a deep learning approach to embedding invisible watermarks in images that can withstand transformations and re-encoding, providing a robust method for creators to prove ownership of their digital assets.
Google Assistant reportedly pivoting to generative AI - Google Assistant is reportedly undergoing a generative AI transformation, with the Assistant team exploring the potential of a supercharged Assistant powered by the latest large language model technology.
The future of AI is video, and it’s coming at us fast - AI video editing software is advancing rapidly, with the ability to copy one style of video to another, raising concerns about the potential for fake news and manipulated content in the future.
Tinder tests AI photo selection feature to help users build profiles - Tinder is testing an AI photo selection feature that selects the five best photos from a user's album to build a dating profile, aiming to eliminate the struggle of finding pictures and accurately represent users.
Songwriters Are Using AI to Sell an Artist’s Sound Back to Them - Songwriters are using AI voice synthesis to create "pitch records" that mimic the voice of the artist being pitched, allowing artists to envision themselves on the track and potentially democratizing the songwriting profession.
GitHub Copilot can now tell developers when its suggestions match code in a public repository - GitHub Copilot now has a code referencing feature that allows developers to see and decide what to do with code suggestions that match existing code in public repositories.
Students can use AI on applications, Arizona State law school says - The Sandra Day O'Connor College of Law at Arizona State University is allowing prospective students to use generative artificial intelligence tools to help draft their applications, as long as they certify their use and the information submitted is truthful.
AI investment forecast to approach $200 billion globally by 2025 - Investment in artificial intelligence is expected to reach around $200 billion globally by 2025, with the potential to boost global labor productivity by more than 1 percentage point a year in the decade following widespread usage, according to Goldman Sachs Economics Research.
The state of AI in 2023: Generative AI’s breakout year - The explosive growth of generative AI (gen AI) tools is confirmed by the latest annual McKinsey Global Survey, with one-third of respondents saying their organizations are using gen AI regularly in at least one business function, and 40 percent of respondents stating that their organizations will increase their investment in AI overall because of advances in gen AI.
Palantir stock up 10% as company rides A.I. craze - Shares of Palantir rose 10% as the company continues to benefit from investor optimism about its AI software, with CEO Alex Karp stating that demand for their AI platform is "without precedent."
AI chip firm Tenstorrent raises $100 mln from Hyundai, Samsung - Canadian AI chip firm Tenstorrent has raised $100 million in funding from Hyundai Motor Group and a Samsung investment fund, among others, as it looks to challenge market leader Nvidia in supplying chips for artificial intelligence products.
Nvidia's AI GPUs Are Selling for up to $70,000 in China - Demand for AI-based services and fears of US restrictions on GPU sales are driving up prices in China, with Nvidia's H800 compute GPU reaching as high as $70,000 per unit, making it difficult to obtain.
Google’s AI search is getting more video and better links - Google's AI-powered Search Generative Experience (SGE) is adding images and videos to the summary box at the top of search results, with plans to make the box appear faster and provide more context to the links it includes.
Alibaba rolls out open-sourced AI model to take on Meta's Llama 2 - Alibaba's cloud computing unit has released two open-sourced AI models, Qwen-7B and Qwen-7B-Chat, in an effort to compete with Meta Platform's Llama 2, marking the first time a major Chinese tech company has open-sourced its large language models.
RapidAI Announces $75 Million Growth Investment Led by Vista Credit Partners - RapidAI, a global leader in developing AI and technology workflow solutions for neurovascular, cardiac, and vascular diseases, has announced $75 million in funding to drive platform innovation and accelerate growth in new regions and product verticals.
Inworld, a generative AI platform for creating NPCs, lands fresh investment - Inworld, a startup launched by the team behind API.AI, has raised over $50 million in funding for its generative AI platform that aims to create more lifelike and immersive gaming experiences by dynamically generating dialogue and allowing NPCs to learn, adapt, and autonomously initiate goals and actions.
Apple has been quiet about ChatGPT. Now Tim Cook says its hefty $22.6 billion research spend is down to generative AI. - Apple's CEO Tim Cook reveals that the company's hefty research spending is driven by generative AI, with research and development spending hitting $22.61 billion for the year so far, a figure that's $3.12 billion higher than last year.
Andy Jassy summed up Amazon’s A.I. gameplan: Every single business unit has ‘multiple generative A.I. initiatives going on’ - Amazon CEO Andy Jassy revealed that every single business unit within Amazon has multiple generative AI initiatives underway, with AI set to be at the heart of the company's future plans.
Dell, Nvidia join forces for next-gen generative AI solutions - Dell Technologies has announced a new portfolio of generative AI solutions, including validated designs with Nvidia, professional services, and Precision workstations, to help customers navigate the generative AI landscape.
Generative AI Startups Might Be Profitable in 2 Years - Generative AI startups are attracting investment from venture capitalists, with many exits occurring through acquisition by big tech firms, but investors face challenges in assessing the unique value and potential return on investment of these startups.
Toyota, Pony.ai plan to mass produce robotaxis in China - Toyota and Pony.ai are partnering to mass produce robotaxis in China, investing over 1 billion yuan ($140 million) to build cars with autonomous driving technology and ride-hailing services.
ChatGPT update brings useful feature known from Microsoft's Bing Chat - ChatGPT now offers contextual questions and answers in each chat, along with topic suggestions, similar to Microsoft's Bing chatbot.
Nvidia GPU shortage is ‘top gossip’ of Silicon Valley - The shortage of Nvidia GPUs, particularly the H100 model, is causing speculation and gossip in Silicon Valley as companies vie for access to the high-performance computing chips for large language model training, with estimates suggesting that demand will continue to rise until at least the end of 2024.
Introducing AudioCraft: A Generative AI Tool For Audio and Music - AudioCraft is an AI tool that generates high-quality audio and music from text prompts, and it consists of three models: MusicGen, AudioGen, and EnCodec, which are now open-sourced to advance the field of AI-generated audio and music.
Pinterest touts Amazon partnership progress, AI in Q2 earnings beat - Pinterest provides an update on its Amazon partnership progress and highlights the role of AI in its Q2 earnings, including improved ad relevance and engagement.
Five months ago Crunchbase vowed to try and avoid layoffs. Since then A.I. has changed everything - Crunchbase CEO Jager McConnell has announced layoffs at the company, citing the need to hire more data scientists and engineers to keep up with recent developments in artificial intelligence.
AI.com flips from ChatGPT to Elon Musk’s X.ai - Elon Musk's X.ai has acquired the valuable domain AI.com, which previously redirected to OpenAI's ChatGPT web interface, sparking speculation about the reasons behind the switch.
What Self-Driving Cars Tell Us About AI Risks - The risks of AI in vehicles, particularly self-driving cars, were highlighted in a plea to regulate its use made to the U.S. Senate Committee on Commerce, Science, and Transportation.
A.I. is on a collision course with white-collar, high-paid jobs — and with unknown impact - Artificial intelligence is expected to target white-collar, higher-paid jobs, potentially causing significant worker displacement and inequality, while also creating new challenges and needs for retraining and reskilling.
Why it’s time to clean up AI’s carbon footprint - The rise of AI has led to concerns about its environmental impact, with researchers attempting to quantify the carbon footprint of AI and highlighting the need for greater awareness and accountability.
An Asian MIT student asked AI to turn an image of her into a professional headshot. It made her white, with lighter skin and blue eyes. - An Asian MIT student was surprised when an AI image generator turned her into a white person with lighter skin and blue eyes for a professional headshot, sparking a conversation about AI bias and the need to mitigate it.
Tech experts are starting to doubt that ChatGPT and A.I. ‘hallucinations’ will ever go away: ‘This isn’t fixable’ - Experts are skeptical that AI systems like ChatGPT will ever be able to eliminate the problem of generating false information, which poses challenges for businesses and organizations relying on these systems for tasks with high stakes.
The generative A.I. battle between companies and hackers is starting - The battle between companies and hackers over generative AI is intensifying, with AI tools being used to both fortify cybersecurity and make attacks more sophisticated and widespread.
Workers at risk of AI radically changing their jobs aren't too worried about it - Over 27 million Americans are in work where key activities involved in their job could "be replaced or aided by artificial intelligence" — or what Pew Research Center defines in a new analysis as being most exposed to AI, and while not all workers in higher exposed industries feel AI's risks outweigh its benefits over the long term, workers who are relatively more exposed to AI also seem relatively less concerned about the impact of AI on them personally.
‘It’s destroyed me completely’: Kenyan moderators decry toll of training of AI models - Former content moderators for Open AI's ChatGPT in Nairobi, Kenya, have filed a petition to the Kenyan government, alleging exploitative conditions, psychological trauma, low pay, and abrupt dismissal while reviewing content for artificial intelligence programs, calling for an investigation into the matter.
Experience: scammers used AI to fake my daughter’s kidnap - A mother recounts her terrifying experience of receiving a phone call from someone claiming to have kidnapped her daughter, only to later discover that it was an AI scam.
Why This AI Moment May Be the Real Deal - The new class of AI, characterized by transformers, is showing remarkable abilities in natural language processing, context understanding, and flexible learning, suggesting that this AI moment may be the real deal.
ChatGPT Is the Perfect Creative Machine—Except For One Big Little Thing - ChatGPT lacks the essential human element of "epsilon" or individual creativity, which is crucial for true artistic expression and the preservation of human soul in creative works.
If Play is a Sign of Intelligence, AI is Unintelligent - Art, play, and pretend are signs of sophisticated intelligence demonstrating profound levels of tacit knowledge and theory of mind, which AI currently lacks.
‘So important’: UK minister endorses Google’s training drive in AI arms race - Google's plans to provide basic AI training to workers and bosses in the UK have been praised by the country's science and innovation secretary, who sees it as a vote of confidence in the UK's ambition to become a leader in AI.
The Stanford University ‘boot camp’ teaching Congress about AI - Congressional staffers from both sides of the aisle will attend a three-day boot camp on AI at Stanford University, where they will learn about the impact of AI on issues such as international security, the future of work, bias, privacy, and healthcare, in an effort to equip them with the knowledge needed to think critically about regulating and governing AI.
Some U.S. government agencies are testing out AI to help fulfill public records requests - Some U.S. government agencies are testing out AI tools to help process Freedom of Information Act requests, but transparency advocates warn that safeguards are needed before wider deployment.
Kickstarter requires generative AI projects to disclose additional info - Kickstarter will require projects that use AI tools to generate content to disclose relevant details about their use of AI and the sources of their training data, in an effort to address concerns about fair use and credit for content creators.
AMD considers making a specific A.I. chip for China to comply with export controls - AMD is considering developing an AI chip specifically for the Chinese market to comply with U.S. export controls, following in the footsteps of Nvidia and Intel.
Generative AI services pulled from Apple App Store in China ahead of new regulations - Generative AI apps have been removed from Apple's China App Store ahead of new regulations requiring AI apps to obtain an administrative license, causing concerns about the impact on developers and the market.
Top Hong Kong university drops ban on ChatGPT in coursework by students - The University of Hong Kong has lifted its ban on students using ChatGPT and will provide generative AI tools to students and teaching staff, allowing them to use the AI application every month without the need for a VPN, although students will have to declare their use of AI as they submit coursework.
A jargon-free explanation of how AI large language models work - Large language models (LLMs) like ChatGPT are powerful but not well understood, as they are built on neural networks trained using billions of words of ordinary language, making their inner workings a deep mystery that researchers are still trying to unravel.
Copyright © 2023 Skynet Today, All rights reserved.