According to initial reports on social media, the recently released Claude 3 Opus is capable of generating coherent code for entire applications. Developer Murat Ayfer shows an impressive demo: He lets Claude 3 generate a multi-user drawing application in the browser with a single prompt on the first try. Ayfer used the following prompt: "Make a multiplayer drawing app where the strokes appear on everyone else's screens in realtime. let user pick a name and color. save users to db on login." If you want to test the app, you can find the code here.
ChatGPT can now read messages out loud. Simply touch and hold a message in the iOS or Android app and select the "Read Aloud" option. The web version has its own read-aloud button below the message. The synthetic voice in the demo doesn't sound quite as convincing as ElvenLabs and the like - probably for cost reasons. We had hoped that OpenAI would respond with GPT-4.5 or GPT-5 to Claude 3 just announced by Anthropic, which is supposed to surpass GPT-4. But reading messages out loud is fine, too.
Meta's AI image generator Imagine creates ahistorical images similar to Google's Gemini, according to a test by Axios. Based on the Emu image synthesis model, Imagine generates images for Instagram and Facebook direct messages. The AI-generated images show diversity in gender and skin color where that diversity did not exist historically. Examples include people of color as founding fathers or the Pope. The system generates these images with a generic prompt, such as "founding fathers," without explicitly requesting the subject. Some critics say this distorts history or is anti-white. The service is currently only available in the United States.
Researchers have developed a generative AI worm called Morris II that can steal data and send spam. The Morris II worm uses two methods: a text-based, self-replicating prompt and a self-replicating prompt embedded in an image file. In the first method, the worm uses Retrieval Augmented Generation (RAG) to "poison" an email program's database, allowing it to steal data and infect new hosts when responding to emails. In the second method, a malicious request embedded in an image triggers the email assistant to forward the message, spreading spam, offensive material, or propaganda. The researchers believe that generative AI worms could appear in the wild in the next two to three years.
