Welcome back!

OpenAI has entered the search engine market with its latest venture, SearchGPT. This new tool provides direct answers and citations, positioning itself as a competitor in the AI search engine space. Let’s unpack…

Today’s Summary:

• 🔍 OpenAI launches SearchGPT

• 🎥 Stable Video 4D multi-view video

• 🥈 DeepMind AI wins Math Olympiad

• 🌲 AI predicts wildfire spread

• 🔍 Bing’s new generative search

• ⚡ Google speeds up Gemini chatbot

• 🎶 Udio 1.5 music model released

• 2 new tools

TOP STORY

OpenAI launches SearchGPT search engine

The Summary: OpenAI has unveiled SearchGPT, an AI-powered search engine prototype that provides direct answers to queries using real-time web information. The tool combines OpenAI models with web search to offer concise responses with source citations. Currently available to a limited group of users and publishers via waitlist.

Key details:

• Provides in-line attributions and links to original content sources

• Publishers can manage appearances and opt out of AI training while remaining in search results

• No current monetization strategy

• Includes visual responses with images and video

• Competes directly with Perplexity, an AI "answer engine" startup

Why it matters: This move positions OpenAI as a competitor in the AI search engine landscape. The focus on publisher partnerships and clear attribution addresses controversies faced recently by Perplexity, potentially setting new standards for AI search practices.

Join the waitlist here

VIDEO

Stable Video 4D multi-view video generator

The Summary: Stability AI introduced Stable Video 4D, an AI model that transforms a single video into multiple novel-view videos from different angles. The model generates 5 frames across 8 views in about 40 seconds, allowing users to specify camera angles. This advancement enables more efficient creation of dynamic content for applications like game development and virtual reality.

Key details:

• Generates 8 novel-view videos from a single input video

• Combines strengths of video and multi-view generation models

• Outperforms baselines on metrics like FVD for video consistency

• Full 4D optimization takes 15-20 minutes per object

Why it matters: This model represents a new advancement in AI-powered content creation, making it faster and easier to generate dynamic assets from 2D video. As the technology progresses, it will accelerate development of more immersive virtual worlds and experiences.

Technical report - Project page

GOOGLE

DeepMind AI models earn Math Olympiad silver medal

The Summary: Google DeepMind AI models, AlphaProof and AlphaGeometry 2, solved four out of six problems from the 2024 International Mathematical Olympiad (IMO). The systems achieved a score equivalent to a silver medal, demonstrating advanced reasoning capabilities in algebra, number theory, and geometry. This is a new AI breakthrough in handling complex mathematical challenges.

Key details:

• AI solved IMO problems in minutes to 3 days, compared to 9 hours for human contestants

• AlphaProof uses reinforcement learning and formal math language

• AlphaGeometry 2 solved 83% of historical IMO geometry problems from the past 25 years

• AI scored 28 out of 42 points, just one point short of the gold medal threshold

Why it matters: This advancement could lead to powerful AI tools that assist mathematicians in exploring hypotheses and solving long-standing problems. It also opens up new possibilities for human-AI collaboration in mathematical research and education.

QUICK NEWS

Quick news

• Bing introduces generative search

• Google makes its Gemini chatbot faster

• AI models collapse when trained on recursively generated data

• USC scientists use AI to predict a wildfire’s next move

• Udio releases version 1.5 of its music generation model

TOOLS

🥇 New tools

• Gendo - AI for architecture and design

• Gobi - Wellness guided by science