AI Updates on 2025-05-31

Google reports massive demand for Veo 3 video generation model with millions of videos generated in recent days, now available on mobile and in more countries including the UK @demishassabis
Google brings Veo 3 to mobile through the Gemini App on Android and iOS for Pro and Ultra members across 71 countries @GoogleAI
TechCrunch reports Google quietly released an app allowing users to download and run AI models locally @TechCrunch

Aravind Srinivas notes AI tools are starting to reduce the number of junior professionals needed in finance, venture capital, investment banking and consulting @AravSrinivas
ChatGPT reaches 1 billion searches per day in just 2 years compared to Google's 11 years to reach similar scale, demonstrating unprecedented technological acceleration @deedydas
Perplexity is being repositioned as a cognitive operating system rather than just a Google competitor, functioning as a Swiss Army knife for thought with retrieval, execution, and synthesis capabilities @soleio
Cursor's AI coding capabilities are creating addictive dopamine rush experiences similar to video games, with users reporting unprecedented coding flow and joy @joulee

Stanford NLP Group warns about AI-generated research papers being submitted to conferences, calling it a terrible evaluation method that burdens the already broken peer review system @stanfordnlp
Dario Hassabis notes the challenge of discussing AI's potential significant impacts without media framing it as product hype @aidan_mclau
Simon Willison introduces the concept of hype coding where developers lose sight of current capabilities by focusing too much on future AI promises, leading to decreased critical thinking @simonw
NAACP calls for halting operations at xAI's data center in Memphis, citing environmental concerns about the dirty data center @TechCrunch

o3 model successfully analyzed 15MB of raw genome data in 4 minutes to provide Polygenic Risk Score assessment for disease risk prediction, though not at clinical diagnostic grade @deedydas
Ethan Mollick tests AI models' ability to create SVG riddles, finding they typically produce either too obvious or too obscure puzzles, with o3 performing best at solving them @emollick
OpenAI's Operator agent successfully found and played a multiplayer tic-tac-toe game online but initially lost, demonstrating both capabilities and limitations of general-purpose AI agents @emollick
Linear introduces AI agents that can be deployed through their mobile app, allowing users to put agents to work while on the go @karrisaarinen
Deedy demonstrates a coding model that generates working code in two seconds through voice commands, calling it the fastest coding model in the world @deedydas

MIT scientists propose that astrocytes, previously considered support cells, might be key to brain's massive memory capacity, potentially revolutionizing understanding of neural memory storage @MIT
Multiple AI research teams successfully submitted AI-generated papers to conferences with some getting accepted, including teams from Sakana, AutoScience, and Intology @stanfordnlp
Jeff Clune proposes a paradigm shift from traditional engineering solutions to engineering evolution, where optimal AI solutions emerge from evolutionary processes rather than human design @jeffclune
Anthropic introduces an interesting tools variant with pre-baked function parameters like str_replace_based_edit_tool that users still need to implement and execute themselves @simonw