Hey {{first_name|TEDchats Fam}}!

Happy Friday! Remember the last time a magic trick completely baffled you, only to leave you even more amazed when you learned the secret behind it? Well, the world of AI has its own bag of tricks, and just like magic, not everything is as it seems.

👀 SNEAK PEEK

🚨FEATURE STORY

Have you ever watched a technology demonstration that left you spellbound, only to discover later that it wasn't quite what it seemed? This is the tale of Google’s Gemini AI demo – a mesmerizing showcase that turned out to be more smoke and mirrors than cutting-edge technology.

Google’s Gemini AI, designed as a multimodal AI model integrating language and visual understanding, recently made waves with its debut. The highlight was a video titled “Hands-on with Gemini: Interacting with multimodal AI” that reached over a million views in less than 24 hours (see video). It showcased the AI using its ‘vision’ to narrate the drawing of a duck, starting from a squiggly line. Gemini impressively identified and described each step, responded to voice queries, and even tracked objects in real-time. For example, the AI noted, “I see you placing a piece of paper on the table”. The video was a spectacle of AI's potential, leading viewers to believe in Gemini's advanced interactive capabilities.

However, the integrity of this technological marvel was soon questioned (see example). Reports emerged revealing that the most impressive parts of the demo were not real-time interactions but a series of pre-captured and carefully edited segments, calling the demo ‘faked’ (read more). While the AI did generate the responses shown in the video, the mode and speed of interaction were significantly embellished. In reality, Gemini doesn't talk back, and its answers aren’t as smooth as they seem in the video. This revelation sparked a debate about the authenticity of AI demonstrations and the ethical implications of misrepresenting technological capabilities to the public.

When Google published the YouTube video, they provided transparency about its nature, stating in the description section: “For the purposes of this demo, latency has been reduced, and Gemini outputs have been shortened for brevity”. Around the same time as the video release, Google also released an in-depth post describing how the video was made (see post). Furthermore, by stating at the start of the video, “This video highlights some of our favourite interactions with Gemini”, overall, it seems that Google may have only wanted to illustrate ‘potential’ user experiences with Gemini.

This incident raises important questions about the future of AI demonstrations and the ethics surrounding them. It spotlights the need for complete transparency in showcasing AI capabilities, to avoid misleading the public. As AI technology continues to evolve, it becomes increasingly crucial for companies to present their innovations in a clear, truthful and responsible manner, ensuring that the line between reality and simulation remains clear.

🍕 QUICK BITES

In a groundbreaking approach, Anthropic's researchers have unveiled an unexpected tactic to combat AI biases – simply asking the AI. By adding a direct request to AI prompts to disregard certain characteristics such as race and gender, the team significantly reduced discrimination in decision-making tasks. The study, while seemingly unorthodox, underscores the responsiveness of AI to straightforward, ethical guidance (read more).

Starting in 2024, McDonald's plans to deploy generative AI in collaboration with Google, targeting improvements in food quality and operational efficiency. This initiative will encompass hardware and software upgrades across thousands of stores worldwide. According to Brian Rice, McDonald’s Executive Vice President, implementing AI solutions will make “restaurants easier to operate”, while also improving “the overall experience for our customers and crew”. This move signals a growing trend of AI integration in the fast food industry (read more).

Meta's latest update brings a novel feature, 'Reimagine', allowing users to alter AI-generated images shared within Messenger and Instagram chats. Users can interact with these images by adding custom text prompts, effectively 'reimagining' the original content. This addition is part of a series of updates to Meta's AI tools, including an online accessible version of their 'imagine' tool and planned invisible watermarking for AI-generated images (read more).

🧰 AI TOOLBOX

🤯 MIND SPARK

A Figure Skating Robot? Even though the figure skating robot isn’t real, it’s fun to watch, and importantly, it was created by AI! (see more)

Fake AI Scientific Research: An article in the leading scientific journal 'Nature' discusses how scientists can't always distinguish between AI-faked work and genuine research (read more).

Want to Solve a Mystery with AI? Play detective and chat with AI characters to solve a mysterious case (read more).

🎨 AI ART