Author HubMatthias Bastian

Online journalist Matthias is the co-founder and publisher of THE DECODER. He believes that artificial intelligence will fundamentally change the relationship between humans and computers.

AI research

Oct 12, 2024Oct 12, 2024

Apple AI researchers question OpenAI's claims about o1's reasoning capabilities

AI research

Oct 6, 2024Oct 6, 2024

Study reveals major reasoning flaws in smaller AI language models

AI and society

Jan 6, 2024Jan 6, 2024

OpenAI claims New York Times' prompting strategy violates its terms of service

AI in practice

Dec 18, 2024Dec 18, 2024

Short

Google has started rolling out an experimental version of its Gemini 2.0 AI model to users of Gemini Advanced. The new model, called Gemini-Exp-1206, can handle more sophisticated tasks, including complex coding challenges, mathematical problem-solving, and advanced reasoning, according to Google. The company stresses that this experimental model is still in early preview testing and may not perform consistently. The model can't access real-time information and doesn't support all of Gemini's standard features. Access to Gemini-Exp-1206 is currently limited to Gemini Advanced subscribers using either desktop computers or mobile Web browsers. The new model may be a preview of an upcoming larger Gemini 2.0 Pro model, as Google has already officially released the smaller Gemini 2.0 Flash version.

GeminiApp

AI in practice

Dec 18, 2024Dec 18, 2024

OpenAI's move away from nonprofit control could cost billions of dollars

AI in practice

Dec 17, 2024Dec 17, 2024

OpenAI's upgraded o1 model brings function calling, image analysis, and more to the API

AI research

Dec 17, 2024Dec 17, 2024

Google DeepMind launches new AI fact-checking benchmark with Gemini in the lead

AI in practice

Dec 17, 2024Dec 17, 2024

YouTube's new controls let creators decide if AI can train on their videos

AI in practice

Dec 16, 2024Dec 16, 2024

Short

OpenAI announced today that ChatGPT's search features are now free for all registered users worldwide, dropping the previous paid-only restriction. The company says it has spent recent months improving search speed and reliability, particularly for mobile users. The update brings several new features, including built-in maps and voice search. Users can also now set ChatGPT as their default browser search engine. To showcase these improvements, OpenAI released several demos highlighting how ChatGPT can help with everyday tasks like finding local events, picking restaurants, and planning vacations. The system pulls current information from across the web and presents results with a mix of text, images, videos, and interactive maps.

OpenAI

AI in practice

Dec 16, 2024Dec 16, 2024

Google's Veo 2 outperforms OpenAI's Sora Turbo in head-to-head AI video generation tests

AI in practice

Dec 15, 2024Dec 15, 2024

Elon Musk's Grok AI on X is now faster and cites legacy media as sources

AI in practice

Dec 15, 2024Dec 15, 2024

Short

Meta announced Video Seal, a new neural watermarking system that helps identify AI-generated videos even after they've been edited. The watermark remains invisible to viewers but can be detected later to verify a video's origin. The company made the entire system available as open-source software under an MIT license, including the Video Seal Watermarking model, research paper, and both training and inference code. Meta also released an interactive demo that lets users test the technology. Along with Video Seal, Meta introduced two additional tools: Omni Seal Bench, which provides comparative rankings for neural watermarking across different types of media, and Meta Watermark Anything, also released under an MIT license.