Fresh from a USD$124M series A raise from notable investors (including Jaan Tallinn, technology investor and co-founder of Skype and ex Googler Eric Schmidt), US startup Anthropic seeks to build more reliable, general AI systems. Headed by ex VP Research @ OpenAI, the Anthropic team are focussing on the reliability of large-scale AI models and the transparency and interpretability of the same. https://www.anthropic.com/#researchImage credit: Gordon Johnson @Pixabay
Determining the most impactful research can be challenging and useful to such a broad audience. Computer scientists at MIT built a machine learning system - Delphi - Dynamic Early Warning by Learning system, trained on 1.6m papers published in 42 Biotechnology journals between 1982 - 2019. Not without critics - how does a system not perpetuate existing academic bias? - could this help identify foundational and impactful areas of research to collaborate and accelerate?https://www.chemistryworld.com/news/artificial-intelligence-system-can-predict-the-impact-of-research/4013750.articleImage credit - intographics @Pixabay
Google (finally) released a few feature updates including the ability to quickly delete the last 15 mins of search history, a new photos folder with password protection on Android devices and reminders on location tracking in Google Maps. An improved password manager (belatedly) integrated into Chrome can be used across devices, and alert users when a password they use is found in a data breach. https://www.protocol.com/google-io-keynote-highlightsImage by Gerd Altmann from Pixabay
To help patients search for information about skin, hair and nail conditions, Google's AI-based dermatology assist tool leverages volumes of images uploaded by patients. Google is one of the tech behemoths that is in a prime (somewhat unbalanced) position to develop fascinating AI tools as it sits at the edge of billions of records and images globally with all its products.Unveiled at the recent Google IO developer conference, it isn't meant to replace medical advice or specialist treatments - but be more of an education and awareness tool.https://blog.google/technology/health/ai-dermatology-preview-io-2021Image credit: Medicinet
Microsoft recently released to the open source community an automation tool for security testing AI systems. Counterfit helps organisations undertake AI security risk assessments to ensure that the algorithms used in their businesses are robust, reliable, and trustworthy.While ML is increasingly used in defence, healthcare, finance and other applications, how do we ensure that cyber security professionals can assess the security state of deployments and tools?Here's a starter...https://github.com/Azure/counterfit/Image by chenspec from Pixabay
