• Identifying AI-generated content is impossible at the moment – current AI watermarking techniques aren’t reliable. - Wired
• Lots of deepfake shenanigans lately, seems like the world (and big tech companies especially) isn’t ready for deepfakes – fake Mr Beast ad on TikTok, fake audio recording that endangered Slovakian elections, and fake Tom Hanks ad and Robin Williams videos.
• But there might be a positive side to deepfakes, as one study found that people who watch tutorials with their own faces deepfaked as the instructor “immediately get better at a task” – although they don’t provide any numbers so I’m skeptical. - Bath University
• Sounds like a prequel to Terminator – AI algorithm designs unique walking robots from scratch in 26 seconds on a standard laptop, the creator said that for him this is “the birth of a brand-new organism." - Tech Xplore
• AI-enabled tools like Microsoft's Power Platform are turning non-coders into "citizen developers" who build dozens of complex, workplace-ready apps – however, software standardization and maintainability could become more difficult because of it. - Semafor
• Canva’s Magic Studio is a rich package of AI-powered tools for design, video creation (editing your clips and even generating short video), writing, repurposing, and more – probably the most comprehensive AI-powered content generator out there at the moment. - Canva
• The fact that Canva added a video generation feature is a clear sign that genAI video is getting better – here are some impressive examples of short videos generated with the Gen-2 model from Runway. - X (Twitter)
• LATS (Language Agent Tree Search) makes LLM agents smarter at decision-making by letting them learn from real-world feedback – it achieved the highest score to date in the HumanEval programming benchmark, with 94.4% accuracy. - arXiv
• One of Anthropic's longest-term research bets is on Mechanistic Interpretability to decipher how LLMs work from the inside and steer their behavior – their new finding shows that focusing on features, not just neurons, makes the models more understandable. - Anthropic
