AI superalignment

Welcome to Daily Zaps — Here’s what we got for you today: 🙌

AI superalignment

OpenAI has started a project called "superalignment." The goal is to figure out how to control super smart AI systems in the future. They want to make sure these AI systems understand and follow human goals. It's a tough problem because these future AI systems are so advanced that we don't really know how they'll work yet.

To start tackling this, OpenAI did an interesting experiment. They used a less advanced AI model (GPT-2) to guide a much more advanced one (GPT-4). The experiment involved giving GPT-4 different tasks, like solving chess puzzles and understanding language, to see if it could do better than the instructions it got from GPT-2. Surprisingly, GPT-4 did really well, even better than GPT-2. It was especially good at language tasks.

This shows that the smarter AI can figure things out on its own, even if the guidance it gets isn't perfect. This result is important for the future. It suggests that super smart AI might be able to understand complex things and make safe decisions, even better than humans can. It's a step towards making sure that when we have these super smart AI systems, they will be safe and helpful.

GPT 4.5 leak speculations

OpenAI CEO Sam Altman has dismissed rumors of an imminent GPT-4.5 release, following speculation sparked by a supposedly leaked screenshot that circulated on social media and Reddit. Screenshot below, claiming to describe the capabilities of "GPT 4.5," was later deemed a hoax, with skepticism arising from inconsistencies in the details it presented.

From Our Partners

Turn your website and product PDFs into customer support chatbots

Here's why website visitors love chatbots:

Quick Problem Resolution: 70% of customers report that chatbots on websites lead to faster and more efficient problem-solving.

24/7 Service Satisfaction: 85% of consumers are more satisfied with businesses that offer 24/7 chatbot support, appreciating the constant availability for assistance.

Enhanced Engagement: 65% of customers feel more engaged with websites offering personalized chatbot interactions.

Build a custom chatbot in 3 minutes →

Is wearable AI the next big thing (again)?

The emerging battlefield in artificial intelligence (AI) technology for the next year is centered around smart glasses and wearable devices, with several major players in the tech industry racing to integrate AI into these products:

Meta Platforms: Ray-Ban Smart Glasses - Enhanced with multimodal AI voice assistant for visual descriptions, clothing suggestions, and text translation.

Google: Gemini AI - Embedded in Pixel phones, with future plans for smart glasses software, focusing on object and action identification.

Microsoft: HoloLens AR Headset - Integrating AI software for interactive environmental understanding and voice assistance.

OpenAI: GPT-4 with Vision - Proposed for integration with Snap’s Spectacles, offering advanced object recognition capabilities.

Apple: Vision Pro Headset - Expected to feature advanced computer vision, with ongoing development in multimodal AI for image and video recognition.

Amazon: Unnamed AI Device (in planning) - A new multimodal AI device being proposed by the Alexa team, separate from Echo voice assistant devices.

AI tax filing assistant

H&R Block has launched a new conversational AI chatbot, AI Tax Assist, integrated into its paid DIY tax software. This chatbot answers questions about tax rules, exemptions, and other related issues, guiding users through their tax filing process. It's particularly useful for those with complicated tax situations, like gig workers, providing specific advice such as deductible business expenses.

The service, developed using models from OpenAI and Microsoft. H&R Block ensures the accuracy of AI Tax Assist through internal content and expert supervision, offering human expert services for double-checking and addressing concerns over AI-provided answers.