🗞 Ex-OpenAI Researcher Challenges ChatGPT's "Fair Use"

🔎 The Latest on the AI Frontier:

• Ex-OpenAI Researcher Challenges ChatGPT's Fair Use 📊

• OpenAI's Policy Expert Exits, Calls for More Open Dialogue Within Company 🔄

• Apple's iOS 18.2 Beta Debuts with AI-Powered Creative Tools 🎯

• Google Makes SynthID Text Watermarking Tool Public After Successful Tests 🔐

• Perplexity Expands to Desktop with Feature-Rich macOS Application 💻

• Midjourney Introduces Selective Access to New AI Image Editor Suite 🎨

• Other news you might find interesting 👀

🔇 Former OpenAI researcher provides detailed analysis arguing ChatGPT's use of training data may not qualify as "fair use" under copyright law. Link

• After four years at OpenAI working on GPT-4, Balaji left the company and published an essay analyzing how copyrighted information from training data appears in model outputs, arguing they don't meet "fair use" standards.

• Balaji demonstrates that ChatGPT creates market harm by substituting original content.

• His information theory analysis suggests that 73-94% of ChatGPT's outputs correspond to information from its training dataset, challenging claims of transformative use.

👥 OpenAI's senior policy researcher Miles Brundage departs amid organizational shifts, citing desire for greater research independence in nonprofit sector. Link

• Brundage's exit prompts restructuring of OpenAI's economic research division, which will move under new chief economist Ronnie Chatterji, while the AGI readiness team will be distributed across other divisions.

• In his farewell message, Brundage urged OpenAI employees to "speak their minds" about company improvement, emphasizing the need to avoid groupthink amid difficult upcoming decisions.

🍎 Apple releases iOS 18.2 beta with expanded AI features including image generation, ChatGPT integration, and Visual Intelligence camera search. Link

• First developer beta introduces Genmoji for AI-generated emoji, Image Playground for image creation, and enhanced AI writing tools.

• Visual Intelligence feature enables camera-based search on iPhone 16 devices, with expanded regional support for Apple Intelligence.

• Follows iOS 18.1's upcoming release with basic AI features, though screen-context Siri upgrade remains pending.

🔍 Google DeepMind open-sources SynthID, its AI text watermarking tool, following successful testing with millions of Gemini users. Link

• Tool works by altering token probability scores during text generation, making AI content identifiable while maintaining output quality and creativity.

• Extensive testing with 20 million Gemini responses showed users found no quality difference between watermarked and unwatermarked content.

• While resistant to basic editing, the tool has limitations with heavily rewritten text, translations, and factual responses, prompting calls for additional AI safeguards.

🔍 Perplexity launches native macOS app, bringing AI-powered search to desktop with Pro features and voice input. Link

• App offers streamlined access to advanced search capabilities using GPT-4 Omni and Claude 3, with cited sources and threaded conversations.

• Pro subscribers ($20/month or $200/year) get 600 daily Pro Searches and ability to analyze uploaded files with multiple AI models.

• Features include built-in library for search history, custom shortcuts, and voice input for hands-free operation without browser requirement.

🎨 Midjourney tests new AI image editor with advanced transformation features, rolling out selective access to prevent misuse. Link

• New web-based tool offers AI editing of uploaded images and object re-texturing capabilities, integrating with existing features like style references and personalized models.

• Initial access restricted to annual subscribers, recent monthly subscribers, and users with 10,000+ image generations to prevent abuse issues similar to March 2023.

• Company also reveals development of video generation model and "Midjourney Orb" hardware project, balancing creative flexibility with processing constraints.

🖨 More news you might find interesting:

• OpenAI researchers develop new continuous-time consistency model (sCM) that generates AI media 50x faster than traditional diffusion models. Link

• OpenAI scientist Noam Brown reveals insight at TED AI: 20 seconds of strategic AI thinking equals performance boost of 100,000x more training data. Link

• Apple CEO Tim Cook and COO Jeff Williams visit China, pledging increased investment and emphasizing supply chain importance. Link

• Nvidia announces major AI partnerships in India, including deals with Reliance, Tech Mahindra, and major infrastructure providers. Link

• Nvidia supplier SK Hynix reports record Q3 operating profit of $5.08 billion, driven by surging AI chip demand. Link

• Microsoft Photos app gets AI-powered super resolution and OCR features in latest Windows Insider update. Link

• Researchers develop 'HarmonyCloak' tool to protect songs from AI training while preserving human listening experience. Link

• Raspberry Pi launches AI HAT+ with up to 26 TOPS performance, offering two variants priced at $70 and $110. Link

• Google DeepMind updates AI music creation suite with improved MusicFX DJ, Music AI Sandbox, and YouTube Dream Track features. Link

• Canva launches 'Dream Lab' text-to-image generator powered by recently acquired Leonardo AI's Phoenix model. Link

• Humane slashes Ai Pin price to $499 from $699, but removes extra accessories and color options. Link

• Chinese tech giant Oppo intensifies AI focus through weekly collaborations with Google and Microsoft, while expanding global market presence. Link

• European scientists develop AI algorithm to decode pig emotions through vocalizations, aiming to revolutionize farm animal welfare monitoring. Link

• ByteDance faces significant slowdown as TikTok Shop struggles and geopolitical tensions impact growth in 2024. Link

Have any feedback? Send us an email