This project extracts text from images of product labels and uses GPT-3 to predict whether the substance is toxic to dogs. There are a number of potentially toxic substances to dogs, including food, rodenticides, plants, drugs (prescription and other), and mycotoxins (e.g., moldy grains). Since ingredients are important in determining the toxicity of a substance, the user can upload a photograph of the ingredient label or manually enter the ingredients. This can also be combined with external applications, e.g. for plant and mushroom identification. Disclaimer: this is a weekend hackathon project. If you think your pet may have been exposed to poison, please contact a professional and not GPT-3.
Common Vulnerabilities and Exposures (CVEs) are a standardized way of cataloging vulnerabilities in software. When a security vulnerability is discovered, it can be assigned a CVE identifier to ensure that information about the vulnerability can be shared and referenced across different platforms and organizations. CVEs are currently maintained as freeform text, but will soon be maintained in JSON format. A Software Bill of Materials (SBOM) is a document that provides a comprehensive inventory of software components and dependencies. It is a structured list that includes information about the software components, libraries, frameworks, and other third-party software used in the development or deployment of a software product. SBOMs can be automatically generated from files that define build information for software, e.g. POM files for Maven, requirements.txt for Python, etc. It would be extremely useful to be able to automatically scan SBOMs to see whether the associated software packages are vulnerable to the CVEs. Google Vertex's Palm turned out to be capable of automatically finding a list of affected versions based on the CVE descriptions. For example, there was a major vulnerability in Apache Log4J in late 2021. Let's take a look at the version information in the CVE (https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2021-44832): "Apache Log4j2 versions 2.0-beta7 through 2.17.0 (excluding security fix releases 2.3.2 and 2.12.4) are vulnerable to a remote code execution (RCE) attack...." That makes sense to a human, but it's not conducive to automatic processing. Google Vertex was able to generate a specific list of Log4J versions that met the criteria in the description. It also shows potential for automatically extracting additional information from text descriptions in CVEs. With this information, we can automatically read an SBOM and determine whether the software is vulnerable to a particular CVE.
Transformers are context-learning neural networks. They learn by tracking relationships in sequential data, for example, words in sentences. You might know them as some of the best AI models in the market - OpenAI’s GPT-3 and Cohere In our 3-day Hackathon, organized along with the Swiss AI Association, you’ll use context-learning neural networks to build solutions with people from all around the globe and have an opportunity to pitch your AI Prototypes worldwide. Jump in to learn more about harnessing the power of Transformers!
A new text-to-image model called Stable Diffusion is set to change the game once again ⚡️. Stable Diffusion is a latent diffusion model that is capable of generating detailed images from text descriptions. It can also be used for tasks such as inpainting, outpainting, and image-to-image translations. Join us for a epic stable diffusion makers event 11-13 November 👊
🚀💻 Be the first to build an AI App on Google's models! Hackathon on July 7-10. 🔬🌐 Try new Vertex AI features from Google Cloud Platform. 🤝🌍 Learn from AI leaders and connect with like-minded people. 🛠️📱 Build apps with the world's best AI tools! 💡🌍 Solve real-world problems with Generative AI models