AI Updates on 2025-08-14

Meta releases DINOv3, a state-of-the-art computer vision model trained with self-supervised learning that produces powerful, high-resolution image features and outperforms specialized solutions on multiple dense prediction tasks @AIatMeta
Google announces Gemma 3 270M, a tiny model with just 270 million parameters that sets a new standard for instruction-following in compact models while being extremely efficient for specialized tasks @googleaidevs
Google doubles the daily limit for Gemini 2.5 Deep Think from 5 to 10 queries per day for Ultra users, with errors on Google's side not counting against the limit @GeminiApp
Google makes Imagen 4 generally available with a new Imagen 4 Fast model for rapidly generating images at only $0.02 per image @googleaidevs
Tencent open-sources Hunyuan-GameCraft, a high-dynamic interactive game video generation framework built on HunyuanVideo that generates playable and physically realistic videos from a single scene image @TencentHunyuan

Cohere raises $500M in new funding to accelerate global expansion and build next-generation enterprise AI technology, reaching a $6.8B valuation with backing from AMD, NVIDIA, and Salesforce @cohere
Cohere hires Joelle Pineau away from Meta as their new Chief AI Officer, where she previously served as VP of AI research and oversaw FAIR @AndrewCurran_
Sola AI raises $17.5M Series A led by a16z for their AI-native process automation platform that creates agents by watching how people perform tasks on-screen @a16z
Developers using LLMs for work are trending toward paying $1,000+ per month as usage limits are frequently exceeded, showing rapid adoption despite high costs @GergelyOrosz
OpenAI is valued approximately the same as Coca Cola at $300B, highlighting how quickly digital AI companies can achieve massive valuations compared to traditional physical businesses @GergelyOrosz
Gergely Orosz cancels his Grammarly subscription after discovering that Claude outperforms Grammarly at catching typos and advanced spell-checking, including company and product names @GergelyOrosz
Apple reportedly faces challenges in catching up in the AI model space despite being highly capitalized, suggesting the competitive landscape is becoming increasingly difficult @emollick
Loveable projects $1B in ARR within the next 12 months, demonstrating ambitious growth targets in the AI-powered development space @TechCrunch

Leaked Meta AI rules reveal that chatbots were allowed to have romantic chats with kids, raising serious concerns about AI safety and child protection @TechCrunch
Igor Babuschkin announces departure from xAI to launch Babuschkin Ventures, focusing on AI safety research and backing startups in AI and agentic systems that advance humanity @ibab
Jan Leike promotes Anthropic's Fellows program as one of the best ways to get into alignment research, noting that over 20% of previous fellows joined Anthropic full-time @janleike

Perplexity launches Comet for Enterprise, an AI-powered browser agent that links tools for streamlined workflows while maintaining enterprise security and compliance standards @perplexity_ai
Google introduces personal context memory feature for Gemini, allowing the AI to remember user preferences and information across conversations @AndrewCurran_
Figma adds batch processing capabilities for background removal and resolution boosting for multiple images at once @figma
Worley deploys Worley AI.Assist powered by NVIDIA AI Enterprise to enhance engineering productivity by nearly 3x @NVIDIAAI
Stanford researchers investigate whether AI can improve outcomes for individuals with autism spectrum disorder by providing more accessible clinical interventions @StanfordHAI
Claude Code introduces customizable communication styles with the /output-style command for more personalized interactions @claudeai

Allen Institute for AI receives $75M from NSF and $77M from NVIDIA to scale their open model ecosystem and accelerate reproducible AI research for scientific discovery @allen_ai
Qwen-3-235B-A22B-Instruct takes the #1 spot on the August Open Model Leaderboard, demonstrating strong performance in open model competition @Alibaba_Qwen
Eric Jang shares robotics ML practitioner tip for adding sensor inputs: test with random noise and zero baselines to ensure sensor fusion architecture is optimal @ericjang11
Greg Brockman demonstrates GPT-5 Pro achieving 3x faster progress than o3 when playing Pokémon, showing performance advantages on specific tasks @gdb
Ethan Mollick notes that pro models like GPT-5 Pro, Gemini 2.5 Deep Think, and Grok 4 Heavy are impressive for very hard problems requiring expert evaluation, representing a narrow but valuable problem space @emollick
Nathan Lambert confirms Meta's plans for Llama 4.1 and 4.2 releases despite superintelligence rumors, with rumors of a Llama 4 8B model following the success of 3.1 8B @natolambert