New

Last Week In Ai #254: Deepmind's Ai Cracks Geometry Problems, Meta Commits To Open-source Ai, China And Us Firms Hold Secret Ai Safety Talks, First Ai Chatbot For Presidential Candidate, And More!

Top News

Google DeepMind’s new AI system can solve complex geometry problems

Google's DeepMind has developed a new AI system, AlphaGeometry, that can solve complex geometry problems. The system combines a language model, which is adept at recognizing patterns and predicting subsequent steps, with a symbolic engine, a more traditional rule-based algorithm. This combination allows for both creative thinking and logical reasoning, mirroring the human approach to solving geometry problems. Importantly, AlphaGeometry was trained purely on synthetically generated data, requiring no human data labeling. Researchers tested AlphaGeometry on 30 problems from the International Mathematical Olympiad, successfully solving 25 within the time limit, a significant improvement over the previous state-of-the-art system developed by Wen-Tsün Wu in 1978, which completed only 10.

Meta is developing open source AGI, says Zuckerberg

author

https%3A%2F%2Fsubstack-post-media.s3.ama

Mark Zuckerberg on Instagram: “Some updates on our AI efforts. Our long term vision is to build general intelligence, open source it responsibly, and make it widely available so everyone can benefit. We’re bringing our two major AI research efforts (FAIR and GenAI) closer together to support this. We’re currently training our next-gen model Llama 3, and we’re building massive compute infrastructure to support our future roadmap, including 350k H100s by the end of this year -- and overall almost 600k H100s equivalents of compute if you include other GPUs. Also really excited about our progress building new AI-centric computing devices like Ray Ban Meta smart glasses. Lots more to come soon.”

January 20, 2024

In an announcement via Instagram Reel, Meta CEO Mark Zuckerberg revealed the company’s dedication to developing open-source artificial general intelligence (AGI), in stark difference from other AI organizations who typically keep AI models proprietary. In the near term, this reaffirms that the next foundation model from Meta, Llama 3, will be open source. Zuckerberg also announced that the two AI teams in Meta, FAIR and GenAI, will be brought together under this strategy, and that the company is almost doubling its state-of-the-art compute cluster to support the training of Llama 3. Meta’s commitment to open source is particularly noteworthy given the backdrop of recent debates over AGI at forums like Davos and amidst Meta's own Chief Scientist Yann LeCun’s skepticism about AGI’s imminent fruition.

US companies and Chinese experts engaged in secret diplomacy on AI safety

In an unprecedented move toward collaboration amid technological rivalry, US AI companies OpenAI, Anthropic, and Cohere, held backchannel discussions with Chinese state-affiliated experts in Geneva, aiming to address mutual concerns over AI safety and the management of its societal impacts. The talks, overseen by the Shaikh Group and had knowledge of key governmental agencies, including the White House, mark a rare moment of Sino-US cooperation. They concentrated on establishing a foundation for safe AI development and the potential hazards associated with deploying AI technologies globally. Despite tension over technological leadership and the US restricting AI-related chip exports, the participants recognized AI safety as crucial, with discussions influencing both the UN Security Council and the UK’s AI summit. This collaboration, though excluding major companies like ByteDance, Tencent, Baidu, and Google DeepMind, may pave the way for setting global standards to ensure AI developments benefit humanity, echoing calls from international AI scientists for stringent regulations and safety measures.

Silicon Valley insiders are trying to unseat Biden with help from AI

A super PAC backed by Silicon Valley insiders has launched an artificial intelligence (AI) bot version of presidential candidate Dean Phillips to spread his ideas, marking one of the first known uses of AI in a political campaign. The bot, named Dean.Bot, was initially powered by the large language model behind ChatGPT, but after OpenAI, the company behind ChatGPT, raised concerns about its use in political campaigns, the PAC switched to other open-source models. The PAC, called We Deserve Better, was formed by Silicon Valley entrepreneurs Matt Krisiloff and Jed Somers in response to President Biden's declining poll numbers. Despite the innovative use of AI, experts warn of the potential risks to elections, including the potential for AI to be used to mislead voters.

Other News

Tools

Amazon brings its AI-powered image generator to Fire TV - Amazon introduces AI-powered image generator on Fire TV, allowing users to create and customize images using voice commands through Alexa.

Microsoft makes its AI-powered reading tutor free - Microsoft's AI-powered Reading Coach, previously available only in Teams for Education, is now free for anyone with a Microsoft account, offering personalized reading practice and integrating with learning management systems.

Microsoft Copilot is now using the previously-paywalled GPT-4 Turbo, saving you $20 a month - Microsoft's Copilot now incorporates the previously-paywalled GPT-4 Turbo, offering users free access to enhanced features and technology, potentially impacting the market dominance of ChatGPT.

Adobe’s new AI-powered Premiere Pro features eradicate boring audio editing tasks - Adobe introduces new AI-powered audio editing features to Premiere Pro, including interactive fade handles, automatic audio category tagging, and Enhanced Speech to clean up dialogue, aiming to streamline the editing process and give editors more time for other tasks.

Samsung’s latest Galaxy phones offer live translation over phone calls, texts - Samsung's latest Galaxy phones introduce a new Live Translation feature powered by AI, allowing users to make or receive calls in different languages and receive live translations both audibly and on the screen.

The Rabbit R1 will offer up-to-date answers powered by Perplexity's AI - The Rabbit R1 AI device, powered by Perplexity, will offer up-to-date search results and a free one-year subscription to Perplexity Pro for the first 100,000 buyers.

Business

Figure announces commercial agreement with BMW Manufacturing to bring general purpose robots into automotive production - Figure and BMW Manufacturing have signed a commercial agreement to deploy general purpose robots in automotive manufacturing environments, aiming to increase productivity, reduce costs, and create a safer and more consistent environment.

FDA Clearance Granted for First AI-Powered Medical Device to Detect All Three Common Skin Cancers - AI-powered medical device for skin cancer detection receives FDA clearance, enabling primary care physicians to provide quantitative, point-of-care testing for all types of skin cancer, potentially accelerating patient access to necessary care.

Samsung's Galaxy AI Set to Transform Foldables and Tablets, Says Company Executive - Samsung's Galaxy AI, powered by generative AI, is set to revolutionize communication, productivity, and content creation on the Galaxy S24 series and will expand to other form factors like tablets and foldables.

Meet the woman who transformed Sam Altman into the avatar of AI - Anna Makanju transformed Sam Altman into the AI industry's ambassador, orchestrating his global diplomatic mission and positioning OpenAI as a trusted partner for policymakers.

Nvidia and AMD shares hit record highs on AI chip surge - Nvidia and AMD stocks reach record highs due to investor demand for AI chip companies, driven by the surge in sales of graphics processors for artificial intelligence.

OpenAI announces team to build ‘crowdsourced’ governance ideas into its models - OpenAI forms a new team to collect and incorporate public input on AI models' behaviors, aiming to align with human values and address regulatory concerns.

The Burro Grande finds the agtech robotics firm going big - Agtech robotics firm Burro, previously known as Augean, has seen significant growth in its agtech offering, with 300 robotics systems operating in fields and nurseries, and has raised $24 million in Series B funding to scale, expand its product and engineering teams, and launch a new autonomous farm vehicle called Burro Grande.

AI Startup Sakana Raises $30 Million to Build Smaller AI Models - AI startup Sakana raises $30 million to develop smaller AI models inspired by the collaborative behavior of animals like fish and bees, in contrast to the trend of building larger AI systems.

AI fever takes over Davos pushing crypto aside as the new cool kid on the block - AI has become the new focus at the World Economic Forum in Davos, with major companies promoting their AI products and services, signaling a shift in interest and investments.

AI models that don’t violate copyright are getting a new certification label - AI models are receiving a new certification label to show they don't violate copyright, as groups offer programs to ensure companies have permission to use copyrighted training data.

Musk Demands Bigger Stake in Tesla as Price for A.I. Work - Elon Musk demands a larger stake in Tesla, worth over $80 billion, in exchange for continuing to develop artificial intelligence products, including robots and self-driving technology.

OpenAI CEO Sam Altman is still chasing billions to build AI chips - OpenAI CEO Sam Altman is seeking billions to build a global network of factories for AI chip fabrication, as the demand for high-powered chips to run complex AI systems intensifies.

Research

AlphaFold found thousands of possible psychedelics. Will its predictions help drug discovery? - AlphaFold's predictions of potential new psychedelic molecules could revolutionize drug discovery, as researchers are learning to effectively deploy the AI protein-structure tool for identifying candidate drug compounds.

Machine learning reveals sources of heterogeneity among cells in our bodies - Machine learning methodology using artificial neural network structures called Density Physics-informed neural networks is developed to reveal the sources of cellular heterogeneity in our bodies, with potential implications for cancer treatment.

Scientists are using AI to study bee behavior, zebra movement, and insects on treadmills - AI and machine learning are being used in diverse sub-disciplines in biology, from neuroscience to animal behavior, to study how animals move, migrate, sense their environment, and behave.

Vision Mamba: Efficient Visual Representation Learning with Bidirectional State Space Model - Efficient visual representation learning with bidirectional state space model, Vision Mamba (Vim), is proposed as a pure-SSM-based method for vision tasks, achieving superior performance on ImageNet classification and dense prediction tasks while being more efficient in terms of GPU memory and inference time compared to Transformer-based models.

SceneVerse: Scaling 3D Vision-Language Learning for Grounded Scene Understanding - On SceneVerse, a large-scale 3D vision-language dataset, and a pre-training framework called GPS, which significantly improves performance in 3D visual grounding tasks.

AI Can Convincingly Mimic A Person's Handwriting Style, Researchers Say - AI researchers have developed technology that can convincingly mimic a person's handwriting style based on just a few paragraphs of written material, using a transformer model to achieve this feat.

Masked Audio Generation using a Single Non-Autoregressive Transformer - A novel non-autoregressive model for audio generation, MAGNeT, uses a single transformer model and a rescoring method to achieve faster inference time and comparable results to autoregressive models, as well as a hybrid version combining autoregressive and non-autoregressive approaches.

The Big Picture of AI Research - AI research is presented as a collaborative and argumentative conversation, with the Big Picture Workshop at EMNLP 2023 showcasing the importance of exploring broader research narratives and consolidating differing contributions on topics such as in-context learning, attention as explanation, and teaching morality to AI models.

Concerns

Slew of deepfake video adverts of Sunak on Facebook raises alarm over AI risk to election - Deepfake video adverts impersonating Rishi Sunak on Facebook raise alarm about AI's risk to the general election, with concerns about manipulation and the spread of misinformation.

IMF warns AI to hit almost 40% of jobs worldwide and worsen overall inequality - IMF chief Kristalina Georgieva urged policymakers to tackle this “troubling trend” and proactively take steps “to prevent the technology from further stoking social tensions.”

Major security flaw in Apple, AMD, and Qualcomm GPUs puts AI data at risk - Security flaw in Apple, AMD, and Qualcomm GPUs poses a significant risk to AI data due to potential data exposure and susceptibility of GPUs to significant data leakage.

Policy

How OpenAI is approaching 2024 worldwide elections - OpenAI is working to ensure their AI tools are used safely and responsibly in the 2024 worldwide elections, aiming to prevent abuse such as misleading "deepfakes" and scaled influence operations.

Exclusive-China's Military and Government Acquire Nvidia Chips Despite US Ban - Chinese military and government entities have been acquiring Nvidia chips banned by the U.S., highlighting the challenges in completely cutting off China's access to advanced U.S. chips for AI and sophisticated computers.

US companies and Chinese experts engaged in secret diplomacy on AI safety - US companies and Chinese experts are engaging in secret diplomacy on AI safety, with potential implications for global technological development and cooperation.

A Closer Look at India's Strategy for Regulating AI and Deepfakes - India is taking proactive steps to regulate AI and deepfakes, recognizing the urgent need for international action and the potential for widespread misinformation and abuse, particularly in the form of deepfake technology.

Australia may ask tech companies to label content generated by AI platforms such as ChatGPT - Australia may consider requiring tech companies to label or watermark content created by AI, such as ChatGPT, due to concerns about the rapid evolution of "high risk" AI products outpacing legislation, as well as public distrust and ethical considerations.

Dutch to Use Europe's AI Act 'Immediately,' Invest $222 Million in Sector - Dutch government to immediately adhere to Europe's AI Act and invest $222 million in the sector, aiming to foster local investment and protect against risks.

Analysis

Clues Say Generative AI’s Future Will Be Revealed in 2024 - Generative AI has experienced significant growth in recent years, with companies like OpenAI and Google DeepMind pushing the boundaries with models like GPT-4 and Gemini, but as the industry approaches 2024, there are questions about whether GPT-4 represents the peak of current technology and if Meta's ambitious plans for AGI will succeed, making it a pivotal year for the future of AI.

Expert Opinions

To Stop AI Killing Us All, First Regulate Deepfakes, Says Researcher Connor Leahy - AI researcher Connor Leahy warns about the existential risk posed by advanced AI, advocating for the regulation of deepfakes as a crucial first step to address the broader societal impact and potential dangers of AI technology.

Back to Listing

credit: