• Google, like Microsoft’s Copilot, will take responsibility for copyright infringement – they’re introducing Training Data Indemnity and Generated Output Indemnity to protect their customers, but only those who follow responsible AI practices. - Google
• Dropbox’s AI playbook – remote work helped them focus on leveraging AI for product development, Dash is their universal search that can comb through Microsoft 365 and Google Workspace, they’re also working on an AI video editing tool called Studio. - Fast Company
• Either it’s just a fun experience, or it’s a worrying sign of the loneliness epidemic – on average, the 4 million monthly active users of Character.AI spend 2hrs/day talking to their chatbots. - Quartz
• LLMs can explain LLMs – attempting to automate the process of alignment research, OpenAI used GPT-4 to generate explanations for the behavior of individual neurons in the GPT-2 model (307,200 neurons to be exact). - OpenAI
• The first “highly integrated bionic hand that can be used independently and reliably in daily life” is a triumph of “sophisticated implanted electrodes, new microsurgical techniques and machine learning” – AI plays the key role of translating signals into hand movements. - Bionics Institute
• New skill unlocked, spotting supernovas – the BTSbot was fed 1.4 million images from 16,000 astronomical sources, which enabled it to identify a new supernova candidate and automatically request a spectrum reading, saving lots of time for researchers. - Gizmodo
• “Love-GPT” is a tool for scammers on dating apps and it’s been active for over a decade, but has now grown in power after LLM integration – it’s active on all leading dating apps, and it’s so convincing that it’s very easy to fall prey to it. - Avast
• Great generative AI FAQ from professor Ethan Mollick, including questions like “can you detect AI writing” (nope), “what’s the best way to get good at using AI”, as well as copyright issue explanations and a science-backed perspective on the future of AI. - One Useful Thing
